CN113836275A

CN113836275A - Conversation model establishing method and device

Info

Publication number: CN113836275A
Application number: CN202010514454.6A
Authority: CN
Inventors: 曹元斌; 吴胜兰; 殷浩; 邵明光; 唐攀攀; 白明智
Original assignee: Cainiao Smart Logistics Holding Ltd
Current assignee: Cainiao Smart Logistics Holding Ltd
Priority date: 2020-06-08
Filing date: 2020-06-08
Publication date: 2021-12-24
Anticipated expiration: 2040-06-08
Also published as: CN113836275B

Abstract

The invention discloses a method and a device for establishing a conversation model. Wherein, the method comprises the following steps: generating a segmentation log by using a question-answer sequence, wherein the segmentation log comprises a plurality of question-answer pairs converted from the question-answer sequence; carrying out duplicate removal and vectorization operation on the text in the segmented log to obtain a vector text; generating a combined log according to the vector text and the segmentation log; generating a conversation flow chart according to the merging log; and executing preset operation through the conversation flow chart. The method and the device for establishing the dialogue model provided by the embodiment of the invention solve the technical problems of low efficiency and inadequacy in establishing the dialogue system purely manually in the prior art.

Description

Conversation model establishing method and device

Technical Field

The invention relates to the field of intelligent learning models, in particular to a method and a device for establishing a dialogue model.

Background

With the continuous development of intelligent learning models, learning models are applied in various fields, such as commerce, military, production, life, and the like. At present, in a customer service conversation robot scene, a business scene needs to be accurately modeled in advance, and how the robot processes business problems through conversation is modeled to simulate a coping strategy of real person customer service, so that the customer service problem is solved.

Generally, a conversation robot system is customized on company business, business modeling of the conversation system is the first step of work, two categories of automatic extraction and manual construction are divided into algorithms in the traditional method, the automatic extraction usually needs a large amount of data, and due to lack of supervision and labeling, the modeled conversation often has insufficient business subdivision or is too delicate, so that important business nodes are omitted, or the user energy is wasted in the absence of so-called conversation, and therefore the method cannot be well applied to real business scenes.

Since a purely automatic solution cannot serve a business well, the industry uses a strategy constructed by human, and the traditional method of constructing by human, i.e. modeling by domain experts, needs to go through the following processes: firstly, various queries are processed by manual customer service on the basis of an existing customer service; at the beginning of a project, a business expert combs out tasks needing to be processed by the robot, combs a conversation process and sorts business requirements; and (4) according to the sorted conversation process and business requirements and the historically accumulated manual customer service data, an engineer constructs a conversation system and an algorithm model, and deploys online.

In the business modeling work of business experts, the business experts often need to complete business modeling by experience or manually combing conversation logs. There are two problems with this: firstly, when the business is huge, the work efficiency of business experts directly influences the progress of the whole project, and due to the fact that the work is purely manual work and expert knowledge is needed, the efficiency is unlikely to be improved by adding one person at any time; secondly, the business experts have their own work centers of gravity in a certain fixed field, and the amount of the dialogue logs which can be combed in a limited time is limited, so that the business modeling of the business experts inevitably deviates from the scene which needs to be covered by the real business scene, or a more biased point is built emphatically, or the important point which needs to be covered is ignored, so that the later algorithm development deviates from the direction; the above two points are inevitable problems of manual work.

In view of the above problems, no effective solution has been proposed.

Disclosure of Invention

The embodiment of the invention provides a method and a device for establishing a conversation model, which are used for at least solving the technical problems of low efficiency and inadequacy in purely manually establishing a conversation system in the prior art.

According to an aspect of the embodiments of the present invention, there is provided a method for building a dialogue model, including: generating a segmentation log by using a question-answer sequence, wherein the segmentation log comprises a plurality of question-answer pairs converted from the question-answer sequence; carrying out duplicate removal and vectorization operation on the text in the segmented log to obtain a vector text; generating a combined log according to the vector text and the segmentation log; generating a conversation flow chart according to the merging log; and executing preset operation through the conversation flow chart.

Optionally, before the generating a segmentation log by using the question-answer sequence, where the segmentation log includes a plurality of question-answer pairs converted from the question-answer sequence, the method further includes: and acquiring the question-answer sequence.

Optionally, the performing deduplication and vectorization operations on the text in the segmented log to obtain a vector text includes: carrying out duplicate removal operation on the repeated dialog text in the segmented log; and performing text vectorization operation on the dialog text subjected to the deduplication operation to generate the vector text.

Optionally, the generating a merged log according to the vector text and the split log includes: merging the vector text and the segmentation log to obtain a first merging result; clustering the first combination result to generate a clustering result; and generating the merging log according to the first merging result and the clustering result.

Optionally, the generating the merged log according to the merged result and the clustering result includes: merging the first merging result and the clustering result to obtain a second merging result; and generating the merging log according to the second merging result.

Optionally, the generating a session flowchart according to the merged log includes: acquiring the number of the same sessions in the merged log; sequencing the number of the same sessions according to a preset rule to obtain a sequencing result; and generating the conversation flow chart according to the sequencing result.

Optionally, the preset operation includes at least one of: sorting, merging and inserting.

Optionally, after the preset operation is executed through the session flowchart, the method further includes: and constructing a conversation model according to the conversation flow chart.

According to another aspect of the embodiments of the present invention, there is also provided a method for establishing a logistics customer service session, applied to a logistics return customer service end, including: generating a return segmentation log by using a return question-answer sequence, wherein the return segmentation log comprises a plurality of return question-answer pairs converted from the return question-answer sequence; performing duplicate removal and vectorization operation on the text in the return cut log to obtain a vector text; generating a combined log according to the vector text and the returned goods segmentation log; generating a return conversion flow chart according to the combined log; and executing preset operation through the return conversion flow chart.

According to another aspect of the embodiments of the present invention, there is also provided a dialog model application method, including: acquiring dialogue model data; extracting training features according to the dialogue model data; and training a dialogue management model and an intention recognition model according to the training characteristics.

Optionally, the acquiring dialog model data includes: acquiring batch conversation models; and acquiring batch dialogue model data according to the batch dialogue models.

Optionally, after the training of the dialogue management model and the intention recognition model according to the training features, the method further includes: sending the trained dialogue management model and the intention recognition model to a dialogue system; and completing the construction of the dialogue system according to the dialogue management model and the intention recognition model.

According to another aspect of the embodiments of the present invention, there is also provided a satisfaction dialogue model application method applied to customer service satisfaction feedback, including: obtaining satisfaction dialogue model data; extracting training characteristics according to the satisfaction degree dialogue model data; training a dialogue management model and an intention recognition model according to the training characteristics; wherein, the satisfaction dialogue model data is established and formed by the dialogue model establishing method.

According to another aspect of the embodiments of the present invention, there is also provided a dialog model building apparatus, including: the segmentation log module is used for generating a segmentation log by using the question-answer sequence, wherein the segmentation log comprises a plurality of question-answer pairs converted from the question-answer sequence; the vector text module is used for carrying out duplication removal and vectorization operations on the text in the segmented log to obtain a vector text; the combined log module is used for generating a combined log according to the vector text and the segmentation log; the generating module is used for generating a session flow chart according to the merging log; and the execution module is used for executing preset operation through the conversation flow chart.

Optionally, the apparatus further comprises: and the acquisition module is used for acquiring the question-answer sequence.

Optionally, the vector text module includes: the duplication removing unit is used for carrying out duplication removing operation on the repeated dialog texts in the segmentation log; and the vector unit is used for performing text vectorization operation on the dialog text subjected to the deduplication operation to generate the vector text.

Optionally, the merge log module includes: the first merging unit is used for merging the vector text and the segmentation log to obtain a first merging result; the clustering unit is used for clustering the first combination result to generate a clustering result; and the generating unit is used for generating the merging log according to the first merging result and the clustering result.

Optionally, the generating unit includes: the second merging unit is used for merging the first merging result and the clustering result to obtain a second merging result; the generating unit is further configured to generate the merging log according to the second merging result.

Optionally, the generating module includes: an obtaining unit, configured to obtain the number of identical sessions in the merged log; the sorting unit is used for sorting the number of the same sessions according to a preset rule to obtain a sorting result; and the generating unit is used for generating the conversation flow chart according to the sequencing result.

Optionally, the apparatus further comprises: and the modeling module is used for constructing a conversation model according to the conversation flow chart.

According to another aspect of the embodiments of the present invention, there is also provided a dialogue model application apparatus, including: the acquisition module is used for acquiring dialogue model data; the extraction module is used for extracting training characteristics according to the dialogue model data; and the training module is used for training a dialogue management model and an intention recognition model according to the training characteristics.

Optionally, the obtaining module includes: the acquisition unit is used for acquiring batch conversation models; the obtaining unit is further configured to obtain batch dialogue model data according to the batch dialogue models.

Optionally, the apparatus further comprises: the transmitting module is used for transmitting the trained dialogue management model and the intention recognition model to a dialogue system; and the construction module is used for finishing the construction of the dialogue system according to the dialogue management model and the intention recognition model.

According to another aspect of the embodiments of the present invention, there is also provided a logistics customer service session establishing apparatus, applied to a logistics return customer service end, including:

the cutting log module is used for generating a return cutting log by utilizing a return question-answer sequence, wherein the return cutting log comprises a plurality of return question-answer pairs converted from the return question-answer sequence;

the vector text module is used for carrying out duplication removal and vectorization operation on the text in the return cut log to obtain a vector text;

the combined log module is used for generating a combined log according to the vector text and the returned goods segmentation log;

the generating module is used for generating a return conversion flow chart according to the combined log;

and the execution module is used for executing preset operation through the return and exchange conversation flow chart.

According to another aspect of the embodiments of the present invention, there is also provided a satisfaction dialogue model application apparatus applied to customer service satisfaction feedback, including:

the acquisition module is used for acquiring satisfaction dialogue model data;

the extraction module is used for extracting training characteristics according to the satisfaction degree dialogue model data;

the training module is used for training a dialogue management model and an intention recognition model according to the training characteristics;

wherein the satisfaction dialogue model data is established by the dialogue model establishing device.

According to another aspect of embodiments of the present invention, there is also provided a computer program product comprising instructions which, when run on a computer, cause the computer to perform a dialogue model building method.

According to another aspect of the embodiments of the present invention, there is also provided a non-volatile storage medium, which is characterized in that the non-volatile storage medium includes a stored program, wherein the program controls a device in which the non-volatile storage medium is located to execute a dialogue model establishment method when running.

According to another aspect of the embodiments of the present invention, there is also provided an electronic device, including a processor and a memory; the memory has stored therein computer readable instructions for execution by the processor, wherein the computer readable instructions when executed perform a method of dialog model building.

In the embodiment of the invention, a question-answer sequence is used for generating a segmentation log, wherein the segmentation log comprises a plurality of question-answer pairs converted from the question-answer sequence; carrying out duplicate removal and vectorization operation on the text in the segmented log to obtain a vector text; generating a combined log according to the vector text and the segmentation log; generating a conversation flow chart according to the merging log; through the mode of executing the preset operation by the conversation flow chart, through the mode of processing and vectorizing the text and constructing the model, the aim of intelligently and automatically constructing the conversation model and manually constructing the conversation model is fulfilled, and the technical problems that the efficiency of purely manually establishing the conversation system is low and the conversation system is not objective in the prior art are solved.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:

FIG. 1 is a flow diagram of a conversation model establishment method in accordance with an embodiment of the present invention;

FIG. 2 is a flow diagram of a method of applying a conversation model in accordance with an embodiment of the present invention;

fig. 3 is a block diagram of a dialogue model application apparatus according to an embodiment of the present invention;

fig. 4 is a block diagram of a dialogue model application apparatus according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of a dialogue model building method according to an embodiment of the invention;

FIG. 6 is a schematic diagram of a conversation model application method in accordance with an embodiment of the present invention;

FIG. 7 is a flowchart of a method for establishing a logistics customer service session according to an embodiment of the invention;

FIG. 8 is a flow diagram of a method of applying a satisfaction dialogue model in accordance with an embodiment of the present invention;

fig. 9 schematically shows a block diagram of a terminal device for performing the method according to the invention; and

fig. 10 schematically shows a storage unit for holding or carrying program code implementing the method according to the invention.

Detailed Description

In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

In accordance with an embodiment of the present invention, there is provided a method embodiment of a dialogue model establishment method, it is noted that the steps illustrated in the flowchart of the figure may be performed in a computer system such as a set of computer-executable instructions and that, although a logical order is illustrated in the flowchart, in some cases the steps illustrated or described may be performed in an order different than here.

Example one

Fig. 1 is a flowchart of a dialogue model building method according to an embodiment of the present invention, as shown in fig. 1, the method includes the following steps:

step S101, generating a segmentation log by using a question-answer sequence, wherein the segmentation log comprises a plurality of question-answer pairs converted from the question-answer sequence.

Specifically, the question-answer sequence may be a dialog log. The dialogue can be segmented and assembled, the question-answer sequence of customer service and the user in the same dialogue is converted into question-answer pairs, namely, one record contains the one-time question-answer interaction of two continuous people in one dialogue and the number of dialogue rounds, and a segmentation log is obtained and can be a segmentation log (2); for example, in one embodiment, the question-and-answer sequence for customer service and user includes the following:

q1: is package posted?

A1: the mail is not wrapped.

Q2: is the Tibet area?

A2: can be used in Tibet region.

Q3: delivery on several days?

A3: and shipping within 24 hours.

The first round of session includes (Q1, a1), the second round of session includes (Q2, a2), and the third round of session includes (Q3, A3). The split log is used to record the number of dialog rounds in which all the question-answer interactions have been performed, such as the first round, the second round, and the third round. The same session may refer to a session between one customer service ID and one user ID, or may refer to a session between one user ID and different customer service IDs for the same question, and the like.

For example, when the question-answer sequence of the embodiment of the invention occurs on the Taobao application software, the dialog between the Taobao customer service personnel and the customer who consumes the Taobao forms the question-answer sequence data, and the dialog model is established according to the question-answer sequence data generated on the Taobao application

The question-answer sequence may be obtained from the user's history data stored in the server. Or from an instant dialogue of user interaction with the customer service.

The question-answer sequence is converted into question-answer pairs, and the question-answer sequence is divided into a plurality of rounds of conversations. The segmentation log generated based on the question-answer sequence comprises question-answer pairs formed by converting the question-answer sequence. Such as the first to third round of sessions previously described.

Specifically, since the embodiment of the present invention needs to process the question-answer sequence, before generating the split log, the server or the client of the instant conversation needs to obtain the historical or real-time question-answer sequence. And sending a request for acquiring the question and answer sequence to a server storage area or a conversation client through a processor of the conversation system, calling the locally existing question and answer sequence according to the request for acquiring by the corresponding server storage area or the conversation client, and sending the question and answer sequence data to the processor to wait for the next step of processing the question and answer sequence.

And step S102, carrying out duplicate removal and vectorization operation on the text in the segmentation log to obtain a vector text.

Specifically, in order to construct the dialogue model according to the embodiment of the present invention, the segmentation log needs to be vectorized, the vectorization may be to perform vectorization on data units in the segmentation log, and perform construction and analysis on the dialogue statements according to the segmentation log after the whole vectorization, for example, after the segmentation log is vectorized, the processor may generate a set of required dialogue statements according to a vectorization segmentation log result, and process the dialogue statements according to the set.

Specifically, when the segmentation log vectorization is processed, all the dialog texts in the dialog log can be extracted, and the de-duplication summarization is performed to form the summarized dialog texts. Secondly, vectorizing the dialog text, wherein a whole sentence of text can be converted into a fixed-length vector by using technologies such as bert or word2vec, and the batch of text and a vector representation corresponding to the batch of text are obtained; wherein, both bert and word2vec are deep learning model algorithms.

It should be noted that, for the dialog log, all dialog texts may be extracted, and if necessary, deduplication operation may be performed, and if the log text itself does not have the deduplication requirement, vectorization operation may be directly performed, so as to obtain a vector text corresponding to the split log.

For example, a deduplication operation for a conversation log may be the following: the customer service representative's dialog is "please confirm for package information. ", the customer service representative continues to send a dialog" please confirm the package information again because the customer did not enter the confirmation information correctly. If "please confirm the package information" and "please confirm the package information again" in the dialog of the customer service representative can be determined as repeated dialog records, and the meaning of the expression indicates that the customer wants to confirm the package information, then one of the two same dialogs can be removed in the duplication removing operation to simplify the dialog log and improve the accuracy of the dialog modeling.

And step S103, generating a combined log according to the vector text and the segmentation log.

Specifically, the merging of the vector text and the split log to obtain the first merging result may be to merge the split log formed by the previously prepared question-answer pairs with the text vector generated just before, and obtain a merged log by vector representation of the supplementary question for each question-answer pair and vector representation of the answer. In the combined log, each log record comprises a group of user questions and answers, the number of turns of the group of user questions and answers in the question-answer conversation between the user and the customer service, and text vectors corresponding to two sentences of the group of questions and answers respectively; for example, in the log after merging, each log record may include the following information: the log data of the dialog segmentation between the customer service representative and the customer and the dialog vector data in all the segmentation logs.

It should be noted that the first merging result is a result of merging vector data and segmented log data, and the clustering operation on the result includes that different rounds can be performed, and the question and answer texts in each round are subjected to text clustering, for example, a k-means algorithm is used to cluster contents with similar semantics into one class, so as to obtain a clustering result of each round of conversation. The k-means algorithm needs to appoint the number k of clusters in advance, the algorithm starts to randomly select k recording points as central points, then traverses each record of the whole data set, puts each record in the cluster where the central point nearest to the record is located, then replaces the previous central point with the mean central point of the records of each cluster, and then continuously iterates until convergence. K-Means is a common clustering algorithm, and compared with other clustering algorithms, the time complexity is low, and the clustering effect is good.

The logs just merged and the generated clustering results may be merged again, and each merged record includes a question-answer pair, a session to which the question-answer pair belongs, a turn number of the question-answer pair, a text vector corresponding to each item of the question-answer pair, and a clustering center corresponding to the vector.

Specifically, each of the merged records may include, for example, the following: question and answer pairs: for example, one question information and one answer information, i.e., the question-answer pair itself; the question-answer pairs belong to the sessions: for example embodied in the form of a session number; number of rounds in which question-answer pairs are located: for example, the question-answer pair is in the second turn of the dialog. Wherein, the text vector corresponding to each item comprises: and utilizing a text vector generated by corresponding to each text item and a clustering center obtained from the text vector.

And step S104, generating a conversation flow chart according to the merged log.

Specifically, after the merged log is generated according to the clustering operation, the embodiment of the present invention needs to generate a session flowchart according to the merged log, where the session flowchart is used for modeling the dialog system in combination with human labor. The conversation flow chart can be a conversation list obtained by calculation and sorting according to conversation frequency, and is used for representing which conversations are frequently occurring and which conversations are not frequently occurring, and the conversations are matched with corresponding answers, namely question-answer pairs of the client and the customer service representative.

Specifically, in order to count and analyze the session data in the merged log, the number of the same sessions in the merged log is obtained first, and the log may be counted according to the following statistics: and counting the number of sessions clustered into sessions with the same question and the same answer under the same session turn. After statistics, all similar question-answer pairs are counted together at the same position of the conversation, then the statistical result is inverted according to the conversation number to represent the current conversation position, and the conversation is sorted according to a preset rule, wherein the preset rule can be a sorting method from high to low, and then the conversation system processor calculates which kinds of conversations appear and the frequency of the conversations according to the statistical data and a sorting algorithm. With such statistics, a flow chart of the conversation can be constructed, wherein the drawing flow chart is a directed graph, and the flow direction is pointed to the next round by the previous round of the question and answer. Finally, in an optional embodiment, the long-tailed low-frequency question-answer pair can be cut off to finally form a conversation flow, and the conversation flow is led into a business modeling system for business expert modeling conversation.

And step S105, executing preset operation through the conversation flowchart.

Specifically, after the session flow chart is established by the session establishing method of the embodiment of the present invention, the manual operation may perform operations of adding, deleting, modifying and checking according to the session flow chart, and the session flow chart is modified to achieve a better session model establishing result.

Specifically, when too many service nodes are found to be clustered together, re-clustering may be performed. Different clustering nodes correspond to the same intent, and can be merged. Alternatively, a function node may be inserted in the conversation flow. When the clustering based on the text is found, too many different service nodes are clustered together, a service expert can select to edit the node, specify the clustering number and re-cluster the node so as to achieve the purpose of service node subdivision, for example, under default clustering, "good", "can", "etc", "I want to wait" are clustered together, two types of "good", "can" be grouped into one type "," etc "," I want to wait "are grouped into one type again, and for the condition that the differentiation cannot be performed, additional features and rules can be added to distinguish the nodes;

in another case, if different cluster nodes are found, which actually correspond to the same intention in the service, the two types can be merged together, so that all the subsequent child nodes are merged into the same parent node in the tool, for example, "do you have what can help you? "and" parent, your good ", have gathered into two kinds, correspond to actually" and user call up "this intention, so can collude and select two nodes from the picture and merge, achieve the purpose of simplifying the procedure.

Finally, functional nodes are inserted in the conversation flow, such as 'order information query', 'express mail position query' and the like, and conversation configuration is completed. The service experts model the service scene conversation flow by adding, deleting and modifying the counted conversation flow, so that the efficiency is high, important service nodes cannot be ignored due to blind spots of people, the importance of the service nodes can be judged according to the statistical indexes, and excessive energy is not wasted on the irrelevant nodes.

Specifically, according to the conversation flow chart generated by the process, a complete conversation model can be constructed by adding, deleting, modifying and checking operations manually, and the model can play a role in automatically replying for different application scenes according to conversation subjects or contents, so that the working efficiency in business application is increased.

Through the steps, the technical effect of objectively and efficiently establishing the dialog system can be realized

Example two

Fig. 2 is a flowchart of a dialogue model application method according to an embodiment of the present invention, as shown in fig. 2, the method includes:

in step S201, dialogue model data is acquired.

Specifically, the dialogue model data may be obtained according to the preceding embodiment, and the contents of the dialogue data and the like in the dialogue model are extracted for the application of the dialogue model in the embodiment.

Specifically, since the number of the dialogue models is more than one in any application scenario, the dialogue models need to be acquired in batch through the dialogue system, and analyzed and processed according to the data of the dialogue models. For example, when a user purchases online by using panning, the user needs to ask and answer a customer service representative for a goods logistics situation, and then ask and answer the customer service representative for a quality problem of goods, two types of asking and answering are needed to be implemented, so that batch acquisition of the logistics and quality dialogue models is needed when acquiring the dialogue models.

Step S202, extracting training characteristics according to the dialogue model data;

specifically, in the foregoing embodiment, the business conversation model is constructed through conversation modeling, and the rules and the cluster model obtained after the construction, which may be referred to as a conversation flow model, are used to distinguish different conversation flows from logs. After the dialog flow model is obtained in the embodiment, the training features required by each model of the dialog can be separated from the log in batches. For example, in dialog management, it is necessary to train a dialog management model with jumps through dialogs; in the idea recognition, different intentions of the user and data of blind child nodes corresponding to each parent node in one round of conversation need to be judged.

Step S203, training a dialogue management model and an intention recognition model according to the training characteristics.

Specifically, the extracted features are used for respectively training a dialogue management model and an intention recognition model, and the trained models are sent to a final dialogue system to complete the construction of the dialogue system. And then, respectively training a dialogue management model and an intention recognition model by the extracted features, and sending the trained models to a final dialogue system to complete the construction of the dialogue system. For example, two models have been trained, wherein a dialog management model is used to manage a user's dialog flow and an intention recognition model is used to recognize the user's intention, and a complete dialog flow can be generated by the dialog management model and the intention recognition model.

The technical effect of establishing the dialog system objectively and efficiently can be achieved by the device.

EXAMPLE III

Fig. 3 is a block diagram of a dialogue model application apparatus according to an embodiment of the present invention, as shown in fig. 3, the apparatus including:

the log splitting module 301 is configured to generate a split log by using a question-answer sequence, where the split log includes a plurality of question-answer pairs converted from the question-answer sequence.

q1: is package posted?

A1: the mail is not wrapped.

Q2: is the Tibet area?

A2: can be used in Tibet region.

Q3: delivery on several days?

A3: and shipping within 24 hours.

The vector text module 302 is configured to perform deduplication and vectorization operations on the text in the split log to obtain a vector text.

And the merging log module 303 is configured to generate a merging log according to the vector text and the split log.

Specifically, in order to obtain a merged log for modeling, the log that has just been merged and the generated clustering result may be merged again, and each record after merging includes a question-answer pair, a session to which the question-answer pair belongs, the number of turns in which the question-answer pair is located, a text vector corresponding to each item of the question-answer pair, and a clustering center corresponding to the vector.

A generating module 304, configured to generate a session flowchart according to the merged log.

And the execution module 305 is configured to execute a preset operation through the session flowchart.

Optionally, a device for establishing a logistics customer service session is further provided, which is applied to a logistics return customer service end, and includes:

Example four

Fig. 4 is a block diagram of a dialogue model application apparatus according to an embodiment of the present invention, as shown in fig. 4, the apparatus including:

an obtaining module 401, configured to obtain the dialogue model data.

An extraction module 402, configured to extract training features according to the dialogue model data;

A training module 404, configured to train a dialogue management model and an intention recognition model according to the training features.

Optionally, the embodiment further provides a satisfaction dialogue model application apparatus, applied to customer service satisfaction feedback, including:

the acquisition module is used for acquiring satisfaction dialogue model data;

EXAMPLE five

Fig. 5 is a schematic diagram of a dialogue model building method according to an embodiment of the present invention, as shown in fig. 5, the method includes:

the dialogue log contains a plurality of dialogue text data, the dialogue log is segmented and assembled to generate a segmentation log, the dialogue text in the log is vectorized, then data merging is carried out according to the vectorized dialogue text and the segmentation log generated by segmentation and assembly, the merged data are clustered, the clustered data and the merged data are merged for the second time, finally, statistics of the number of the same terms of question and answer is carried out according to the merged data after the second time merging, a directed graph is constructed, the directed graph is a conversation flow graph, a person constructing the dialogue model can carry out addition and deletion check according to the conversation flow graph to complete the final dialogue model and is applied to an operation tool, and the operation tool can be an online shopping tool such as Taobao and the like.

Fig. 6 is a schematic diagram of a dialogue model application method according to an embodiment of the present invention, as shown in fig. 5, the method includes: .

The dialog log comprises a plurality of characteristic values which are used for training a dialog management model and an intention recognition model, wherein the dialog management model is used for controlling the accuracy and the answering efficiency of the robot customer service when answering the user questions, the intention recognition model is used for recognizing what answers the user wants to obtain, and the whole user communication service process, namely the construction of the whole dialog system, can be completed by combining the dialog management model according to the intention of the user. The extraction of the feature values needs to be extracted and called from a dialog flow model established by a dialog log, which parameters are specifically defined as the feature values, and needs to be determined according to an actual application scenario by a user, for example, when the user performs online shopping, the user can extract data such as package-oriented dialog question-answer data and goods-oriented dialog question-answer data in the dialog flow model and define the data as the feature values to train a dialog management model and an intention recognition model, and establish a dialog system.

EXAMPLE six

Fig. 7 is a flowchart of a logistics customer service session establishment method according to an embodiment of the present invention, as shown in fig. 7, the method includes:

step S701, generating a return cutting log by using the return question-answer sequence, wherein the return cutting log comprises a plurality of return question-answer pairs converted from the return question-answer sequence.

Step S702, the text in the return cutting log is subjected to duplication removal and vectorization operation to obtain a vector text.

And step S703, generating a combined log according to the vector text and the returned goods split log.

Step S704, a return conversion flow chart is generated according to the combined log.

Step S705, executing a preset operation through the return exchange session flowchart.

Specifically, the present embodiment is used for the customer service to perform a dialogue with the user and perform a dialogue model establishment for the return goods according to the method of the present embodiment when the return goods occurs, for example: the return transaction log contains several text data of the transaction, such as: "ask: can goods be returned? (ii) a Answering: may! "" ask: when to return the goods? (ii) a Answering: this evening! "" ask: is the return package postage? (ii) a Answering: of course, the package is stamped! Then the return commodity conversation log is cut and assembled to generate a return commodity cutting log, and the return commodity conversation text in the log is vectorized, then data merging is carried out according to the vectorized dialogue text and the returned goods segmentation log generated by segmentation and assembly, at the same time, the combined data is clustered, the clustered data and the combined data are combined for the second time, finally, the number statistics of the same entries of question and answer is carried out according to the combined data after the second time combination, and a directed graph is constructed, the directed graph is a return conversation flow chart, specifically, for the dialogue model in the return flow, the person who establishes the return dialogue model can perform the modification and check according to the conversation flow chart to complete the final return dialogue model, and the model is applied to an operation tool, which can be an online shopping tool such as panning.

EXAMPLE seven

Fig. 8 is a flowchart of a satisfaction dialogue model application method according to an embodiment of the present invention, as shown in fig. 8, the method comprising:

in step S801, satisfaction dialogue model data is acquired.

And S802, extracting training characteristics according to the satisfaction degree dialogue model data.

Step S803, training a dialogue management model and an intention recognition model according to the training features.

Specifically, the present embodiment is applied to train the existing dialogue model by using the response result of the user to the customer service satisfaction as the training value, so as to achieve the technical effect of perfecting the existing dialogue model through the feedback of the user. For example: the satisfaction dialogue log contains a plurality of characteristic values which are used for training dialogue management models and intention recognition models, such as question: do you also satisfy? (ii) a Answering: unsatisfied! "" ask: where is dissatisfied? (ii) a Answering: i want no one to do me to return the exchange goods, feel ignored! ", there is a satisfaction feedback about the user's satisfaction with the current service, such as a return dissatisfaction. The dialogue management model is used for controlling the answer accuracy and the answer efficiency of the robot customer service in answering the user questions, the intention identification model is used for identifying what answers the user wants to obtain, and the whole user communication service process, namely the construction of the whole dialogue system, can be completed by combining the dialogue management model according to the intention of the user. The extraction of the satisfaction degree characteristic value needs to be extracted and called from a conversation flow model established by a satisfaction degree conversation log, which satisfaction degree parameters are specifically defined as the characteristic values, and needs to be determined by a merchant according to an actual application scene.

Specifically, the method comprises the following steps: generating a segmentation log by using a question-answer sequence, wherein the segmentation log comprises a plurality of question-answer pairs converted from the question-answer sequence; carrying out duplicate removal and vectorization operation on the text in the segmented log to obtain a vector text; generating a combined log according to the vector text and the segmentation log; generating a conversation flow chart according to the merging log; and executing preset operation through the conversation flow chart.

By the method, the technical effect of objectively and efficiently establishing the dialog system can be achieved.

The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.

In the above embodiments of the present invention, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.

Fig. 9 is a schematic diagram of a hardware structure of a terminal device according to an embodiment of the present application. As shown in fig. 9, the terminal device may include an input device 90, a processor 91, an output device 92, a memory 93, and at least one communication bus 94. The communication bus 94 is used to enable communication connections between the elements. The memory 93 may comprise a high speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, in which various programs may be stored in the memory 93 for performing various processing functions and implementing the method steps of the present embodiment.

Alternatively, the processor 91 may be implemented by, for example, a Central Processing Unit (CPU), an Application Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), a Programmable Logic Device (PLD), a Field Programmable Gate Array (FPGA), a controller, a microcontroller, a microprocessor, or other electronic components, and the processor 91 is coupled to the input device 90 and the output device 92 through a wired or wireless connection.

Alternatively, the input device 90 may include a variety of input devices, such as at least one of a user-oriented user interface, a device-oriented device interface, a software-programmable interface, a camera, and a sensor. Optionally, the device interface facing the device may be a wired interface for data transmission between devices, or may be a hardware plug-in interface (e.g., a USB interface, a serial port, etc.) for data transmission between devices; optionally, the user-facing user interface may be, for example, a user-facing control key, a voice input device for receiving voice input, and a touch sensing device (e.g., a touch screen with a touch sensing function, a touch pad, etc.) for receiving user touch input; optionally, the programmable interface of the software may be, for example, an entry for a user to edit or modify a program, such as an input pin interface or an input interface of a chip; optionally, the transceiver may be a radio frequency transceiver chip with a communication function, a baseband processing chip, a transceiver antenna, and the like. An audio input device such as a microphone may receive voice data. The output device 92 may include a display, a sound, or other output device.

In this embodiment, the processor of the terminal device includes a module for executing the functions of the modules of the data processing apparatus in each device, and specific functions and technical effects may refer to the foregoing embodiments, which are not described herein again.

Fig. 10 is a schematic diagram of a hardware structure of a terminal device according to another embodiment of the present application. FIG. 10 is a specific embodiment of the implementation of FIG. 9. As shown in fig. 8, the terminal device of the present embodiment includes a processor 101 and a memory 102.

The processor 101 executes the computer program code stored in the memory 102 to implement the methods of fig. 1-2 in the above embodiments.

The memory 102 is configured to store various types of data to support operations at the terminal device. Examples of such data include instructions for any application or method operating on the terminal device, such as messages, pictures, videos, and so forth. The memory 102 may include a Random Access Memory (RAM) and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

Optionally, the processor 101 is provided in the processing assembly 100. The terminal device may further include: a communication component 103, a power component 104, a multimedia component 105, an audio component 106, an input/output interface 107 and/or a sensor component 108. The specific components included in the terminal device are set according to actual requirements, which is not limited in this embodiment.

The processing component 100 generally controls the overall operation of the terminal device. The processing assembly 100 may include one or more processors 101 to execute instructions to perform all or part of the steps of the above-described method. Further, the processing component 100 can include one or more modules that facilitate interaction between the processing component 100 and other components. For example, the processing component 100 may include a multimedia module to facilitate interaction between the multimedia component 105 and the processing component 100.

The power supply component 104 provides power to the various components of the terminal device. The power components 104 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the terminal device.

The multimedia component 105 includes a display screen that provides an output interface between the terminal device and the user. In some embodiments, the display screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the display screen includes a touch panel, the display screen may be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation.

The audio component 106 is configured to output and/or input audio signals. For example, the audio component 106 may include a Microphone (MIC) configured to receive external audio signals when the terminal device is in an operational mode, such as a voice recognition mode. The received audio signal may further be stored in the memory 102 or transmitted via the communication component 103. In some embodiments, the audio component 106 also includes a speaker for outputting audio signals.

The input/output interface 107 provides an interface between the processing component 100 and peripheral interface modules, which may be click wheels, buttons, etc. These buttons may include, but are not limited to: a volume button, a start button, and a lock button.

The sensor component 108 includes one or more sensors for providing various aspects of status assessment for the terminal device. For example, the sensor component 108 can detect the open/closed status of the terminal device, the relative positioning of the components, the presence or absence of user contact with the terminal device. The sensor assembly 108 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact, including detecting the distance between the user and the terminal device. In some embodiments, the sensor assembly 108 may also include a camera or the like.

The communication component 103 is configured to facilitate wired or wireless communication between the terminal device and other devices. The terminal device may access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In one embodiment, the terminal device may include a SIM card slot for inserting a SIM card therein, so that the terminal device can log on to a GPRS network and establish communication with the server via the internet.

From the above, the communication component 103, the audio component 106, the input/output interface 107 and the sensor component 108 involved in the embodiment of fig. 10 can be implemented as the input device in the embodiment of fig. 9.

In the embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units may be a logical division, and in actual implementation, there may be another division, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, units or modules, and may be in an electrical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.

The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes.

The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.

Claims

1. A conversation model building method, comprising:

generating a segmentation log by using a question-answer sequence, wherein the segmentation log comprises a plurality of question-answer pairs converted from the question-answer sequence;

carrying out duplicate removal and vectorization operation on the text in the segmented log to obtain a vector text;

generating a combined log according to the vector text and the segmentation log;

generating a conversation flow chart according to the merging log;

and executing preset operation through the conversation flow chart.

2. The method of claim 1, wherein before the generating a split log using a question-answer sequence, wherein the split log comprises a plurality of question-answer pairs transformed from the question-answer sequence, the method further comprises: and acquiring the question-answer sequence.

3. The method of claim 1, wherein the performing de-duplication and vectorization operations on the text in the split log to obtain a vector text comprises:

carrying out duplicate removal operation on the repeated dialog text in the segmented log;

and performing text vectorization operation on the dialog text subjected to the deduplication operation to generate the vector text.

4. The method of claim 1, wherein generating a merged log from the vector text and the sliced log comprises:

merging the vector text and the segmentation log to obtain a first merging result;

clustering the first combination result to generate a clustering result;

and generating the merging log according to the first merging result and the clustering result.

5. The method of claim 4, wherein the generating the merged log according to the merged result and the clustering result comprises:

merging the first merging result and the clustering result to obtain a second merging result;

and generating the merging log according to the second merging result.

6. The method of claim 1, wherein generating a session flow graph from the merged log comprises:

acquiring the number of the same sessions in the merged log;

sequencing the number of the same sessions according to a preset rule to obtain a sequencing result;

and generating the conversation flow chart according to the sequencing result.

7. The method of claim 1, wherein the preset operation comprises at least one of: sorting, merging and inserting.

8. The method of claim 7, wherein after the performing the preset operation through the session flow diagram, the method further comprises: and constructing a conversation model according to the conversation flow chart.

9. A logistics customer service dialogue establishing method is applied to a logistics goods-returning customer service end and is characterized by comprising the following steps:

generating a return segmentation log by using a return question-answer sequence, wherein the return segmentation log comprises a plurality of return question-answer pairs converted from the return question-answer sequence;

performing duplicate removal and vectorization operation on the text in the return cut log to obtain a vector text;

generating a combined log according to the vector text and the returned goods segmentation log;

generating a return conversion flow chart according to the combined log;

and executing preset operation through the return conversion flow chart.

10. A dialogue model application method, comprising:

acquiring dialogue model data;

extracting training features according to the dialogue model data;

training a dialogue management model and an intention recognition model according to the training characteristics;

wherein the dialogue model data is created by the dialogue model creation method of any one of claims 1 to 8.

11. The method of claim 10, wherein the obtaining dialogue model data comprises:

acquiring batch conversation models;

and acquiring batch dialogue model data according to the batch dialogue models.

12. The method of claim 10, wherein after training a dialogue management model and an intent recognition model based on the training features, the method further comprises:

sending the trained dialogue management model and the intention recognition model to a dialogue system;

and completing the construction of the dialogue system according to the dialogue management model and the intention recognition model.

13. A satisfaction dialogue model application method is applied to customer service satisfaction feedback and is characterized by comprising the following steps:

obtaining satisfaction dialogue model data;

extracting training characteristics according to the satisfaction degree dialogue model data;

wherein the satisfaction dialogue model data is formed by the dialogue model building method of any one of claims 1-8.

14. A dialogue model creation apparatus, comprising:

the segmentation log module is used for generating a segmentation log by using the question-answer sequence, wherein the segmentation log comprises a plurality of question-answer pairs converted from the question-answer sequence;

the vector text module is used for carrying out duplication removal and vectorization operations on the text in the segmented log to obtain a vector text;

the combined log module is used for generating a combined log according to the vector text and the segmentation log;

the generating module is used for generating a session flow chart according to the merging log;

and the execution module is used for executing preset operation through the conversation flow chart.

15. The apparatus of claim 14, further comprising: and the acquisition module is used for acquiring the question-answer sequence.

16. The apparatus of claim 14, wherein the vector text module comprises:

the duplication removing unit is used for carrying out duplication removing operation on the repeated dialog texts in the segmentation log;

and the vector unit is used for performing text vectorization operation on the dialog text subjected to the deduplication operation to generate the vector text.

17. The apparatus of claim 14, wherein the merge log module comprises:

the first merging unit is used for merging the vector text and the segmentation log to obtain a first merging result;

the clustering unit is used for clustering the first combination result to generate a clustering result;

and the generating unit is used for generating the merging log according to the first merging result and the clustering result.

18. The apparatus of claim 17, wherein the generating unit comprises:

the second merging unit is used for merging the first merging result and the clustering result to obtain a second merging result;

the generating unit is further configured to generate the merging log according to the second merging result.

19. The apparatus of claim 14, wherein the generating module comprises: an obtaining unit, configured to obtain the number of identical sessions in the merged log;

the sorting unit is used for sorting the number of the same sessions according to a preset rule to obtain a sorting result;

and the generating unit is used for generating the conversation flow chart according to the sequencing result.

20. The apparatus of claim 14, wherein the preset operation comprises at least one of: sorting, merging and inserting.

21. The apparatus of claim 20, further comprising:

and the modeling module is used for constructing a conversation model according to the conversation flow chart.

22. A logistics customer service dialogue establishing device is applied to a logistics goods-returning customer service end and is characterized by comprising:

23. A dialogue model application apparatus, comprising:

the acquisition module is used for acquiring dialogue model data;

the extraction module is used for extracting training characteristics according to the dialogue model data;

the dialogue model data is dialogue model data created by the dialogue model creation apparatus according to claims 12 to 19.

24. The apparatus of claim 23, wherein the obtaining module comprises:

the acquisition unit is used for acquiring batch conversation models;

the obtaining unit is further configured to obtain batch dialogue model data according to the batch dialogue models.

25. The apparatus of claim 23, further comprising:

the transmitting module is used for transmitting the trained dialogue management model and the intention recognition model to a dialogue system;

and the construction module is used for finishing the construction of the dialogue system according to the dialogue management model and the intention recognition model.

26. A satisfaction dialogue model application apparatus for customer service satisfaction feedback, comprising:

the acquisition module is used for acquiring satisfaction dialogue model data;

wherein the satisfaction dialogue model data is created by the dialogue model creation apparatus of any one of claims 12 to 19.

27. A computer program product comprising instructions which, when run on a computer, cause the computer to perform the method of any one of claims 1 to 13.

28. A non-volatile storage medium, comprising a stored program, wherein the program, when executed, controls an apparatus in which the non-volatile storage medium is located to perform the method of any of claims 1 to 13.

29. An electronic device comprising a processor and a memory; the memory has stored therein computer readable instructions for execution by the processor, wherein the computer readable instructions when executed perform the method of any of claims 1 to 13.