CN115203381A

CN115203381A - Marketing call recommendation system, method, device and equipment

Info

Publication number: CN115203381A
Application number: CN202110343858.8A
Authority: CN
Inventors: 王兴军
Original assignee: Alibaba Singapore Holdings Pte Ltd
Current assignee: Alibaba Innovation Co
Priority date: 2021-03-26
Filing date: 2021-03-26
Publication date: 2022-10-18

Abstract

The application discloses a marketing tactical recommendation system, a related method, a device and equipment. The system server module learns the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; and determining a target marketing session for the first user from a plurality of marketing sessions for the received marketing session data of the first user and the second user, and sending the target marketing session to the first client module; and displaying the target marketing communication through the first client module so that the second user can conveniently talk with the first user according to the target marketing communication. By adopting the processing mode, the marketing quality and efficiency can be effectively improved, so that the marketing success rate is improved.

Description

Marketing conversation recommendation system, method, device and equipment

Technical Field

The application relates to the technical field of data processing, in particular to a marketing speech recommendation system, a marketing speech recommendation method and device, a marketing speech extraction model processing method and device, a marketing speech processing method and device, a customer service robot system, a robot dialogue method and device and electronic equipment.

Background

In the field of marketing, including traditional manual traffic marketing, and new customer service robot systems, a front-line customer service staff typically replies to consumers based on their own experience and Standard Operating Procedure (SOP).

A typical process of developing a marketing campaign is as follows. The standard phonetics library is formed by manually inputting the standard phonetics and gradually accumulating the standard phonetics in a standard phonetics library mode. When in use, the method adopts a manual query mode to find the standard dialect of the reply. That is, customer service personnel reply to the customer by manually querying the standard dialogs in the standard operating process.

However, in the process of implementing the present invention, the inventors found that the above solution has at least the following problems: on one hand, the marketing response speed can be influenced by manually determining the proper dialect; on the other hand, customer service personnel levels vary, and manually determined standards are extremely limited and often slow to update, thus often giving consumers a better answer. In addition, standard dialogs do not necessarily fit well with current marketing scenarios. In conclusion, the prior art has the problem that the marketing quality and the marketing efficiency are low, so that the marketing failure rate is high.

Disclosure of Invention

The application provides a marketing tactics recommendation system to solve the problem that marketing quality and marketing efficiency are low in the prior art. The application further provides a marketing speech recommendation method and device, a marketing speech extraction model processing method and device, a marketing speech processing method and device, a customer service robot system, a robot dialogue method and device and electronic equipment.

The application provides a marketing tactics recommendation system, includes:

the server module is used for learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; and determining a target marketing session for the first user from a plurality of marketing sessions for the received marketing session data of the first user and the second user, and sending the target marketing session to the first client module;

and the first client module is used for displaying the target marketing communication so that the second user can conveniently talk with the first user according to the target marketing communication.

Optionally, the method further includes:

the server module is specifically used for sending the marketing dialogues determined by the marketing dialogues extraction model to the second client module; and receiving an audit result returned by the second client module, and if the audit result is yes, taking the marketing word as an effective marketing word;

and the second client module is used for displaying the marketing telephone operation, determining the auditing result of the third user on the marketing telephone operation and sending the auditing result to the server module.

Optionally, the server module is further configured to determine a marketing scenario to which the marketing session belongs; determining a marketing scene to which the marketing conversation belongs; and is specifically configured to determine the targeted marketing session from a plurality of marketing sessions corresponding to the marketing scenario to which the marketing session belongs, according to the marketing session data.

Optionally, the server module is specifically configured to learn from training data of multiple marketing scenes in a multitask learning manner to obtain the marketing tactical prediction model and the scene classification model; the scene classification model and the marketing tactical prediction model share a coding layer; the training data includes: marketing telephone traffic data, marking data of whether the marketing telephone traffic is the marketing telephone traffic or not and marking data of a marketing scene; and determining the marketing scene to which the marketing conversation belongs and the marketing scene to which the marketing conversation belongs through the scene classification model.

The application also provides a marketing tactics recommendation method, which comprises the following steps:

learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic to obtain a marketing telephone traffic extraction model;

selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model;

according to the marketing dialogue data of the first user and the second user, a target marketing dialogue aiming at the first user is determined from a plurality of marketing dialogues, so that the second user can conveniently dialogue with the first user according to the target marketing dialogue.

Optionally, the method further includes:

determining the auditing result of the marketing dialect determined by the marketing dialect extraction model by the third user;

if the result of the audit is yes, the marketing session is used as an effective marketing session.

Optionally, the method further includes:

determining a marketing scene to which a marketing conversation belongs;

determining a marketing scene to which the marketing dialogue belongs;

the method for determining the target marketing dialogs recommended to the second user for the first user from the plurality of marketing dialogs according to the marketing dialog data of the first user and the second user comprises the following steps:

and determining the target marketing communication from a plurality of marketing communication corresponding to the marketing scene to which the marketing communication belongs according to the marketing communication data.

Optionally, the learning of the marketing communication extraction model from the marketing communication data and the annotation data of whether the marketing communication data is the marketing communication comprises:

learning from training data of a plurality of marketing scenes in a multi-task learning mode to obtain the marketing tactical prediction model and the scene classification model; the scene classification model and the marketing tactical prediction model share a coding layer; the training data includes: marketing telephone traffic data, marking data of whether the marketing telephone traffic is the marketing telephone traffic or not and marking data of a marketing scene;

and determining the marketing scene to which the marketing conversation belongs and the marketing scene to which the marketing conversation belongs through the scene classification model.

Optionally, if the marketing session extraction condition is satisfied, selecting the marketing traffic data from the historical marketing traffic data as the marketing session through the marketing session extraction model.

Optionally, the marketing dialog extraction condition includes: the time length between the current time and the last marketing communication extraction is greater than a time length threshold value;

the historical marketing telephone traffic data comprises newly added marketing telephone traffic data after the last time of marketing telephone traffic extraction.

Optionally, the determining a targeted marketing session for the first user from a plurality of marketing sessions according to marketing session data of the first user and the second user includes:

determining the matching degree of the contextual dialogue data associated with the marketing dialogue and the marketing dialogue data;

and determining the target marketing skills according to the matching degree.

The application also provides a marketing tactics extraction model processing method, which comprises the following steps:

determining a training data set; the training data includes: marketing telephone traffic data, whether the marketing telephone traffic data is marking data of the marketing telephone traffic;

constructing a network structure of a marketing tactic extraction model;

and learning from a training data set to obtain the network parameters of the marketing strategy extraction model.

Optionally, the network structure includes: a phonetics feature extractor and a phonetics discriminator;

the conversational feature extractor is used for determining conversational feature data of the marketing telephone traffic data;

and the conversational technology discriminator is used for judging whether the marketing telephone traffic data is the marketing conversational technology or not according to the conversational technology feature data.

Optionally, the linguistic feature extractor includes: a word embedding layer, a text segment embedding layer, a coding layer and a language feature aggregation layer;

the word embedding layer is used for determining word vectors in the marketing traffic data;

the text segment embedding layer is used for determining text segment vectors in the marketing telephone traffic data according to the word vectors;

the coding layer is used for determining the coding data of the marketing telephone traffic data according to the text segment vector;

the speech characteristic aggregation layer is used for determining the speech characteristic data according to the coded data.

Optionally, the training data further includes: marketing scene annotation data;

the network architecture further comprises: a scene feature extractor and a scene classifier;

the scene feature extractor is used for determining scene feature data of the marketing telephone traffic data;

the scene classifier is used for determining a marketing scene to which the marketing telephone traffic data belongs according to the scene characteristic data;

the scene feature extractor includes: the word embedding layer, the text segment embedding layer, the coding layer and the scene feature aggregation layer;

the scene feature aggregation layer is used for determining the scene feature data according to the coded data;

learning from training data of a multi-marketing scene in a multi-task learning mode to obtain the marketing tactical prediction model and a scene classification model; the scene classification model includes the scene feature extractor and the scene classifier.

Optionally, the training data further includes: marketing domain annotation data;

the network architecture further comprises: a domain feature extractor and a domain classifier;

the domain feature extractor is used for determining domain feature data of the marketing telephone traffic data;

the domain classifier is used for determining the marketing domain to which the marketing telephone traffic data belongs according to the domain feature data;

the domain feature extractor includes: the word embedding layer, the text segment embedding layer, the coding layer and the domain feature aggregation layer;

the domain feature aggregation layer is used for determining the domain feature data according to the coded data;

learning from training data in a multi-marketing field in a multi-task learning mode to obtain the marketing tactical prediction model and the field classification model; the domain classification model comprises the domain feature extractor and the domain classifier.

Optionally, the multiple marketing areas include: the communication operator field, the E-business field, the education and training field and the insurance field.

The application also provides a marketing communication processing method, which comprises the following steps:

determining historical marketing traffic data;

extracting a conversational feature extractor in the model through marketing conversational to determine conversational feature data of historical marketing traffic data;

and extracting a word technology discriminator in the model through the marketing word technology, and judging whether the historical marketing word traffic data is the marketing word technology or not according to the characteristic data.

Optionally, the method further includes:

determining a marketing scene to which the historical marketing traffic data belongs;

and taking the historical marketing traffic data which is judged as the marketing traffic of the marketing scene.

Optionally, the method further includes:

learning from training data of a plurality of marketing scenes in a multi-task learning mode to obtain the marketing word operation extraction model and the scene classification model; the scene classification model includes: the scene feature extractor and the scene classifier share a coding layer; the training data includes: marketing telephone traffic data, marking data of whether the marketing telephone traffic is a marketing telephone operation or not and marketing scene marking data;

and determining the marketing scene to which the historical marketing traffic data belongs through the scene classification model.

Optionally, the method further includes:

learning from training data of a plurality of marketing fields in a multi-task learning mode to obtain the marketing tactical extraction model and the field classification model; the domain classification model comprises: the system comprises a domain feature extractor and a domain classifier, wherein the domain feature extractor and the conversational feature extractor share a coding layer; the training data includes: marketing telephone traffic data, marking data of whether the marketing telephone traffic is a marketing telephone operation or not and marking data of a marketing field;

and determining the marketing field to which the historical marketing traffic data belongs through a field classification model.

Optionally, the determining historical marketing traffic data includes:

acquiring marketing dialogue data between a first user and a second user in a historical marketing process;

and taking the historical marketing conversation data of the second user as historical marketing traffic data.

Optionally, the historical marketing traffic data includes: marketing traffic voice data;

the method further comprises the following steps:

and converting the marketing telephone traffic voice data into a marketing telephone traffic text through a voice recognition algorithm.

Optionally, the method further includes:

the noise data in the historical marketing traffic data is cleared according to the noise data filtering rules.

Optionally, the method further includes:

and performing standardization processing on the historical marketing traffic data according to the marketing traffic data standardization rule.

receiving a marketing telephone operation determined from historical marketing telephone operation data by a marketing telephone operation extraction model sent by a server module; the marketing communication extraction model is obtained by learning from a plurality of marketing communication data and the marking data of whether the marketing communication data is the marketing communication;

determining the auditing result of the marketing session of a third user;

and sending the auditing result to the server module so that the server module determines whether the marketing session is effective or not according to the auditing result.

receiving a target marketing conversation aiming at a first user and sent by a server module; the server module determines the target marketing session in the following manner: learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; and determining a targeted marketing session from a plurality of marketing sessions for the received marketing session data of the first user and the second user;

and displaying the target marketing communication so that the second user can conveniently talk with the first user according to the target marketing communication.

The present application further provides a marketing conversation recommendation system, including:

the system comprises at least one first service end module, a first server end module and a second server end module, wherein the first service end module is used for sending historical marketing traffic data of a target field; and sending marketing dialog data of the first user and the second user;

the second server module is used for learning the marketing telephone traffic data of the multiple marketing fields and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multiple marketing fields; selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic of a target field; and determining a target marketing conversation for the first user from a plurality of marketing conversations in a target field according to the marketing conversation data, and transmitting the target marketing conversation;

and the client module is used for displaying the target marketing communication so that a second user can conveniently talk with the first user according to the target marketing communication.

Optionally, the first service module is specifically configured to send the historical marketing traffic data according to a preset time interval, where the historical marketing traffic data includes marketing traffic data newly added to a target field in the time interval.

Optionally, the target fields corresponding to different first service end modules include: the communication operator field, the E-business field, the education and training field and the insurance field.

sending the historical marketing telephone traffic data of the target field to the server module, so that the server module extracts the model through the marketing telephone traffic, and selects the marketing telephone traffic data from the historical marketing telephone traffic data as the marketing telephone traffic of the target field; the server module learns the marketing telephone traffic data of the multiple marketing fields and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multiple marketing fields;

and sending the marketing dialogue data of the first user and the second user to the server module, so that the server module determines a target marketing dialogue aiming at the first user from a plurality of marketing dialogues in a target field according to the marketing dialogue data, and the second user can conveniently talk with the first user according to the target marketing dialogue.

learning from marketing telephone traffic data of multiple marketing fields and marking data of whether the marketing telephone traffic data is marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multiple marketing fields;

aiming at the received historical marketing telephone traffic data of the target field, selecting the marketing telephone traffic data from the historical marketing telephone traffic data as the marketing telephone traffic of the target field through a marketing telephone traffic extraction model;

and aiming at the received marketing dialogue data of the first user and the second user, determining a target marketing dialogue aiming at the first user from a plurality of marketing dialogues in a target field so as to facilitate the second user to dialogue with the first user according to the target marketing dialogue.

receiving a target marketing conversation aiming at a first user and sent by a server module; the server module determines a target marketing session in the following way: learning from marketing telephone traffic data of multiple marketing fields and marking data of whether the marketing telephone traffic data is marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multiple marketing fields; aiming at the received historical marketing telephone traffic data of the target field, selecting the marketing telephone traffic data from the historical marketing telephone traffic data as the marketing telephone traffic of the target field through a marketing telephone traffic extraction model; determining a target marketing conversation for the first user from a plurality of marketing conversations in a target field for the received marketing conversation data of the first user and the second user;

The application also provides a customer service robot system, including:

the server module is used for learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; and determining a target marketing conversation for the target user from the plurality of marketing conversations for the received marketing conversation data of the target user, and sending the target marketing conversation to the client module;

and the client module is used for displaying the target marketing words.

The application also provides a robot dialogue method, which comprises the following steps:

and determining a target marketing dialogue aiming at the target user from the plurality of marketing dialogs according to the marketing dialogue data of the target user.

receiving a target marketing conversation aiming at a target user and sent by a server module; the server module determines the target marketing communication in the following way: learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; determining a target marketing conversation aiming at the target user from a plurality of marketing conversations according to the marketing conversation data of the target user;

and displaying the target marketing words.

The application also provides a marketing word recommendation device, including:

the model construction unit is used for learning and obtaining a marketing telephone operation extraction model from the marketing telephone operation data and the marking data of whether the marketing telephone operation data is the marketing telephone operation;

the system comprises a dialogue mining unit, a marketing dialogue extracting unit and a marketing dialogue extracting unit, wherein the dialogue mining unit is used for extracting a model through marketing dialogue and selecting marketing dialogue data from historical marketing dialogue data as marketing dialogue;

and the word operation recommending unit is used for determining a target marketing word operation aiming at the first user from a plurality of marketing word operations according to the marketing dialogue data of the first user and the second user so as to facilitate the second user to dialogue with the first user according to the target marketing word operation.

the system comprises a recommended word receiving unit, a target marketing word sending unit and a target word receiving unit, wherein the recommended word receiving unit is used for receiving the target marketing word aiming at a first user and sent by a server module; the server module determines the target marketing session in the following manner: learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic; and determining a targeted marketing session from a plurality of marketing sessions for the received marketing session data of the first user and the second user;

and the recommendation language and technology display unit is used for displaying the target marketing language and facilitating the conversation between the second user and the first user according to the target marketing language and technology.

The application also provides a marketing words art extraction model processing apparatus, includes:

a training data determination unit for determining a training data set; the training data includes: marketing telephone traffic data, whether the marketing telephone traffic data is marking data of the marketing telephone traffic;

the network structure construction unit is used for constructing a network structure of the marketing tactics extraction model;

and the model training unit is used for learning from a training data set to obtain the network parameters of the marketing strategy extraction model.

The present application further provides a marketing communication processing apparatus, including:

the traffic data determining unit is used for determining historical marketing traffic data;

the conversational feature extraction unit is used for extracting a conversational feature extractor in the model through marketing conversational to determine conversational feature data of historical marketing traffic data;

and the conversational judging unit is used for extracting the conversational discriminator in the model through the marketing conversational technology and judging whether the historical marketing conversational data is the marketing conversational technology or not according to the characteristic data.

the system comprises a speech receiving unit, a marketing speech extracting unit and a speech processing unit, wherein the speech receiving unit is used for receiving the marketing speech determined by a marketing speech extracting model from historical marketing speech data sent by a server module; the marketing telephone operation extraction model is obtained by learning from a plurality of marketing telephone operation data and the marking data of whether the marketing telephone operation data is the marketing telephone operation or not;

the word operation auditing unit is used for determining the auditing result of the marketing word operation of a third user;

and the auditing result sending unit is used for sending the auditing result to the server module so that the server module can determine whether the marketing session is effective or not according to the auditing result.

the historical traffic data sending unit is used for sending the historical marketing traffic data of the target field to the server module, so that the server module extracts the model through marketing traffic, and selects the marketing traffic data from the historical marketing traffic data as the marketing traffic of the target field; the server module learns the marketing telephone traffic data of the multiple marketing fields and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multiple marketing fields;

the current dialogue data sending unit is used for sending marketing dialogue data of the first user and the second user to the server module, so that the server module determines a target marketing dialogue for the first user from a plurality of marketing dialogues in a target field according to the marketing dialogue data, and the second user can conveniently talk with the first user according to the target marketing dialogue.

the model construction unit is used for learning from marketing telephone traffic data of multiple marketing fields and marking data of whether the marketing telephone traffic data is marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multiple marketing fields;

the system comprises a word operation mining unit, a word operation extracting unit and a word operation extracting unit, wherein the word operation mining unit is used for extracting a model according to received historical marketing telephone traffic data of a target field and selecting the marketing telephone traffic data from the historical marketing telephone traffic data as marketing word operations of the target field;

and the word operation recommending unit is used for determining a target marketing word operation aiming at the first user from a plurality of marketing word operations in a target field aiming at the received marketing dialogue data of the first user and the second user so as to facilitate the second user to dialogue with the first user according to the target marketing word operation.

the system comprises a recommended word receiving unit, a target marketing word sending unit and a target word receiving unit, wherein the recommended word receiving unit is used for receiving the target marketing word aiming at a first user and sent by a server module; the server module determines a target marketing session in the following way: learning from marketing telephone traffic data of multiple marketing fields and marking data of whether the marketing telephone traffic data is marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multiple marketing fields; aiming at the received historical marketing telephone traffic data of the target field, selecting the marketing telephone traffic data from the historical marketing telephone traffic data as the marketing telephone traffic of the target field through a marketing telephone traffic extraction model; determining a target marketing conversation for the first user from a plurality of marketing conversations in a target field for the received marketing conversation data of the first user and the second user;

The present application further provides a robot dialogue device, including:

the traffic mining unit is used for extracting a model through marketing traffic, and selecting marketing traffic data from historical marketing traffic data as marketing traffic;

and the word operation recommending unit is used for determining the target marketing word operation aiming at the target user from a plurality of marketing word operations according to the marketing word operation data of the target user.

The present application further provides a robot dialogue device, including:

the recommendation conversation receiving unit is used for receiving the target marketing conversation aiming at the target user and sent by the server module; the server module determines the target marketing communication in the following way: learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic; determining a target marketing conversation aiming at the target user from a plurality of marketing conversations according to the marketing conversation data of the target user;

and the recommendation language and technology display unit is used for displaying the target marketing language and technology.

The present application further provides an electronic device, comprising:

a processor and a memory;

a memory for storing a program for implementing the above method, the apparatus being powered on and the program for executing the method by the processor.

constructing a marketing conversation library according to historical marketing conversation data;

acquiring current marketing dialogue data;

and determining a target marketing conversation from the marketing conversation library according to the current marketing conversation data.

Optionally, the determining a target marketing conversation from a marketing conversation library according to the current marketing conversation data includes:

determining the matching degree of the historical contextual dialog data associated with the marketing dialog and the current marketing dialog data;

and determining the target marketing words according to the matching degree.

The present application also provides a computer-readable storage medium having stored therein instructions, which when run on a computer, cause the computer to perform the various methods described above.

The present application also provides a computer program product comprising instructions which, when run on a computer, cause the computer to perform the various methods described above.

Compared with the prior art, the method has the following advantages:

the marketing communication recommendation system provided by the embodiment of the application learns the marketing communication data and the marking data of whether the marketing communication data is the marketing communication through the server module to obtain a marketing communication extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; and determining a target marketing session for the first user from a plurality of marketing sessions for the received marketing session data of the first user and the second user, and sending the target marketing session to the first client module; and displaying the target marketing communication through the first client module so that the second user can conveniently talk with the first user according to the target marketing communication. By adopting the processing mode, the telephone traffic containing the excellent speech technology is screened and filtered through mining the daily telephone traffic data of excellent customer service staff groups, and higher-quality and richer marketing speech technologies are recommended to the customer service marketing staff in real time based on the mined excellent speech technology in the process of carrying out marketing activities on the customers (first users) by the customer service marketing staff (second users), so that the customer service marketing staff is assisted to carry out marketing activities; therefore, the marketing quality and efficiency can be effectively improved, and the marketing success rate is improved. In addition, the language recommendation processing process adopted by the system is high in reusability, only data in different fields (such as the operator field, the e-commerce field, the education institution field, the insurance field and the like) need to be replaced, and the structure of the marketing language extraction model can be directly reused, so that the expandability of the marketing field can be effectively improved.

Further, the marketing session recommendation system can also send the marketing session determined by the marketing session extraction model to the second client module through the server module; displaying the marketing talk through the second client module, determining the auditing result of the third user to the marketing talk, and sending the auditing result to the server module; if the result of the audit is yes, the server-side module takes the marketing communication as an effective marketing communication. By adopting the processing mode, after the telephone traffic possibly containing the excellent speech operation is screened and filtered from the daily telephone traffic data of the excellent and beautiful customer service personnel group, the speech operation processing personnel manually compares the speech operation with the reduced range, and finally determines the excellent speech operation which is beneficial to improving the marketing service capability.

Further, the server module can also determine a marketing scenario to which a marketing session belongs; determining a marketing scene to which the marketing conversation belongs; and determining the target marketing session from a plurality of marketing sessions corresponding to the marketing scene to which the marketing session belongs according to the marketing session data. By adopting the processing mode, the context scene of customer service in the telephone traffic can be identified and sensed in real time, and the conversational recommendation service can be provided for marketing activities in a plurality of scenes; therefore, the scene service capability can be effectively improved.

Furthermore, the server module can learn from training data of multiple marketing scenes in a multi-task learning mode to obtain a marketing tactical prediction model and a scene classification model; the scene classification model and the marketing tactical prediction model share a coding layer; the training data includes: marketing telephone traffic data, marking data of whether the marketing telephone traffic is the marketing telephone traffic or not and marking data of a marketing scene; and determining the marketing scene to which the marketing conversation belongs and the marketing scene to which the marketing conversation belongs through the scene classification model. By adopting the processing mode, a marketing dialect extraction model and a scene classification model can be obtained by learning at the same time through one-time model training processing, and the marketing dialect extraction model and the scene classification model share a coding layer; therefore, the application efficiency of the cross-scene can be effectively improved.

Furthermore, if the coding layer adopts a network structure which can learn the long-distance semantic dependence of the context, the problem of keyword ambiguity can be solved, and therefore the speech mining accuracy can be further improved.

Further, the server module may further select, through the marketing session extraction model, marketing traffic data serving as the marketing session from the historical marketing traffic data when the marketing session extraction condition is satisfied, where the marketing session extraction condition is that a time between a current time and a last time of the marketing session extraction is greater than a time threshold (e.g., 1 day, 7 days, 1 month, and the like), and the historical marketing traffic data includes newly added marketing traffic data (e.g., newly added traffic data within a time period of 1 day, 7 days, 1 month, and the like) after the last marketing session extraction. By adopting the processing mode, the marketing communication can be supplemented according to the incremental traffic data at regular intervals; therefore, the richness of the dialects can be effectively improved, the marketing quality is improved, and meanwhile, the computing resources can be saved.

The marketing communication processing method provided by the embodiment of the application determines historical marketing communication data; extracting a conversational feature extractor in the model through marketing conversational to determine conversational feature data of historical marketing traffic data; and extracting a conversational technology discriminator in the model through marketing conversational technology, and judging whether the historical marketing telephone traffic data is the marketing conversational technology or not according to the conversational technology feature data. By adopting the processing mode, at least the following technical effects can be achieved:

1) Because the marketing telephone traffic extraction model is established by using the prior knowledge, and certain specific templates/styles, such as characteristic data of excellent telephone traffic, in the marketing telephone traffic data are automatically learned and fitted by adopting the neural network, the accuracy of telephone traffic mining can be effectively improved;

2) By the model, the daily telephone traffic data of excellent customer service personnel groups can be mined, and telephone traffic containing excellent dialogs can be screened and filtered out, so that the richness of the dialogs can be effectively improved;

3) The stability of the dialect mining can be effectively improved because the adjustment of a heuristic algorithm threshold value is not involved;

4) The method is an extraction-type word mining method, existing expressions can be extracted from the marketing word original text and standardized display can be carried out, and the situation that the word mining result is uncontrollable does not exist, so that the word quality and readability can be effectively improved;

5) Because the marketing dialect is determined without depending on a word bank, the generalization of the marketing dialect mining can be effectively improved.

6) If the training data of the marketing word operation extraction model comprises marketing word operation texts in multiple scenes, different marketing scenes can share one marketing word operation extraction model to carry out mining processing on the marketing word operation, and construction of multiple marketing word operation extraction models suitable for different scenes is avoided, so that cross-scene application efficiency of marketing word operation mining can be improved.

7) If the training data of the marketing dialect extracting model comprises a multi-scene marketing telephone traffic text, and the marketing dialect extracting model and the scene classification model are simultaneously learned in a multi-task learning mode, the two models share the encoder, so that the cross-scene application efficiency of marketing dialect mining can be further improved.

8) If the encoder adopts a network structure which can learn the long-distance semantic dependence of the context, the problem of keyword ambiguity can be solved, so that the speech mining accuracy can be further improved;

9) The reusability of the dialect mining process is high, data in different fields (such as an operator field and an e-commerce field) only need to be replaced, and the model structure can be directly reused, so that the expandability of the dialect mining field can be effectively improved.

The marketing tactics extraction model processing method provided by the embodiment of the application determines a training data set; constructing a network structure of a marketing tactic extraction model; and learning from the training data set to obtain the network parameters of the marketing strategy extraction model. By adopting the processing mode, a marketing speech extraction model is established by using the prior knowledge, the model adopts a neural network to automatically learn and fit certain specific templates/styles, such as characteristic data of excellent speech, in a marketing speech text, and the model can be used for mining daily speech data of excellent customer service personnel groups, and screening and filtering out the speech with excellent speech; therefore, the accuracy, richness, stability and generalization of the dialect mining can be effectively improved, and the quality and readability of the dialect can be improved.

The marketing telephone traffic recommendation system provided by the embodiment of the application sends historical marketing telephone traffic data of a target field through at least one first service end module; learning from the marketing telephone traffic data of the multi-marketing field and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not through a second server module to obtain a marketing telephone traffic extraction model of the multi-marketing field; selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic of a target field; sending marketing dialogue data of the first user and the second user through at least one first service end module; determining a target marketing conversation aiming at the first user from a plurality of marketing conversations in a target field through a second server module according to the marketing conversation data, and sending the target marketing conversation; and displaying the target marketing communication through the client module so that a second user can conveniently talk with the first user according to the target marketing communication. By adopting the processing mode, real-time speech technology recommendation service is provided for third-party marketing activities in multiple fields through the universal speech technology recommendation platform, cross-field real-time speech technology recommendation is realized, wherein the speech technology recommendation platform is used for mining daily traffic data of excellent customer service personnel groups of the third-party marketing activities in multiple fields, screening and filtering out the traffic containing excellent speech technology to form a marketing speech technology library in multiple marketing fields, and in the process of carrying out marketing activities on customers by marketing personnel, higher-quality and richer marketing speech technologies are recommended to the customer service marketing personnel in real time based on the mined excellent speech technology to assist the customer service personnel to carry out marketing activities; therefore, the field service capability can be effectively improved, the marketing quality and efficiency can be improved, and the marketing success rate can be improved.

Further, the first server module may send the historical marketing traffic data according to a preset time interval, where the historical marketing traffic data includes marketing traffic data newly added to a target field in the time interval, for example, the newly added traffic data is sent once a day, and the newly added traffic data is mined once a day. By adopting the processing mode, the marketing communication can be supplemented according to the incremental traffic data at regular intervals; therefore, the language and skill richness can be effectively improved, and the marketing quality is improved.

The customer service robot system provided by the embodiment of the application learns the marketing telephone operation extraction model from the marketing telephone operation data and the marking data of whether the marketing telephone operation data is the marketing telephone operation or not through the server module; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; and determining a target marketing conversation for the target user from the plurality of marketing conversations for the received marketing conversation data of the target user, and sending the target marketing conversation to the client module; and displaying the target marketing words through the client module. By adopting the processing mode, the telephone traffic containing the excellent speech technology is screened and filtered by mining the daily telephone traffic data of excellent customer service personnel groups, and in the process of carrying out marketing activities on customers (target users) by the customer service robot, the higher-quality and richer marketing speech technology is determined in real time based on the mined excellent speech technology, and the marketing activities are carried out automatically by the customer service robot; therefore, the marketing quality and efficiency can be effectively improved, and the marketing success rate is improved.

According to the marketing conversation recommendation method provided by the embodiment of the application, a marketing conversation library is constructed according to historical marketing conversation data; acquiring current marketing dialogue data; and determining a target marketing conversation from the marketing conversation library according to the current marketing conversation data. By adopting the processing mode, the telephone traffic containing the excellent speech technology is screened and filtered by mining the daily telephone traffic data of excellent customer service personnel groups, and the marketing speech technology is recommended in real time based on the mined excellent speech technology when marketing activities are carried out on consumers, so that a customer service robot system capable of carrying out marketing activities fully automatically can be realized, and a marketing speech recommendation system capable of assisting customer service marketing personnel to carry out marketing activities can also be realized; therefore, the marketing quality and efficiency can be effectively improved, and the marketing success rate is improved.

Drawings

FIG. 1 is a schematic diagram of an application scenario of an embodiment of a marketing tactical recommendation system provided in the present application;

FIG. 2 is a schematic diagram of a device interaction of an embodiment of a marketing tactical recommendation system provided by the present application;

FIG. 3 is a schematic diagram of a marketing tactical extraction model of an embodiment of a marketing tactical recommendation system provided by the present application;

FIG. 4 is a schematic flow chart diagram illustrating an embodiment of a marketing strategy recommendation method provided by the present application;

FIG. 5 is a schematic flow chart diagram of an embodiment of a marketing tactical extraction model processing method provided in the present application;

FIG. 6 is a schematic flow chart diagram of an embodiment of a marketing session processing method provided by the present application;

FIG. 7 is a schematic flow chart diagram of an embodiment of a marketing dialog processing method provided by the present application;

FIG. 8 is a schematic flow chart diagram of an embodiment of a marketing tactical recommendation method provided herein;

FIG. 9 is a schematic diagram of another application scenario of an embodiment of a marketing tactical recommendation system provided in the present application;

fig. 10 is a schematic view of an application scenario of an embodiment of a customer service robot system provided in the present application.

Detailed Description

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. This application is capable of implementation in many different ways than those herein set forth and of similar import by those skilled in the art without departing from the spirit of this application and is therefore not limited to the specific implementations disclosed below.

In the application, a marketing communication recommendation system, a marketing communication recommendation method and device, a marketing communication extraction model processing method and device, a marketing communication processing method and device, a customer service robot system, a robot conversation method and device and electronic equipment are provided. Each of the schemes is described in detail in the following examples.

First embodiment

The marketing tactics recommendation system provided by the embodiment comprises: the system comprises a server module and a first client module. The server module can be deployed on the server side, and the first client module can be deployed on the terminal equipment side used by a marketer (a second user).

The server may be a server deployed on a cloud server, or a server deployed on a server dedicated to providing marketing call recommendation service.

The terminal device includes but is not limited to: personal computer, also including mobile communication equipment, namely: the mobile phone or the smart phone, and the terminal devices such as the smart speaker, the PAD, and the iPad.

Please refer to fig. 1, which is a schematic view of an application scenario of the marketing tactical recommendation system of the present application. In this embodiment, a customer service marketing person (a second user) of a certain communication operator needs to market a service package (such as a 5G package, broadband installation, recharging and paying) to a mobile phone user (a first user), and in a process of a call between the customer service marketing person and the mobile phone user, a server side can send a recommended marketing communication to a personal computer used by the marketing person in real time according to a real-time conversation flow between two users and a plurality of marketing communication technologies corresponding to the service package mined from daily traffic data of a premium customer service person group, and the marketing person can reply to a consumer according to the recommended marketing communication. Wherein, the marketing communication is the model document of communication between the sales personnel and the customers, which depends on the category of the sales industry. Sales are largely science of language, and the core wisdom of sales is the conversational skill of sales.

As shown in fig. 1, the server may first establish a marketing telephone traffic extraction model by using whether a part of historical marketing telephone traffic data related to the service package is labeled data of marketing telephone traffic, and the model may automatically learn and fit some specific templates/styles in the marketing telephone traffic data by using a neural network; then, a marketing telephone art corresponding to the service package can be mined out from another part of historical marketing telephone traffic text related to the service package through the model, so that the mining of daily telephone traffic data of excellent customer service personnel groups can be realized, and telephone traffic containing the excellent telephone art can be screened and filtered out.

Please refer to fig. 2, which is a schematic diagram illustrating an apparatus interaction of the marketing tactical recommendation system of the present application. In this embodiment, the server module learns the marketing telephone traffic extraction model from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; and determining a target marketing session for the first user from a plurality of marketing sessions for the received marketing session data of the first user and the second user, and sending the target marketing session to the first client module; the first client module displays the target marketing communication so that the second user can conveniently talk with the first user according to the target marketing communication.

It can be seen that the process of the system comprises three phases. Wherein, the first stage is a stage of constructing a marketing conversation extraction model for a server; the second stage is that the server side extracts a model through marketing communication and digs marketing communication from historical marketing communication data; the third stage is a stage of recommending the real-time conversation based on the marketing conversation mined in the previous stage when the marketing personnel develop marketing activities to the consumers. The processing procedure of these three stages will be described below.

The first stage is as follows: and constructing a marketing tactics extraction model.

The marketing tactical prediction model can be obtained by learning from a training data set by adopting a supervised machine learning algorithm. The training data may include: marketing traffic data, and whether the marketing traffic data is tagging data for marketing traffic. Wherein the marketing traffic data may be derived from historical marketing traffic data of customer service marketers. Specifically, the marketing traffic data may be marketing dialog data that can be precipitated as excellent customer service marketers of marketing dialogs, such as a consumer (first user): "we have done the broadband", second user (customer service marketer): "this is irrelevant, one surfs the net, one watches TV, two is not wrong, it is very smooth, and we have paid the preference of xxx yuan in the present broadband (keep the bore); or general marketing dialogue data of excellent customer service marketers or common customer service marketers, such as customers: "hello", customer service: "do you ask you about the owner of 139 xxx". In specific implementation, whether the marketing telephone traffic data is the marketing telephone traffic can be identified and marked manually.

In this embodiment, the server module may construct the marketing tactical prediction model as follows: 1) Determining a training data set; 2) Constructing a network structure of a marketing tactical prediction model; 3) And taking the marketing telephone traffic data as input data of the marketing telephone traffic prediction model, taking the marked data of whether the marketing telephone traffic data is the marketing telephone traffic as output data of the marketing telephone traffic prediction model, and training network parameters of the marketing telephone traffic prediction model.

1) A training data set is determined.

In one example, marketing traffic data in the training data may be determined as follows: acquiring marketing dialogue data between a first user and a second user in a historical marketing process, wherein the marketing dialogue data comprises dialogue data of the first user and dialogue data of the second user; and taking the marketing dialogue data of the second user as the marketing traffic data. By adopting the processing mode, the marketing conversation data of the customer service staff can be extracted from the historical marketing conversation data to form the marketing telephone traffic data in the training data, so that the occurrence of the conversation data of the consumers irrelevant to the marketing telephone traffic is avoided, and the effectiveness of the training data can be effectively improved.

In practical applications, the marketing session data between the first user and the second user is typically multi-turn question and answer data. In this case, to extract the marketing dialogue data of the second user from the multiple rounds of question and answer data, the following method can be adopted: executing clause processing on the multi-turn question-answer data to obtain marketing clause data; determining user role information of marketing clause data; and taking the marketing clause data of which the user role information is the second user as marketing telephone traffic data in the training data.

In practical applications, the original historical marketing traffic data of the customer service marketer is usually the marketing traffic data in the form of voice, that is: marketing traffic voice data. In this case, to determine the marketing traffic data in the training data, the following steps may be further included: and converting the marketing telephone traffic voice data into marketing telephone traffic text data serving as the marketing telephone traffic data in the training data through a voice recognition algorithm. Since the speech recognition algorithm belongs to the mature prior art, it is not described here in detail.

In practice, the original historical marketing traffic data of the customer service marketer also typically includes some noise data, such as web page link addresses, order identifications, communication numbers, emoticons, order numbers, and the like. In this case, to determine the marketing traffic data in the training data, the following steps may be further included: and clearing the noise data in the marketing traffic data according to the noise text filtering rule. The noise data filtering rules include, but are not limited to: and clearing the webpage link address, the order identification, the communication number, the emoticon, the order number and the like.

In practical applications, the original historical marketing traffic data of the customer service marketer also usually includes some non-standardized conversation contents, such as stop words, multiple synonyms for expressing the same meaning, spoken language expression forms (mainly spoken words, dialect common phrases, etc.), and the like. In this case, to determine the marketing traffic data in the training data, the following steps may be further included: and executing standardization processing on the marketing traffic data according to the marketing traffic data standardization rule. The marketing traffic data normalization rules include, but are not limited to: filtering stop words, normalizing synonyms, and performing normalization expression on the form of spoken expression (mainly language words, dialect common phrases and the like).

Table 1 shows training data for a single marketing scenario.

As can be seen from table 1, in the specific implementation, the marketing telephone traffic prediction model can be obtained by learning from the training data set of a single marketing scenario, and the marketing telephone traffic prediction model thus constructed is suitable for mining the marketing telephone traffic from the historical marketing telephone traffic data of the single marketing scenario.

2) And constructing a network structure of the marketing tactical prediction model.

In this embodiment, the marketing tactical prediction model is learned from a training data set using a supervised machine learning algorithm. The marketing telephone traffic extracting model can adopt a neural network structure to automatically learn and fit some specific templates/styles in the input marketing telephone traffic data, such as characteristic data of excellent telephone traffic. The marketing language extraction model comprises a language feature extractor and a language discriminator. The conversational feature extractor is used for determining conversational feature data of the marketing telephone traffic data, and the conversational feature data comprise marketing conversational features of the fitted labeling data; and the session discriminator is used for judging whether the marketing telephone traffic data is the marketing session according to the characteristic data.

The marketing tactics extraction model can adopt a neural network structure with a layered design so as to improve the design flexibility and reusability. In one example, the feature extractor comprises: word embedding layer, coding layer, speech technology characteristic aggregation layer, the speech technology discriminator includes: a full connection layer and a speech discrimination layer.

Wherein the word embedding layer is to determine word vectors in the model input data "marketing traffic data". The marketing traffic data comprises a plurality of words, different words correspond to different word vectors, and the word vectors comprise semantic information of the words. The word vector may be a word vector determined during training of the marketing word extraction model.

In specific implementation, an existing Word segmentation technology (such as jieba Word segmentation) can be selected to determine a plurality of words in the marketing traffic data, algorithms such as BERT, ELMO, word2vec, fastText and the like which are common in the prior art can be selected to generate Word vectors, and a FastText algorithm is adopted to generate the Word vectors in the embodiment. Since the word vector and the determination method thereof belong to the mature prior art, they are not described herein again.

And the coding layer is used for determining the coding data of the marketing traffic data according to the word vector. That is, the coding layer may be responsible for vectoring characterization and characterization of the incoming marketing traffic data. The coding layer can adopt a bidirectional long-short term time sequence model (BilSTM) and can also adopt a neural network with other structures as long as the coding function of the marketing telephone traffic data can be realized.

The conversational feature aggregation layer is to determine conversational feature data for the marketing traffic data based on the encoded data. That is, the aggregation layer may output marketing tactical features of the fitted annotation data.

The telephone operation discriminator can judge whether the marketing telephone traffic data is the marketing telephone operation according to the characteristic data output by the full connection layer. The phonetics judging layer completes a two-classification task and judges whether the input data of the model is excellent phonetics or not. In specific implementation, the speech discrimination layer may be implemented by using a logistic regression function (e.g., softmax function), or may be implemented by using other classifiers.

In another example, the tactical feature extractor may further comprise: a text segment embedding layer. The text segment embedding layer is used for determining text segment vectors in the marketing telephone traffic data according to the word vectors; correspondingly, the coding layer is used for determining the coding data of the marketing traffic data according to the text segment vector. The text segment vector may be a vector of text segments formed by a plurality of adjacent words in the marketing traffic data.

Please refer to fig. 3, which is a specific diagram of a marketing tactical extraction model of the marketing tactical recommendation system of the present application. As can be seen from fig. 3, the coding layer may employ a Depth Pyramid Convolutional Neural Network (DPCNN), and the morphological feature extractor includes: a text segment embedding layer. By adopting the processing mode, the long-distance semantic dependence of the marketing telephone traffic data context can be learned, so that the problem of keyword ambiguity can be solved, and the accuracy of the telephone traffic mining can be effectively improved. For example: the word "apple" can refer to both mobile phones and fruits, and the word is ambiguous and needs to see what the context of its surroundings says.

3) And taking the marketing telephone traffic data as input data of the marketing telephone traffic prediction model, taking the marked data of whether the marketing telephone traffic data is the marketing telephone traffic as output data of the marketing telephone traffic prediction model, and training network parameters of the marketing telephone traffic prediction model.

After the training data set is determined and the network structure of the marketing telephone traffic text model is established, network parameters of the marketing telephone traffic text model can be obtained by learning from the training data set in a supervised machine learning mode. In specific implementation, the marketing telephone traffic data can be used as input data of the model, the marking data of whether the marketing telephone traffic data is the marketing telephone traffic data or not is used as output data of the model, the loss value is calculated, and the training is finished after the loss value reaches the optimization target. Since the machine learning method belongs to the mature prior art, it is not described here again.

So far, the basic network structure and model construction method of the marketing tactics extraction model are explained. The following describes a preferred embodiment of constructing a marketing science and technology extraction model in connection with practical application.

In practical applications, a typical situation is that a plurality of products or services are generally marketed to consumers by an enterprise, for example, a communication carrier may provide services such as 5G package, broadband installation, recharging and paying for a mobile phone user. In this embodiment, a plurality of marketing objects of an enterprise are referred to as marketing scenarios, for example, the marketing scenarios associated with a communication carrier include a 5G package scenario, a broadband installation scenario, a recharge payment scenario, and the like.

In the face of the above practical application, the marketing tactical prediction model and the scene classification model are obtained by learning from training data of a plurality of marketing scenes in a multi-task learning manner. In this case, the training data further includes: marketing scene annotation data, the scene classification model comprising a scene feature extractor and a scene classifier. The scene feature extractor is used for determining scene feature data of the marketing traffic data; the scene classifier is used for determining the marketing scene to which the marketing telephone traffic data belongs according to the feature data, so that the excavated marketing telephone traffic can be further determined to belong to the marketing scene, and the corresponding relation between the marketing scene and the marketing telephone traffic is constructed.

In particular implementation, the scene feature extractor may include: the word embedding layer, the text segment embedding layer, the coding layer and the scene feature aggregation layer; and the scene characteristic aggregation layer is used for determining the scene characteristic data according to the coded data. The optimization objective for the entire model may include two loss functions, one being a marketing tactical loss function and the other being a scenario classification loss function. Table 2 shows training data for a multiple marketing scenario.

The system provided by the embodiment obtains the marketing communication extraction models and the marketing scene classification models suitable for a plurality of marketing scenes by simultaneously learning from training data of a plurality of marketing scenes in one model training process, so that different marketing scenes can share one marketing communication extraction model to carry out mining processing of marketing communication, and the mined marketing communication can be determined through the scene classification models.

In specific implementation, different marketing tactics extraction models can be constructed for different marketing scenes, and an independent scene classification model can be constructed at the same time. For example, different marketing language extraction models are learned from training data of different marketing scenes, and scene classification models are learned from scene marking data of marketing language data of a plurality of marketing scenes. The marketing conversation extraction model and the scene classification model can adopt the same coding layer network structure so as to improve the application efficiency of the cross-scene.

In practical application, another typical situation is that different enterprises need to market products or services to consumers, for example, communication operators can provide services such as 5G package, broadband installation, recharging and paying for mobile phone users, insurance enterprises need to sell dangerous varieties such as health insurance, accident insurance, life insurance and annuity insurance to consumers, education institutions like consumers recommend courses such as english system courses, novel reading courses and examination levels, and mobile phone sellers need to sell multiple types of mobile phones to consumers. The present embodiment refers to marketing objects of a plurality of enterprises as marketing fields, such as a communication operator field, an e-commerce field, an educational training field, an insurance field, and the like.

In the face of the above practical application, the marketing tactical prediction model and the field classification model are obtained by learning from training data in a multi-marketing field in a multi-task learning manner. In this case, the training data further includes: marketing domain annotation data, the domain classification model includes a domain feature extractor and a domain classifier. Wherein the domain feature extractor is to determine domain feature data of the marketing traffic data; the domain classifier is used for determining the marketing domain to which the marketing telephone traffic data belongs according to the domain feature data, so that the marketing domain to which the mined marketing telephone traffic belongs can be further determined, and the corresponding relation between the marketing domain and the marketing telephone traffic is constructed.

In particular, the domain feature extractor may include: the word embedding layer, the text segment embedding layer, the coding layer and the domain feature aggregation layer; and the domain feature aggregation layer is used for determining the domain feature data according to the coded data. The optimization objective for the entire model may include two loss functions, one for marketing tactical loss functions and the other for domain classification loss functions. Table 3 shows training data for multiple marketing domains, multiple marketing scenarios.

The system provided by the embodiment obtains the marketing jargon extracting models and the marketing jargon classifying models suitable for a plurality of marketing fields by simultaneously learning from the training data of the plurality of marketing fields in one model training process, so that different marketing fields can share one marketing jargon extracting model to carry out the mining processing of the marketing jargon, the marketing field to which the mined marketing jargon belongs can be determined through the field classifying model, and therefore the system can avoid the construction of the plurality of marketing jargon extracting models suitable for different marketing fields, and can also enable the field classifying model and the marketing jargon predicting model to share the encoder, and therefore the cross-field application efficiency of the marketing jargon mining can be improved.

So far, a construction method of the marketing tactical prediction model is explained, and a marketing tactical mining method is explained below.

The second stage is as follows: and mining the marketing jargon from the historical marketing telephone traffic text by the constructed marketing jargon extraction model.

According to the embodiment, through a constructed marketing conversation extraction model, daily traffic data of excellent customer service personnel groups are mined, and traffic containing excellent conversations is screened and filtered out. The result of the model training in the first stage is stored in a certain mode, and in the second stage, through loading and restoring network parameters, corresponding marketing telephone traffic scores (such as probabilities) are given to any input historical marketing telephone traffic data to be identified as marketing telephone traffic, and if the score meets a threshold value, the historical marketing telephone traffic data is used as a marketing telephone traffic library.

For example, the historical marketing traffic data to be identified as whether the marketing traffic is "you are relieved, the broadband is almost not open-line at present, the family of the cell is generally open, and the house of the house owner is not affected. ", the marketing tactic data is determined to be the marketing tactic if the model score is extracted by the marketing tactic 95.

In this embodiment, to mine marketing jargon from the historical marketing traffic text through the constructed marketing jargon extraction model, the following method may be adopted: 1) Determining historical marketing traffic data; 2) Extracting a conversational feature extractor in the model through marketing conversational to determine conversational feature data of historical marketing traffic data; 3) And extracting a conversational operation discriminator in the model through a marketing conversational operation, and judging whether the historical marketing telephone traffic data is the marketing conversational operation or not according to the conversational characteristic data.

1) Historical marketing traffic data is determined.

The historical marketing telephone traffic data is the marketing telephone traffic data to be identified whether the marketing telephone traffic data is the marketing telephone traffic.

In one example, the historical marketing traffic data to be identified as marketing traffic may be determined as follows: obtaining marketing dialogue data between a first user and a second user in a historical marketing process, wherein the marketing dialogue data comprises dialogue data of the first user and dialogue data of the second user; and taking the marketing dialogue data of the second user as historical marketing dialogue data to be identified as whether the marketing dialogue is the marketing dialogue or not. By adopting the processing mode, the marketing conversation data of the customer service staff can be extracted from the historical marketing conversation data to be used as the historical marketing conversation data to be identified whether the marketing conversation is a marketing conversation, and the conversation data of consumers irrelevant to the marketing conversation is avoided, so that the conversation mining efficiency can be effectively improved.

In practical applications, the marketing session data between the first user and the second user is typically multi-turn question and answer data. The multi-turn question and answer data can be information of a user session (session). In user session, the data of multiple questions and answers between customer service and user are usually included, including both the conversation data of the customer service and the conversation data of the customer service. And taking the dialogue data of the customer service staff in the multi-round question-answer data as historical marketing traffic data to be identified as whether the marketing traffic data is a marketing traffic. In this case, to extract the marketing dialogue data of the second user from the multiple rounds of question and answer data, the splitting and marking processes need to be performed on the dialogue data corresponding to the user role, and the context can be kept. The context preservation means that the order of each sentence of the first user and the second user in a session is preserved, so that which question of the first user corresponds to the marketing session can be determined.

In particular implementation, the following method can be adopted to extract historical marketing traffic data from the multiple rounds of question and answer texts: executing clause processing on the multi-turn question-answer data to obtain marketing clause data; determining user role information of marketing clause data; and taking the marketing clause data of which the user role information is the second user as historical marketing telephone traffic data to be identified whether the marketing telephone traffic is the marketing telephone traffic.

In specific implementation, the distinction of user roles can see the data condition: when the voice data has distinguished the audio track, the character mark (the first user or the second user) is automatically carried after being converted into the text; when the voice data is not distinguished by the audio track, after the voice data is converted into the text, it cannot be directly distinguished which one is the first user to speak, and which one is the second user to speak, and at this time, a binary model is needed to distinguish. Since the sentence segmentation technique and the role marking technique are both mature prior art, they are not described herein again.

In practical application, marketing personnel mainly develop marketing activities to consumer users in a voice mode, so that the original historical marketing traffic data of customer service marketing personnel is generally the marketing traffic data in a voice form, namely: marketing traffic voice data. In this case, to determine whether the historical marketing traffic data to be identified is marketing traffic, the following steps may be further included: and converting the historical marketing telephone traffic voice data into historical marketing telephone traffic text data through a voice recognition algorithm, wherein the historical marketing telephone traffic text data is used as the historical marketing telephone traffic data to be recognized whether the marketing telephone traffic is the marketing telephone traffic or not. Since the speech recognition algorithm belongs to the mature prior art, it is not described here in detail.

In practice, the original historical marketing traffic data of the customer service marketer also typically includes some noise data, such as web page link addresses, order identifications, communication numbers, emoticons, order numbers, and the like. In this case, to determine whether the historical marketing traffic data to be identified is marketing traffic, the following steps may be further included: and clearing the noise data in the historical marketing traffic data according to the noise text filtering rule. The noise data filtering rules include, but are not limited to: and clearing the webpage link address, the order identification, the communication number, the emoticon, the order number and the like. The order number is usually in the e-commerce system, and privacy data filtering processing is performed on the order number related to the consumer.

In practical applications, the original historical marketing traffic data of the customer service marketer also usually includes some non-standardized conversation contents, such as stop words, multiple synonyms for expressing the same meaning, spoken language expression forms (mainly spoken words, dialect common phrases, etc.), and the like. In this case, to determine whether the historical marketing traffic data to be identified is marketing traffic, the following steps may be further included: and executing standardization processing on the historical marketing traffic data according to the marketing traffic data standardization rule. The marketing traffic data normalization rules include, but are not limited to: filtering stop words, normalizing synonyms, and performing normalization expression on the form of spoken expression (mainly language words, dialect common phrases and the like).

2) And determining the conversational feature data of the historical marketing traffic data through a conversational feature extractor in the marketing conversational extraction model.

In one example, the determining the conversational feature data of the historical marketing traffic data by the conversational feature extractor in the marketing conversational extraction model may be implemented as follows: determining word vectors in historical marketing traffic data through a word embedding layer in the conversational feature extractor; determining the coded data of the historical marketing traffic data according to the word vector through a coding layer in the feature extractor; determining, by a speech feature aggregation layer in the speech feature extractor, speech feature data from the encoded data.

In another example, the determining the linguistic feature data of the historical marketing traffic data by the linguistic feature extractor in the marketing-linguistic extraction model may further include: determining a text segment vector in historical marketing traffic data according to the word vector through a text segment embedding layer in the jargon feature extractor; correspondingly, the coded data is determined according to the text segment vector through the coding layer. By adopting the processing mode, the long-distance semantic dependence of the marketing telephone traffic data context can be determined, so that the problem of keyword ambiguity can be solved, and the accuracy of the telephone traffic mining can be effectively improved.

3) And extracting a conversational operation discriminator in the model through a marketing conversational operation, and judging whether the historical marketing telephone traffic data is the marketing conversational operation or not according to the conversational characteristic data.

In one example, the conversational discriminator may determine whether the marketing traffic data is a marketing conversational gesture according to conversational feature data output by a full connectivity layer.

So far, a basic processing mode of mining the marketing communication from the historical marketing communication text by the constructed marketing communication extraction model is explained. The following describes a preferred embodiment of the mining marketing session in conjunction with practical application.

In one example, if the marketing session extraction condition is satisfied, the marketing traffic data is selected from the historical marketing traffic data as the marketing session through the marketing session extraction model. The marketing session extraction conditions include, but are not limited to: the time length between the current time and the last marketing communication extraction is greater than a time length threshold value; correspondingly, the historical marketing traffic data comprises newly added marketing traffic data after the last marketing traffic extraction. For example, the time duration threshold is 1 day, 7 days, or 1 month, and the historical marketing traffic data includes traffic data newly added within the last 1 day, 7 days, or 1 month. By adopting the processing mode, the marketing communication can be supplemented according to the incremental traffic data regularly; therefore, the richness of the dialects can be effectively improved, the marketing quality is improved, and meanwhile, the computing resources can be saved.

In one example, the marketing session recommendation system may further send, by the server module, the marketing session determined by the marketing session extraction model to the second client module; displaying the marketing talk through the second client module, determining the auditing result of the third user to the marketing talk, and sending the auditing result to the server module; if the result of the audit is yes, the server-side module takes the marketing communication as an effective marketing communication. By adopting the processing mode, after the telephone traffic possibly containing the excellent speech operation is screened and filtered from the daily telephone traffic data of the excellent and beautiful customer service personnel group, the speech operation after the range is reduced is manually compared by the speech operation processing personnel (the third user), and finally the excellent speech operation which is beneficial to improving the marketing service capability is determined, so that the marketing speech quality can be effectively improved.

In practical application, the historical marketing telephone traffic data to be identified as whether the marketing telephone traffic is a marketing telephone traffic is not limited to the marketing telephone traffic data of one marketing scene, and in this case, the marketing scene to which the historical marketing telephone traffic data belongs needs to be determined; and taking the historical marketing traffic data which is judged as the marketing traffic of the marketing scene.

In specific implementation, the marketing scenario to which the user belongs can be determined according to the source of the historical marketing traffic data, and if the historical marketing traffic data is derived from the marketing traffic log data of the 5G package, the marketing scenario to which the user belongs can be directly determined to be the 5G package.

In one example, the marketing tactical extraction model and the scene classification model are obtained by learning from training data of a multi-marketing scene in a multi-task learning mode; therefore, the marketing scenario to which the historical marketing traffic data belongs can be determined through the scenario classification model.

Table 4 shows marketing dialogs for multiple marketing scenarios.

As can be seen from Table 4, the marketing dialog library includes the consumer dialog content of the marketing context. In a specific implementation, the marketing techniques library may not include the content of the consumer conversation of the marketing context, but includes the user session identifier sessionID and the conversation identifier chatID corresponding to the marketing techniques, for example, sessionID "20210324-userA" indicates the user a session at the date of 20210324, and chatID "5" indicates the 5 th session in this session. In this way, the marketing context's consumer conversation content can be found in the original marketing conversation data (log data) by reverse searching. By adopting the processing mode, the storage resources can be effectively saved.

In practical applications, another typical case is that the historical marketing traffic data to be identified as whether the marketing traffic is a marketing traffic is not limited to the marketing traffic data of one marketing field, and in this case, the marketing field to which the historical marketing traffic data belongs needs to be determined; and taking the historical marketing telephone traffic data which is judged to be the marketing telephone traffic as the marketing telephone traffic of the marketing field.

In specific implementation, the marketing field can be determined according to the source of the historical marketing traffic data, and if the historical marketing traffic data is from the marketing traffic log data of a communication operator, the marketing scene can be directly determined to be the field of the communication operator.

In one example, the marketing tactical extraction model and the domain classification model are obtained by learning from training data of a multi-marketing domain in a multi-task learning mode; therefore, the marketing domain to which the historical marketing traffic data belongs can be determined through the domain classification model.

Table 5 shows marketing dialogs for multiple marketing areas, multiple marketing scenarios.

So far, a processing mode of mining marketing dialogues from historical marketing traffic texts through a constructed marketing traffic extraction model is described, and a marketing traffic implementation recommendation mode is described below.

The third stage: and when the marketing campaign is developed, real-time conversational recommendation is carried out based on the mined marketing conversational language.

The system provided by the embodiment can determine a target marketing session for a first user from a plurality of marketing sessions according to the current marketing session data of the first user and a second user, and send the target marketing session to a first client module; the first client module displays the target marketing communication so that the second user can conveniently talk with the first user according to the target marketing communication. When the method is specifically implemented, the first client module can display the recommended dialect in real time in the forms of an embedded window, a popup window and the like.

In one example, targeted marketing campaigns may be determined as follows: 1) Determining the matching degree of the contextual dialogue data associated with the marketing dialogue and the marketing dialogue data; 2) And determining the target marketing words according to the matching degree. The contextual dialog data includes marketing context of the marketing dialog, such as dialog data of the first user during the marketing process.

In specific implementation, the matching degree of the current marketing dialogue data of the first user and the contextual dialogue data (marketing dialogue data of the first user) associated with each candidate marketing dialogue can be determined through a text matching algorithm, and the marketing dialogue with the high matching degree is taken as the target marketing dialogue.

For example, in a currently conducted marketing campaign, a consumer (first user): the current conversation content and the pre-mined marketing communication are unrelated, one online and one television are not wrong, and are smooth, the current broadband is used for handling the marketing context which is associated with the preference (reserved caliber) of the xxx element, the matching degree of the current broadband is higher, so that the target marketing communication is determined to be unrelated, one online and one television are not wrong, and is smooth, the current broadband is used for handling the preference (reserved caliber) of the xxx element, and the current customer service (a second user) can reply to the current consumer according to the target marketing communication.

The contextual dialogue data associated with the marketing dialogues may be stored in a marketing dialogues repository, such as the marketing contexts shown in table 4. In this case, the current dialog content can be directly matched to the marketing context in the marketing dialog library. By adopting the processing mode, the computing resources can be effectively saved. However, this processing method causes redundancy of the context dialog data, and thus consumes a large amount of memory resources, and cannot ensure consistency of redundant data.

In this embodiment, the marketing tactic library may also not include the consumer dialogue content of the marketing context, but rather include the user session identification sessionID and the dialogue identification chatID corresponding to the marketing tactic. In this way, contextual dialog data associated with each candidate dialog can be found in the original marketing dialog data (log data) by a reverse search. By adopting the processing mode, the storage resources can be effectively saved, and the data accuracy can be improved.

In practical application, the marketing dialogue data to be processed by the server module is not limited to marketing dialogue data of one marketing scenario, and in this case, the marketing scenario to which the marketing dialogue belongs needs to be determined; and determining the target marketing session from a plurality of marketing sessions corresponding to the marketing scene to which the marketing session belongs according to the marketing session data.

In a specific implementation, the target marketing session is determined from a plurality of marketing sessions corresponding to the marketing scenario to which the marketing session belongs according to the marketing session data, and the following implementation may be adopted: 1) Telephone traffic retrieval: matching scene identification results of the real-time QA data and existing excellent dialect scene labels in a marketing dialect library, retrieving and recalling excellent marketing dialects in corresponding scenes, and generating a candidate set; 2) Ranking the excellent dialogues in the candidate set according to the current dialog context; 3) The score top-N (N is the number of excellent talks) was taken as the final returned talks result.

According to the system provided by the embodiment of the application, the marketing scene to which the marketing conversation belongs is determined; according to the marketing dialogue data, determining the target marketing dialogue from a plurality of marketing dialogues corresponding to marketing scenes to which the marketing dialogue belongs, so that context scenes of customer service in the telephone traffic can be recognized and sensed in real time, and a dialogue recommendation service can be provided for marketing activities of a plurality of scenes; therefore, the scene service capability can be effectively improved.

In specific implementation, the marketing scene to which the marketing conversation belongs can be determined through a scene classification model constructed based on a multi-task learning mode. By adopting the processing mode, the application efficiency of the cross-scene can be effectively improved.

In practical application, the marketing dialogue data to be processed by the server module is not limited to marketing dialogue data of one marketing field, and in this case, the marketing field to which the marketing dialogue belongs needs to be determined; determining the target marketing session from a plurality of marketing sessions corresponding to a marketing domain to which the marketing session belongs according to the marketing session data.

In specific implementation, the marketing field to which the marketing conversation belongs can be determined through a field classification model constructed based on a multi-task learning mode. By adopting the processing mode, the cross-domain application efficiency can be effectively improved.

As can be seen from the above embodiments, the marketing telephone operation recommendation system provided in the embodiments of the present application learns the marketing telephone operation extraction model from the marketing telephone operation data and the marking data indicating whether the marketing telephone operation data is the marketing telephone operation or not through the server module; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; and determining a target marketing session for the first user from a plurality of marketing sessions for the received marketing session data of the first user and the second user, and sending the target marketing session to the first client module; and displaying the target marketing communication through the first client module so that the second user can conveniently talk with the first user according to the target marketing communication. By adopting the processing mode, the telephone traffic containing the excellent speech technology is screened and filtered out by mining the daily telephone traffic data of excellent customer service staff groups, and in the conversation process of the customer service marketer (the second user) for developing the marketing activity to the consumer (the first user), the higher-quality and richer speech technology is recommended to the customer service marketer in real time based on the mined excellent speech technology, so that the customer service marketer is assisted to develop the marketing activity; therefore, the marketing quality and efficiency can be effectively improved, and the marketing success rate is improved. In addition, the language recommendation processing process adopted by the system is high in reusability, only data in different fields (such as the operator field, the e-commerce field, the education institution field, the insurance field and the like) need to be replaced, and the structure of the marketing language extraction model can be directly reused, so that the expandability of the marketing field can be effectively improved.

Second embodiment

In the above embodiment, a marketing conversation recommendation system is provided, and correspondingly, the present application further provides a marketing conversation processing method, which can mine the daily traffic data of the excellent customer service staff group, and filter the traffic containing the excellent conversation from the excellent customer service staff group.

In order to explain the technical effect of the marketing word processing method provided in the embodiment of the present application, first, a word mining technique in the prior art and technical problems thereof are briefly described below.

In the marketing field, front-line customer service personnel usually reply to consumers according to own experiences, and the customers are difficult to give better answers due to different levels of customer service personnel. In order to improve marketing effect, more and more enterprises start to construct a marketing conversation base, and customer service personnel can develop marketing activities to consumers based on the conversation base.

The phone library may include Standard phones in an artificially defined Standard Operating Procedure (SOP) and may also include excellent phones determined based on phone mining techniques. Currently, commonly used excavation schemes include: a conversational mining scheme based on heuristic search, a generative conversational mining scheme that generates excellent conversational language from conversational scripts, and a scheme based on keywords and scene templates. The following briefly describes the implementation of the above three schemes and their advantages and disadvantages.

Scheme one, a conversational mining scheme based on heuristic search.

The search space of the scheme is full traffic data, and heuristic search is performed by utilizing manually labeled seed data (manually labeled excellent dialect), and the core of the heuristic search is gradual iterative mining by defining the measurement between the seed data and the target traffic data. However, it suffers from several significant disadvantages: 1) The method is very sensitive to a threshold value of similarity measurement, and is difficult to adjust in actual landing, so that the result is unstable; 2) The upper limit of the accuracy rate is lower; 3) The scalability is not sufficient due to the influence of the data set and the similarity measurement mode.

And a generating-type grammar mining scheme for generating excellent vocabularies according to the vocabularies original texts.

The scheme adopts a speech-to-speech (seq-to-seq) method, generates excellent speech according to a speech original text, and is characterized in that the generated speech does not necessarily come from the original text. However, there are also disadvantages as follows: 1) The generated result is uncontrollable, and the quality is difficult to ensure; 2) Readability is insufficient.

Scheme two, scheme based on keyword and scene template.

The scheme adopts a manually-combed keyword library and manually realizes the excavation of dialogues according to different templates defined by specific scenes. The defects of the method comprise: 1) The method depends on word banks seriously and is not generalized enough; 2) Different data and scenes are required to be reset, so that the efficiency is low; 3) The keyword matching mode cannot solve the semantic ambiguity problem of the context.

In summary, the existing speech excavation technology has the problems of low speech excavation accuracy and the like.

Aiming at the technical problems in the prior art, the application provides a marketing telephone operation processing method which can be used for mining daily telephone operation data of excellent customer service staff groups, and screening and filtering telephone operations containing excellent telephone operations. The method corresponds to the second stage of the process in the embodiment of the system described above. Since the method embodiment is substantially similar to the system embodiment, the description is relatively simple, and reference may be made to the partial description of the system embodiment for relevant points. The method embodiments described below are merely illustrative.

Please refer to fig. 4, which is a flowchart illustrating a marketing communication processing method according to the present application. In this embodiment, the marketing communication processing method includes the following steps:

step S401: historical marketing traffic data is determined.

In one example, step S401 may include the following sub-steps: acquiring marketing dialogue data between a first user and a second user in a historical marketing process; and taking the historical marketing conversation data of the second user as historical marketing traffic data.

In one example, the historical marketing traffic data includes: marketing traffic voice data; the method may further comprise the steps of: and converting the marketing telephone traffic voice data into a marketing telephone traffic text through a voice recognition algorithm.

In one example, the method may further comprise the steps of: the noise data in the historical marketing traffic data is cleared according to the noise data filtering rules.

In one example, the method may further include the steps of: and executing standardization processing on the historical marketing traffic data according to the marketing traffic data standardization rule.

Step S403: and determining the conversational feature data of the historical marketing traffic data through a conversational feature extractor in the marketing conversational extraction model.

Step S405: and extracting a word technology discriminator in the model through the marketing word technology, and judging whether the historical marketing word traffic data is the marketing word technology or not according to the characteristic data.

In one example, the method may comprise the steps of: determining a marketing scene to which the historical marketing traffic data belongs; and taking the historical marketing traffic data which is judged as the marketing traffic of the marketing scene.

In one example, the method may comprise the steps of: learning from training data of a plurality of marketing scenes in a multi-task learning mode to obtain the marketing word operation extraction model and the scene classification model; the scene classification model comprises: the scene feature extractor and the scene classifier share a coding layer; the training data includes: marketing telephone traffic data, marking data of whether the marketing telephone traffic is the marketing telephone traffic or not and marking data of a marketing scene; correspondingly, the marketing scene to which the historical marketing traffic data belongs is determined through the scene classification model.

In one example, the method may comprise the steps of: learning from training data of a plurality of marketing fields in a multi-task learning mode to obtain the marketing tactical extraction model and the field classification model; the domain classification model comprises: the system comprises a domain feature extractor and a domain classifier, wherein the domain feature extractor and the conversational feature extractor share a coding layer; the training data includes: marketing telephone traffic data, marking data of whether the marketing telephone traffic is the marketing telephone traffic or not and marking data of marketing fields; correspondingly, the marketing field to which the historical marketing traffic data belongs is determined through the field classification model.

As can be seen from the foregoing embodiments, the marketing communication processing method provided in the embodiments of the present application determines historical marketing communication data; extracting a feature extractor in the model through marketing telephone operation, and determining feature data of historical marketing telephone traffic data; and extracting a word technology discriminator in the model through the marketing word technology, and judging whether the historical marketing word traffic data is the marketing word technology or not according to the characteristic data. By adopting the processing mode, at least the following technical effects can be achieved:

4) The method is a removable type word operation mining method, existing expressions can be extracted from the original text of the marketing word operation and are displayed in a standardized mode, and the situation that the word operation mining result is uncontrollable does not exist, so that the word operation quality and readability can be effectively improved;

7) If the training data of the marketing session extraction model comprises a multi-scene marketing traffic text, and the marketing session extraction model and the scene classification model are simultaneously learned in a multi-task learning mode, the two models share the encoder, so that the cross-scene application efficiency of marketing session mining can be further improved.

9) The reusability of the dialect mining process is strong, only data in different fields (such as an operator field and an e-commerce field) need to be replaced, and the model structure can be directly reused, so that the expandability of the dialect mining field can be effectively improved.

Third embodiment

In the above embodiment, a marketing communication processing method is provided, and correspondingly, the application further provides a marketing communication processing device. The apparatus corresponds to an embodiment of the method described above. Since the apparatus embodiments are substantially similar to the method embodiments, they are described in a relatively simple manner, and reference may be made to some of the descriptions of the method embodiments for relevant points. The device embodiments described below are merely illustrative.

The present application additionally provides a marketing communication processing apparatus, comprising:

Fourth embodiment

In the above embodiment, a marketing tactical recommendation system is provided, and correspondingly, the application further provides a marketing tactical extraction model processing method, which can be used for constructing a marketing tactical extraction model. The method corresponds to the first stage of the process in the embodiment of the system described above. Since the method embodiment is basically similar to the system embodiment, the description is simple, and the relevant points can be referred to the partial description of the system embodiment. The method embodiments described below are merely illustrative.

Please refer to fig. 5, which is a flowchart illustrating a marketing strategy extraction model processing method according to the present application. In this embodiment, the marketing tactical extraction model processing method includes the following steps:

step S501: a training data set is determined.

The training data includes: marketing traffic data, and marking data of whether the marketing traffic data is marketing traffic.

Step S503: and constructing a network structure of the marketing tactics extraction model.

In one example, the network architecture includes: a phonetics feature extractor and a phonetics discriminator; the conversational feature extractor is used for determining conversational feature data of the marketing traffic data; and the conversational technology discriminator is used for judging whether the marketing telephone traffic data is the marketing conversational technology or not according to the conversational characteristic data.

In one example, the tactical feature extractor comprises: a word embedding layer, a text segment embedding layer, a coding layer and a language feature aggregation layer; the word embedding layer is used for determining word vectors in the marketing traffic data; the text segment embedding layer is used for determining text segment vectors in the marketing telephone traffic data according to the word vectors; the coding layer is used for determining the coding data of the marketing telephone traffic data according to the text segment vector; the speech characteristic aggregation layer is used for determining the speech characteristic data according to the coded data.

Step S505: and learning from the training data set to obtain the network parameters of the marketing strategy extraction model.

In one example, the training data further comprises: marketing scene annotation data; the network architecture further comprises: a scene feature extractor and a scene classifier; the scene feature extractor is used for determining scene feature data of the marketing telephone traffic data; the scene classifier is used for determining a marketing scene to which the marketing telephone traffic data belongs according to the scene characteristic data; the scene feature extractor includes: the word embedding layer, the text segment embedding layer, the coding layer and the scene feature aggregation layer; the scene feature aggregation layer is used for determining the scene feature data according to the coded data; learning from training data of a plurality of marketing scenes in a multi-task learning mode to obtain the marketing tactical prediction model and the scene classification model; the scene classification model includes the scene feature extractor and the scene classifier.

In one example, the training data further comprises: marketing domain annotation data; the network architecture further comprises: a domain feature extractor and a domain classifier; the domain feature extractor is used for determining domain feature data of the marketing telephone traffic data; the domain classifier is used for determining the marketing domain to which the marketing telephone traffic data belongs according to the domain feature data; the domain feature extractor includes: the word embedding layer, the text segment embedding layer, the coding layer and the domain feature aggregation layer; the domain feature aggregation layer is used for determining the domain feature data according to the coded data; learning from training data in a multi-marketing field in a multi-task learning mode to obtain the marketing tactical prediction model and the field classification model; the domain classification model comprises the domain feature extractor and the domain classifier. The multi-marketing domain includes, but is not limited to: the communication operator field, the E-business field, the education and training field and the insurance field.

As can be seen from the above embodiments, the marketing communication extraction model processing method provided by the embodiment of the present application determines a training data set; constructing a network structure of a marketing tactic extraction model; and learning from the training data set to obtain the network parameters of the marketing strategy extraction model. By adopting the processing mode, a marketing speech extraction model is established by using the prior knowledge, the model adopts a neural network to automatically learn and fit certain specific templates/styles, such as characteristic data of excellent speech, in a marketing speech text, and the model can be used for mining daily speech data of excellent customer service personnel groups, and screening and filtering out the speech with excellent speech; therefore, the accuracy, richness, stability and generalization of the dialect mining can be effectively improved, and the dialect quality and readability can be improved.

Fifth embodiment

In the above embodiment, a marketing word extraction model processing method is provided, and correspondingly, the application further provides a marketing word extraction model processing device. The apparatus corresponds to an embodiment of the method described above. Since the apparatus embodiments are substantially similar to the method embodiments, they are described in a relatively simple manner, and reference may be made to some of the descriptions of the method embodiments for relevant points. The device embodiments described below are merely illustrative.

The present application additionally provides a marketing communication extraction model processing apparatus, comprising:

a training data determination unit for determining a training data set; the training data includes: marketing telephone traffic data, and marking data of whether the marketing telephone traffic data is marketing telephone traffic;

Sixth embodiment

In the above embodiment, a marketing tactical recommendation system is provided, and correspondingly, the application also provides a marketing tactical recommendation method, which can be used for constructing a marketing tactical extraction model. The method corresponds to the embodiment of the system described above. Since the method embodiment is basically similar to the system embodiment, the description is simple, and the relevant points can be referred to the partial description of the system embodiment. The method embodiments described below are merely illustrative.

Please refer to fig. 6, which is a flowchart illustrating a marketing strategy recommendation method according to the present application. In this embodiment, the marketing strategy recommendation method includes the following steps:

step S601: and learning the marketing communication operation data and the marking data which is whether the marketing communication operation data is the marketing communication operation or not to obtain the marketing communication operation extraction model.

Step S603: and selecting marketing telephone traffic data from the historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model.

Step S605: according to the marketing dialogue data of the first user and the second user, a target marketing dialogue aiming at the first user is determined from a plurality of marketing dialogues, so that the second user can conveniently dialogue with the first user according to the target marketing dialogue.

In one example, the method may further comprise the steps of: determining the auditing result of the marketing dialect determined by the marketing dialect extraction model by the third user; if the result of the audit is yes, the marketing session is used as an effective marketing session.

In one example, the method may further comprise the steps of: determining a marketing scene to which a marketing conversation belongs; determining a marketing scene to which the marketing conversation belongs; step S605 can be implemented as follows: and determining the target marketing session from a plurality of marketing sessions corresponding to the marketing scene to which the marketing session belongs according to the marketing session data.

In one example, step S601 can be implemented as follows: learning from training data of a multi-marketing scene in a multi-task learning mode to obtain the marketing tactical prediction model and a scene classification model; the scene classification model and the marketing tactical prediction model share a coding layer; the training data includes: marketing telephone traffic data, marking data of whether the marketing telephone traffic is the marketing telephone traffic or not and marking data of a marketing scene; correspondingly, the marketing scene to which the marketing conversation belongs and the marketing scene to which the marketing conversation belongs are determined through the scene classification model.

In one example, if the marketing dialogs extraction condition is satisfied, the marketing traffic data is selected from the historical marketing traffic data through a marketing dialogs extraction model as the marketing dialogs. For example, the marketing dialog extraction conditions include: the time length between the current time and the last marketing communication extraction is greater than a time length threshold value; correspondingly, the historical marketing telephone traffic data comprises newly added marketing telephone traffic data after the last time of marketing telephone traffic extraction.

In one example, step S605 may include the following sub-steps: determining the matching degree of the contextual dialogue data associated with the marketing dialogue and the marketing dialogue data; and determining the target marketing words according to the matching degree.

As can be seen from the above embodiments, the marketing telephone operation recommendation method provided in the embodiments of the present application learns the marketing telephone operation extraction model from the plurality of marketing telephone operation data and the label data indicating whether the marketing telephone operation data is the marketing telephone operation; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; according to the marketing dialogue data of the first user and the second user, a target marketing dialogue aiming at the first user is determined from a plurality of marketing dialogues, so that the second user can conveniently dialogue with the first user according to the target marketing dialogue. By adopting the processing mode, the telephone traffic containing the excellent speech technology is screened and filtered out by mining the daily telephone traffic data of excellent customer service staff groups, and in the conversation process of the customer service marketer (the second user) for developing the marketing activity to the consumer (the first user), the higher-quality and richer speech technology is recommended to the customer service marketer in real time based on the mined excellent speech technology, so that the customer service marketer is assisted to develop the marketing activity; therefore, the marketing quality and efficiency can be effectively improved, and the marketing success rate is improved. In addition, the language recommendation processing process adopted by the system is high in reusability, only data in different fields (such as the operator field, the e-commerce field, the education institution field, the insurance field and the like) need to be replaced, and the structure of the marketing language extraction model can be directly reused, so that the expandability of the marketing field can be effectively improved.

Seventh embodiment

In the above embodiment, a marketing tactical recommendation method is provided, and correspondingly, the application further provides a marketing tactical recommendation device. The apparatus corresponds to an embodiment of the method described above. Since the apparatus embodiments are substantially similar to the method embodiments, they are described in a relatively simple manner, and reference may be made to some of the descriptions of the method embodiments for relevant points. The device embodiments described below are merely illustrative.

The present application additionally provides a marketing conversation recommendation device, comprising:

Eighth embodiment

In the above embodiment, a marketing jargon recommendation system is provided, and correspondingly, the application further provides a marketing jargon processing method, which can perform manual review processing on marketing jargon mined by a marketing jargon extraction model. The method corresponds to the embodiment of the system described above. Since the method embodiment is basically similar to the system embodiment, the description is simple, and the relevant points can be referred to the partial description of the system embodiment. The method embodiments described below are merely illustrative.

Please refer to fig. 7, which is a flowchart illustrating a marketing communication processing method according to the present application. In this embodiment, the marketing communication processing method includes the following steps:

step S701: and receiving the marketing telephone operation determined from the historical marketing telephone operation data by the marketing telephone operation extraction model sent by the server side module.

The marketing communication extraction model is obtained by learning from a plurality of marketing communication data and the marking data of whether the marketing communication data is the marketing communication.

Step S703: determining the auditing result of the marketing session of a third user;

step S705: and sending the auditing result to the server module so that the server module determines whether the marketing session is effective or not according to the auditing result.

As can be seen from the foregoing embodiments, the marketing communication processing method provided in the embodiments of the present application receives the marketing communication determined from the historical marketing communication data by the marketing communication extraction model sent by the server; the marketing telephone operation extraction model is obtained by learning from a plurality of marketing telephone operation data and the marking data of whether the marketing telephone operation data is the marketing telephone operation or not; determining an auditing result of a third user on the marketing session; and sending the auditing result to the server so that the server determines whether the marketing session is effective or not according to the auditing result. By adopting the processing mode, after the telephone traffic possibly containing the excellent speech operation is screened and filtered from the daily telephone traffic data of the excellent and beautiful customer service personnel group, the speech operation processing personnel manually compares the speech operation with the reduced range, and finally determines the excellent speech operation which is beneficial to improving the marketing service capability.

Ninth embodiment

In the above embodiment, a marketing dialect processing method is provided, and correspondingly, a marketing dialect processing device is also provided. The device corresponds to the embodiment of the method. Since the apparatus embodiments are substantially similar to the method embodiments, they are described in a relatively simple manner, and reference may be made to some of the descriptions of the method embodiments for relevant points. The device embodiments described below are merely illustrative.

The present application additionally provides a marketing tactics processing apparatus, comprising:

the system comprises a speech receiving unit, a marketing speech extracting unit and a speech processing unit, wherein the speech receiving unit is used for receiving the marketing speech determined by a marketing speech extracting model from historical marketing speech data sent by a server module; the marketing communication extraction model is obtained by learning from a plurality of marketing communication data and the marking data of whether the marketing communication data is the marketing communication;

and the auditing result sending unit is used for sending the auditing result to the server module so as to ensure that the server module determines whether the marketing session is effective according to the auditing result.

Tenth embodiment

In the above embodiments, a marketing tactical recommendation system is provided, and correspondingly, the application also provides a marketing tactical recommendation method, where an execution subject of the method includes, but is not limited to, a terminal device of a second user. The method corresponds to the embodiment of the system described above. Since the method embodiment is basically similar to the system embodiment, the description is simple, and the relevant points can be referred to the partial description of the system embodiment. The method embodiments described below are merely illustrative.

Please refer to fig. 8, which is a flowchart illustrating a marketing tactical recommendation method according to the present application. In this embodiment, the marketing tactical recommendation method includes the following steps:

step S801: and receiving the target marketing words sent by the server module and aiming at the first user.

The server module determines the target marketing communication in the following way: learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; and determining a targeted marketing session from the plurality of marketing sessions for the received marketing session data of the first user and the second user.

Step S803: and displaying the target marketing communication so that the second user can conveniently talk with the first user according to the target marketing communication.

As can be seen from the foregoing embodiments, the marketing communication recommendation method provided in the embodiments of the present application receives a target marketing communication sent by a server and addressed to a first user; the server side determines the target marketing communication in the following mode: learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic; and determining a targeted marketing session from a plurality of marketing sessions for the received marketing session data of the first user and the second user; and displaying the target marketing communication so that the second user can conveniently talk with the first user according to the target marketing communication. By adopting the processing mode, the telephone traffic containing the excellent speech technology is screened and filtered through mining the daily telephone traffic data of excellent customer service staff groups, and higher-quality and richer marketing speech technologies are recommended to the customer service marketing staff in real time based on the mined excellent speech technology in the process of carrying out marketing activities on the customers (first users) by the customer service marketing staff (second users), so that the customer service marketing staff is assisted to carry out marketing activities; therefore, the marketing quality and efficiency can be effectively improved, and the marketing success rate is improved. In addition, the language recommendation processing process adopted by the system is high in reusability, only data in different fields (such as the operator field, the e-commerce field, the education institution field, the insurance field and the like) need to be replaced, and the structure of the marketing language extraction model can be directly reused, so that the expandability of the marketing field can be effectively improved.

Eleventh embodiment

the system comprises a recommended word receiving unit, a target marketing word sending unit and a target word receiving unit, wherein the recommended word receiving unit is used for receiving the target marketing word aiming at a first user and sent by a server module; the server module determines the target marketing session in the following manner: learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; and determining a targeted marketing session from a plurality of marketing sessions for the received marketing session data of the first user and the second user;

Twelfth embodiment

In the above embodiment, a marketing tactics recommendation system is provided, and correspondingly, the application further provides a marketing tactics recommendation system. The system corresponds to the embodiments of the system described above. The similarity between the two system embodiments can be found in the partial description of the first embodiment. The system embodiments described below are merely illustrative.

A marketing tactics recommendation system of this embodiment, includes: the system comprises at least one first server module, a second server module and a client module.

Please refer to fig. 9, which is a schematic diagram of another application scenario of the marketing tactical recommendation system of the present application. In this embodiment, at least one first server module may be deployed at first server sides of different marketing enterprises, a second server module is deployed at a server of a general marketing tactic recommendation platform that provides marketing tactic recommendation services to a plurality of marketing enterprises, and a client is deployed at a terminal device side of a marketing person and may display marketing tactics recommended by the second server, so that cross-domain and cross-scene real-time tactic recommendation may be implemented.

For example, a traffic system of a certain communication carrier is deployed on a first service end a side, and the traffic system includes a first service end module a ', and the module a' sends historical marketing traffic data of the communication carrier (communication carrier field) to a second service end module, and sends marketing conversation data of a first user and a second user of the communication carrier; the telephone traffic system of an education and training institution is deployed on the side of a first service end B, and comprises a first service end module B ', and the module B' sends historical marketing telephone traffic data of the education and training institution (education and training field) to a second service end module and sends marketing conversation data of a first user and a second user of the education and training institution; the telephone traffic system of a certain insurance company is deployed at the side of a first service end C, and comprises a first service end module C ', and the module C' sends the historical marketing telephone traffic data of the insurance company (insurance field) to a second service end module and sends the marketing dialogue data of a first user and a second user of the insurance company.

In this embodiment, the first service module is configured to send historical marketing traffic data of a target field; and sending marketing dialog data of the first user and the second user; the second server module is used for learning marketing telephone traffic data of the multi-marketing field and the marking data of whether the marketing telephone traffic data is marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multi-marketing field; selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic of a target field; and determining a target marketing conversation for the first user from a plurality of marketing conversations in a target field according to the marketing conversation data, and transmitting the target marketing conversation; the client module is used for displaying the target marketing communication so that the second user can conveniently talk with the first user according to the target marketing communication. The target fields corresponding to different first service end modules include but are not limited to: the communication operator field, the e-commerce field, the educational training field, the insurance field, and the like.

In specific implementation, one general marketing telephone traffic extraction model applicable to multiple marketing fields can be learned from marketing telephone traffic data of multiple marketing fields and label data of whether the marketing telephone traffic data are marketing telephone traffic, and a plurality of general marketing telephone traffic extraction models applicable to different marketing fields can be learned from marketing telephone traffic data of different marketing fields and label data of whether the marketing telephone traffic data are marketing telephone traffic.

In an example, the first server module may send the historical marketing traffic data according to a preset time interval, where the historical marketing traffic data includes newly added marketing traffic data in a target field in the time interval, for example, the newly added traffic data is sent once a day, and the newly added traffic data is mined once a day. By adopting the processing mode, the marketing communication can be supplemented according to the incremental traffic data regularly; therefore, the language and skill richness can be effectively improved, and the marketing quality is improved.

As can be seen from the foregoing embodiments, the marketing telephone traffic recommendation system provided in the embodiments of the present application sends historical marketing telephone traffic data of a target field through at least one first service-side module; learning from the marketing telephone traffic data of the multiple marketing fields and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not through a second server module to obtain a marketing telephone traffic extraction model of the multiple marketing fields; selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic of a target field; sending marketing dialogue data of the first user and the second user through at least one first service end module; determining a target marketing conversation for the first user from a plurality of marketing conversations in a target field according to the marketing conversation data through a second server module, and sending the target marketing conversation; and displaying the target marketing communication through the client module so that a second user can conveniently talk with the first user according to the target marketing communication. By adopting the processing mode, real-time speech technology recommendation service is provided for third-party marketing activities in multiple fields through the universal speech technology recommendation platform, cross-field real-time speech technology recommendation is realized, wherein the speech technology recommendation platform is used for mining daily traffic data of excellent customer service personnel groups of the third-party marketing activities in multiple fields, screening and filtering out the traffic containing excellent speech technology to form a marketing speech technology library in multiple marketing fields, and in the process of carrying out marketing activities on customers by marketing personnel, higher-quality and richer marketing speech technologies are recommended to the customer service marketing personnel in real time based on the mined excellent speech technology to assist the customer service personnel to carry out marketing activities; therefore, the field service capability can be effectively improved, the marketing quality and efficiency can be improved, and the marketing success rate can be improved.

Thirteenth embodiment

In the above embodiment, a marketing tactical recommendation system is provided, and correspondingly, the application further provides a marketing tactical recommendation method. The system corresponds to the embodiment of the method described above. The method corresponds to the processing procedure of the first service module in the embodiment of the system. Since the method embodiment is substantially similar to the system embodiment, the description is relatively simple, and reference may be made to the partial description of the system embodiment for relevant points. The system embodiments described below are merely illustrative.

The application also provides a marketing strategy recommendation method, which comprises the following steps:

1. sending the historical marketing telephone traffic data of the target field to the server module, so that the server module extracts the model through the marketing telephone traffic, and selects the marketing telephone traffic data from the historical marketing telephone traffic data as the marketing telephone traffic of the target field; the server module learns the marketing telephone traffic data of the multiple marketing fields and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multiple marketing fields;

2. and sending the marketing dialogue data of the first user and the second user to the server module, so that the server module determines a target marketing dialogue aiming at the first user from a plurality of marketing dialogues in a target field according to the marketing dialogue data, and the second user can conveniently talk with the first user according to the target marketing dialogue.

Fourteenth embodiment

In the above embodiment, a marketing tactical recommendation method is provided, and correspondingly, a marketing tactical recommendation device is also provided in the present application. The device corresponds to the embodiment of the method. Since the apparatus embodiments are substantially similar to the method embodiments, they are described in a relatively simple manner, and reference may be made to some of the descriptions of the method embodiments for relevant points. The device embodiments described below are merely illustrative.

Fifteenth embodiment

In the above embodiment, a marketing tactical recommendation system is provided, and correspondingly, the application further provides a marketing tactical recommendation method. The system corresponds to the embodiment of the method described above. The method corresponds to the processing procedure of the second server module in the embodiment of the system. Since the method embodiment is substantially similar to the system embodiment, the description is relatively simple, and reference may be made to the partial description of the system embodiment for relevant points. The system embodiments described below are merely illustrative.

1. learning from marketing telephone traffic data of multiple marketing fields and marking data of whether the marketing telephone traffic data is marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multiple marketing fields;

2. aiming at the received historical marketing telephone traffic data of the target field, selecting the marketing telephone traffic data from the historical marketing telephone traffic data as the marketing telephone traffic of the target field through a marketing telephone traffic extraction model;

3. and aiming at the received marketing dialogue data of the first user and the second user, determining a target marketing dialogue aiming at the first user from a plurality of marketing dialogues in a target field so as to facilitate the second user to dialogue with the first user according to the target marketing dialogue.

Sixteenth embodiment

In the above embodiment, a marketing tactical recommendation method is provided, and correspondingly, the application further provides a marketing tactical recommendation device. The device corresponds to the embodiment of the method. Since the apparatus embodiments are substantially similar to the method embodiments, they are described in a relatively simple manner, and reference may be made to some of the descriptions of the method embodiments for relevant points. The device embodiments described below are merely illustrative.

The present application additionally provides a marketing tactics recommendation device, comprising:

the model construction unit is used for learning from marketing telephone traffic data of the multi-marketing field and marking data of whether the marketing telephone traffic data is marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multi-marketing field;

Seventeenth embodiment

In the above embodiment, a marketing tactical recommendation system is provided, and correspondingly, the application further provides a marketing tactical recommendation method. The system corresponds to the embodiment of the method described above. The method corresponds to the processing procedure of the client module in the embodiment of the system. Since the method embodiment is substantially similar to the system embodiment, the description is relatively simple, and reference may be made to the partial description of the system embodiment for relevant points. The system embodiments described below are merely illustrative.

1. receiving a target marketing conversation aiming at a first user and sent by a server module; the server module determines a target marketing session in the following way: learning from marketing telephone traffic data of multiple marketing fields and marking data of whether the marketing telephone traffic data is marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multiple marketing fields; aiming at the received historical marketing telephone traffic data of the target field, selecting the marketing telephone traffic data from the historical marketing telephone traffic data as the marketing telephone traffic of the target field through a marketing telephone traffic extraction model; determining a target marketing conversation for the first user from a plurality of marketing conversations in a target domain for the received marketing conversation data of the first user and the second user;

2. and displaying the target marketing communication so that the second user can conveniently talk with the first user according to the target marketing communication.

Eighteenth embodiment

In the above embodiment, a marketing tactical recommendation method is provided, and correspondingly, a marketing tactical recommendation device is also provided in the present application. The apparatus corresponds to an embodiment of the method described above. Since the apparatus embodiments are substantially similar to the method embodiments, they are described in a relatively simple manner, and reference may be made to some of the description of the method embodiments for relevant points. The device embodiments described below are merely illustrative.

the system comprises a recommended word receiving unit, a target marketing word sending unit and a target word receiving unit, wherein the recommended word receiving unit is used for receiving the target marketing word aiming at a first user and sent by a server module; the server module determines a target marketing session in the following way: learning from the marketing telephone traffic data of the multi-marketing field and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model of the multi-marketing field; aiming at the received historical marketing telephone traffic data of the target field, selecting the marketing telephone traffic data from the historical marketing telephone traffic data as the marketing telephone traffic of the target field through a marketing telephone traffic extraction model; determining a target marketing conversation for the first user from a plurality of marketing conversations in a target domain for the received marketing conversation data of the first user and the second user;

and the recommended speech presentation unit is used for presenting the target marketing speech so as to facilitate the second user to have a conversation with the first user according to the target marketing speech.

Nineteenth embodiment

In the embodiment, a marketing conversation recommendation system is provided, and correspondingly, a customer service robot system is further provided. The system corresponds to the embodiments of the system described above. The similarity between the two system embodiments can be found in the partial description of the first embodiment. The system embodiments described below are merely illustrative.

A customer service robot system of this embodiment includes: a server module and a client module.

Please refer to fig. 10, which is a schematic view of an application scenario of the customer service robot system according to the present application. In this embodiment, the server module may be deployed at a server side of the customer service robot platform, and the client module is deployed at a terminal device (such as a smart speaker, a smart phone, a personal computer, etc.) side of a consumer user, and may display information replied by the customer service robot based on a marketing session, thereby implementing development of a marketing campaign by the customer service robot.

In this embodiment, the server module is configured to learn a marketing telephone operation extraction model from a plurality of marketing telephone operation data and label data of whether the marketing telephone operation data is a marketing telephone operation; selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic; and determining a target marketing conversation for the target user from the plurality of marketing conversations for the received marketing conversation data of the target user, and sending the target marketing conversation to the client module; the client module is used for displaying the target marketing words.

As can be seen from the above embodiments, the customer service robot system provided in the embodiment of the application learns the marketing telephone traffic extraction model from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic is the marketing telephone traffic through the server module; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; and determining a target marketing conversation for the target user from the plurality of marketing conversations for the received marketing conversation data of the target user, and sending the target marketing conversation to the client module; and displaying the target marketing words through the client module. By adopting the processing mode, the telephone traffic containing excellent speech is screened and filtered through mining the daily telephone traffic data of excellent customer service personnel groups, and in the conversation process of the customer service robot for developing marketing activities to consumers (target users), the marketing speech with higher quality and richer content is determined in real time based on the mined excellent speech, and the marketing activities are automatically developed through the customer service robot; therefore, the marketing quality and efficiency can be effectively improved, and the marketing success rate is improved.

Twentieth embodiment

In the embodiment, a customer service robot system is provided, and correspondingly, the application also provides a robot dialogue method. The system corresponds to the embodiment of the method described above. The method corresponds to the processing procedure of the server module in the embodiment of the system. Since the method embodiment is basically similar to the system embodiment, the description is simple, and the relevant points can be referred to the partial description of the system embodiment. The system embodiments described below are merely illustrative.

The present application also provides a robot dialogue method, which may include the steps of:

1. learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic to obtain a marketing telephone traffic extraction model;

2. selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic;

3. and determining a target marketing conversation aiming at the target user from the plurality of marketing conversations according to the marketing conversation data of the target user.

Twenty-first embodiment

In the above embodiment, a robot dialogue method is provided, and correspondingly, the present application further provides a robot dialogue device. The apparatus corresponds to an embodiment of the method described above. Since the apparatus embodiments are substantially similar to the method embodiments, they are described in a relatively simple manner, and reference may be made to some of the description of the method embodiments for relevant points. The device embodiments described below are merely illustrative.

The present application further provides a robot dialogue apparatus including:

the model building unit is used for learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model;

Twenty-second embodiment

In the embodiment, a customer service robot system is provided, and correspondingly, the application also provides a robot dialogue method. The system corresponds to the embodiment of the method described above. The method corresponds to the processing procedure of the client module in the embodiment of the system. Since the method embodiment is substantially similar to the system embodiment, the description is relatively simple, and reference may be made to the partial description of the system embodiment for relevant points. The system embodiments described below are merely illustrative.

The application also provides a robot dialogue method, which can comprise the following steps:

1. receiving a target marketing conversation aiming at a target user and sent by a server module; the server module determines the target marketing communication in the following way: learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; determining a target marketing conversation aiming at the target user from a plurality of marketing conversations according to the marketing conversation data of the target user;

2. and displaying the target marketing words.

Twenty-third embodiment

The present application further provides a robot dialogue apparatus including:

the recommendation conversation receiving unit is used for receiving the target marketing conversation aiming at the target user and sent by the server module; the server module determines the target marketing communication in the following way: learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic; determining a target marketing conversation aiming at the target user from a plurality of marketing conversations according to the marketing conversation data of the target user;

and the recommended word presentation unit is used for presenting the target marketing words.

Twenty fourth embodiment

In the above embodiment, a marketing tactical recommendation system implementation is provided, and correspondingly, the application further provides a marketing tactical recommendation method. The method corresponds to the embodiment implemented by the system. Since the method embodiment is basically similar to the system embodiment, the description is simple, and the relevant points can be referred to the partial description of the system embodiment. The system embodiments described below are merely illustrative.

1. constructing a marketing session library according to historical marketing session data;

2. acquiring current marketing dialogue data;

3. and determining a target marketing conversation from the marketing conversation library according to the current marketing conversation data.

In this embodiment, the step of determining a target marketing conversation from a marketing conversation library according to the current marketing conversation data may include the following sub-steps: determining the matching degree of the historical contextual dialog data associated with the marketing dialog and the current marketing dialog data; and determining the target marketing skills according to the matching degree.

As can be seen from the above embodiments, the marketing session recommendation method provided in the embodiments of the present application constructs a marketing session library according to historical marketing session data; acquiring current marketing dialogue data; and determining a target marketing conversation from the marketing conversation library according to the current marketing conversation data. By adopting the processing mode, the telephone traffic containing the excellent speech technology is screened and filtered by mining the daily telephone traffic data of excellent customer service personnel groups, and the marketing speech technology is recommended in real time based on the mined excellent speech technology when marketing activities are carried out on consumers, so that a customer service robot system capable of carrying out marketing activities fully automatically can be realized, and a marketing speech recommendation system capable of assisting customer service marketing personnel to carry out marketing activities can also be realized; therefore, the marketing quality and efficiency can be effectively improved, and the marketing success rate is improved.

Twenty-fifth embodiment

the conversation library construction unit is used for constructing a marketing conversation library according to historical marketing conversation data;

the marketing dialogue acquisition unit is used for acquiring current marketing dialogue data;

and the word operation recommending unit is used for determining the target marketing word operation from the marketing word operation library according to the current marketing word operation data.

Twenty-sixth embodiment

In the above embodiments, a plurality of processing methods related to marketing dialogues are provided, and correspondingly, the present application also provides an electronic device. The device corresponds to the embodiment of the method. Since the apparatus embodiments are substantially similar to the method embodiments, they are described in a relatively simple manner, and reference may be made to some of the descriptions of the method embodiments for relevant points. The device embodiments described below are merely illustrative.

An electronic device of this embodiment, the electronic device includes: a processor and a memory; a memory for storing a program for implementing the above method, the device being powered on and the program for the method being run by the processor.

Although the present application has been described with reference to the preferred embodiments, it is not intended to limit the present application, and those skilled in the art can make variations and modifications without departing from the spirit and scope of the present application, therefore, the scope of the present application should be determined by the appended claims.

In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include forms of volatile memory in a computer readable medium, random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

1. Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include non-transitory computer readable media (transient media), such as modulated data signals and carrier waves.

2. As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

Claims

1. A marketing tactical recommendation system, comprising:

the server module is used for learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic; and determining a target marketing session for the first user from a plurality of marketing sessions for the received marketing session data of the first user and the second user, and sending the target marketing session to the first client module;

2. The system of claim 1, further comprising:

the server module is specifically used for sending the marketing dialogues determined by the marketing dialogues extraction model to the second client module; and receiving an auditing result returned by the second client module, and if the auditing result is yes, taking the marketing communication as an effective marketing communication;

3. The system of claim 1,

the server module is also used for determining the marketing scene to which the marketing session belongs; determining a marketing scene to which the marketing conversation belongs; and is specifically configured to determine the targeted marketing session from a plurality of marketing sessions corresponding to the marketing scenario to which the marketing session belongs, according to the marketing session data.

4. The system of claim 3,

the server module is specifically used for learning from training data of a plurality of marketing scenes in a multitask learning mode to obtain the marketing tactical prediction model and the scene classification model; the scene classification model and the marketing tactical prediction model share a coding layer; the training data includes: marketing telephone traffic data, marking data of whether the marketing telephone traffic is the marketing telephone traffic or not and marking data of a marketing scene; and determining the marketing scene to which the marketing conversation belongs and the marketing scene to which the marketing conversation belongs through the scene classification model.

5. A marketing tactic recommendation method is characterized by comprising the following steps:

learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model;

selecting marketing telephone traffic data from historical marketing telephone traffic data through a marketing telephone traffic extraction model to serve as marketing telephone traffic;

6. The method of claim 5, further comprising:

determining the auditing result of the marketing session determined by the marketing session extraction model by the third user;

7. The method of claim 5, further comprising:

determining a marketing scene to which a marketing conversation belongs;

determining a marketing scene to which the marketing conversation belongs;

the method for determining the target marketing dialogues recommended to the second user for the first user from the plurality of marketing dialogues according to the marketing dialogue data of the first user and the second user comprises the following steps:

and determining the target marketing session from a plurality of marketing sessions corresponding to the marketing scene to which the marketing session belongs according to the marketing session data.

8. The method according to claim 7,

the method for learning the marketing telephone operation extraction model from the plurality of marketing telephone operation data and the marking data of whether the marketing telephone operation data is the marketing telephone operation or not comprises the following steps:

9. The method according to claim 5,

and if the marketing session extraction condition is satisfied, selecting the marketing traffic data from the historical marketing traffic data as the marketing session through a marketing session extraction model.

10. The method according to claim 9,

the marketing session extraction conditions include: the current time is longer than the time threshold value from the time of extracting the marketing telephone operation last time;

the historical marketing traffic data comprises newly added marketing traffic data after the last marketing traffic extraction.

11. The method of any one of claims 5 to 10, wherein determining a targeted marketing dialog for the first user from a plurality of marketing dialogs based on marketing dialog data of the first user with the second user comprises:

and determining the target marketing skills according to the matching degree.

12. A marketing dialect extraction model processing method is characterized by comprising the following steps:

determining a training data set; the training data includes: marketing telephone traffic data, and marking data of whether the marketing telephone traffic data is marketing telephone traffic;

constructing a network structure of a marketing tactics extraction model;

and learning from the training data set to obtain the network parameters of the marketing strategy extraction model.

13. The method according to claim 12,

the network architecture comprises: a phonetics feature extractor and a phonetics discriminator;

and the conversational technology discriminator is used for judging whether the marketing telephone traffic data is the marketing conversational technology or not according to the conversational characteristic data.

14. The method of claim 13,

the dialogistic feature extractor includes: a word embedding layer, a text segment embedding layer, a coding layer and a language feature aggregation layer;

15. The method according to claim 14,

the training data further comprises: marketing scene annotation data;

learning from training data of a plurality of marketing scenes in a multi-task learning mode to obtain the marketing tactical prediction model and the scene classification model; the scene classification model includes the scene feature extractor and the scene classifier.

16. The method according to any one of claims 13 to 15,

the training data further comprises: marketing domain annotation data;

17. The method of claim 16,

the multi-marketing domain includes: the communication operator field, the E-business field, the education and training field and the insurance field.

18. A marketing conversation processing method, comprising:

determining historical marketing traffic data;

19. The method of claim 18, further comprising:

20. The method of claim 19, further comprising:

learning from training data of a multi-marketing scene in a multi-task learning mode to obtain the marketing tactical extraction model and the scene classification model; the scene classification model comprises: the scene feature extractor and the scene classifier share a coding layer; the training data includes: marketing telephone traffic data, marking data of whether the marketing telephone traffic is a marketing telephone operation or not and marketing scene marking data;

21. The method of claim 18 or 19, further comprising:

learning from training data of a plurality of marketing fields in a multi-task learning mode to obtain the marketing tactical extraction model and the field classification model; the domain classification model comprises: the system comprises a domain feature extractor and a domain classifier, wherein the domain feature extractor and the conversational feature extractor share a coding layer; the training data includes: marketing telephone traffic data, marking data of whether the marketing telephone traffic is the marketing telephone traffic or not and marking data of marketing fields;

22. The method of claim 18,

the determining historical marketing traffic data includes:

23. The method according to any one of claims 18 to 22,

the historical marketing traffic data includes: marketing traffic voice data;

the method further comprises the following steps:

24. The method of any of claims 18 to 22, further comprising:

25. The method of any of claims 18 to 22, further comprising:

26. A marketing conversation processing method, comprising:

receiving a marketing telephone operation determined from historical marketing telephone operation data by a marketing telephone operation extraction model sent by a server module; the marketing telephone operation extraction model is obtained by learning from a plurality of marketing telephone operation data and the marking data of whether the marketing telephone operation data is the marketing telephone operation or not;

determining an auditing result of a third user on the marketing session;

27. A marketing tactic recommendation method is characterized by comprising the following steps:

28. A marketing tactical recommendation system, comprising:

the system comprises at least one first service end module, at least one second service end module and a server, wherein the first service end module is used for sending historical marketing traffic data of a target field; and sending marketing dialog data of the first user and the second user;

29. The system of claim 28,

and the first service end module is specifically used for sending the historical marketing traffic data according to a preset time interval, wherein the historical marketing traffic data comprises newly added marketing traffic data in a target field in the time interval.

30. The system of claim 28,

the target fields corresponding to different first service end modules comprise: the communication operator field, the E-business field, the education and training field and the insurance field.

31. A marketing conversation recommendation method is characterized by comprising the following steps:

32. A marketing conversation recommendation method is characterized by comprising the following steps:

33. A marketing tactic recommendation method is characterized by comprising the following steps:

34. A customer service robot system, comprising:

and the client module is used for displaying the target marketing words.

35. A robotic dialog method, comprising:

and determining a target marketing conversation aiming at the target user from the plurality of marketing conversations according to the marketing conversation data of the target user.

36. A robotic dialog method, comprising:

receiving a target marketing conversation aiming at a target user and sent by a server module; the server module determines the target marketing session in the following manner: learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; determining a target marketing conversation aiming at the target user from a plurality of marketing conversations according to the marketing conversation data of the target user;

and displaying the target marketing words.

37. A marketing tactic recommendation apparatus, comprising:

38. A marketing conversation recommendation apparatus, comprising:

and the recommended speech presentation unit is used for presenting a target marketing speech so that the second user can conveniently talk with the first user according to the target marketing speech.

39. A marketing tactics extraction model processing apparatus, comprising:

40. A marketing tactical processing apparatus, comprising:

the telephone traffic data determining unit is used for determining historical marketing telephone traffic data;

the conversational feature extraction unit is used for extracting the conversational feature extractor in the model through marketing conversational to determine the conversational feature data of the historical marketing telephone traffic data;

and the conversational judging unit is used for extracting the conversational technology discriminator in the model through the marketing conversational technology and judging whether the historical marketing telephone traffic data is the marketing conversational technology or not according to the characteristic data.

41. A marketing tactical processing apparatus, comprising:

the word operation auditing unit is used for determining the auditing result of the third user on the marketing word operation;

42. A marketing conversation recommendation apparatus, comprising:

the current dialogue data sending unit is used for sending marketing dialogue data of the first user and the second user to the server module, so that the server module determines a target marketing dialogue for the first user from a plurality of marketing dialogues in a target field according to the marketing dialogue data, and the second user can conveniently dialogue with the first user according to the target marketing dialogue.

43. A marketing conversation recommendation apparatus, comprising:

the system comprises a dialogue mining unit, a marketing dialogue extracting unit and a marketing dialogue extracting unit, wherein the dialogue mining unit is used for extracting a marketing dialogue data from historical marketing dialogue data of a target field through a marketing dialogue extracting model and taking the marketing dialogue data as the marketing dialogue of the target field;

44. A marketing tactic recommendation apparatus, comprising:

45. A robotic dialog device, comprising:

46. A robotic dialog device, comprising:

the recommendation language and technology receiving unit is used for receiving the target marketing language and technology aiming at the target user and sent by the server-side module; the server module determines the target marketing communication in the following way: learning from the marketing telephone traffic data and the marking data of whether the marketing telephone traffic data is the marketing telephone traffic or not to obtain a marketing telephone traffic extraction model; selecting marketing telephone traffic data from historical marketing telephone traffic data as marketing telephone traffic through a marketing telephone traffic extraction model; determining a target marketing conversation aiming at the target user from a plurality of marketing conversations according to the marketing conversation data of the target user;

47. An electronic device, comprising:

a processor and a memory;

a memory for storing a program for implementing the method of any one of claims 5 to 27, 31 to 33, 35, 36, the device being powered on and the program for the method being run by the processor.

48. A marketing conversation recommendation method is characterized by comprising the following steps:

acquiring current marketing dialogue data;

49. The method of claim 48, wherein determining a targeted marketing session from a library of marketing sessions based on current marketing session data comprises:

determining the matching degree of the historical contextual dialogue data associated with the marketing tactics and the current marketing dialogue data;

and determining the target marketing skills according to the matching degree.