CN110909166A

CN110909166A - Method, apparatus, medium, and electronic device for improving session quality

Info

Publication number: CN110909166A
Application number: CN201911192314.5A
Authority: CN
Inventors: 杨蕴凯
Original assignee: Beike Technology Co Ltd
Current assignee: Seashell Housing Beijing Technology Co Ltd
Priority date: 2019-11-28
Filing date: 2019-11-28
Publication date: 2020-03-24
Anticipated expiration: 2039-11-28
Also published as: CN110909166B

Abstract

A method, apparatus, medium, and electronic device for improving session quality are disclosed. The method comprises the following steps: obtaining emotion information of a conversation statement of a predetermined conversation party in a plurality of conversations; determining respective quality parameters of the plurality of conversations according to the emotional information of the conversation sentences; determining quality categories to which the plurality of sessions belong according to respective quality parameters of the plurality of sessions; and determining the conversation feature distribution information of a predetermined conversation party according to the emotion information of the conversation sentences of the predetermined conversation party in all the conversations belonging to the same quality class. The technical scheme provided by the disclosure is beneficial to improving the session quality of the predetermined session party, thereby being beneficial to promoting the service to realize the purpose.

Description

Method, apparatus, medium, and electronic device for improving session quality

Technical Field

The present disclosure relates to a conversation technology, and more particularly, to a method for improving conversation quality, an apparatus for improving conversation quality, a storage medium, and an electronic device.

Background

For some services, the session often affects the result of the service, for example, during the session between the customer service and the user, inappropriate terms of the customer service may cause dissatisfaction of the user, which may result in the user ending the session with the customer service in advance, and further may make the service lose the possibility of achieving its purpose.

How to improve the quality of the session so as to make the service achieve the final purpose is a technical problem worthy of attention.

Disclosure of Invention

The present disclosure is proposed to solve the above technical problems. Embodiments of the present disclosure provide a method for improving session quality, an apparatus for improving session quality, a storage medium, and an electronic device.

According to an aspect of the embodiments of the present disclosure, there is provided a method for improving session quality, including: obtaining emotion information of a conversation statement of a predetermined conversation party in a plurality of conversations; determining respective quality parameters of the plurality of conversations according to the emotional information of the conversation sentences; determining quality categories to which the plurality of sessions belong according to respective quality parameters of the plurality of sessions; and determining the conversation feature distribution information of a predetermined conversation party according to the emotion information of the conversation sentences of the predetermined conversation party in all the conversations belonging to the same quality class.

In an embodiment of the present disclosure, the acquiring emotion information of a conversational sentence of a predetermined conversation party in a plurality of conversations includes: providing the conversation sentences of a predetermined conversation party in the plurality of conversations to an emotion polarity prediction model respectively; obtaining the emotion information of each conversation statement according to the information output by the emotion polarity prediction model; wherein, the emotion information of the conversational sentence comprises: at least one of positive, negative, and neutral.

In another embodiment of the present disclosure, the emotion polarity prediction model is obtained by training using a historical conversational sentence, and the training process of the emotion polarity prediction model includes: respectively providing a plurality of historical conversation sentences in the training set to an emotion polarity prediction model; adjusting model parameters of the emotion polarity prediction model according to the emotion polarity prediction result output by the emotion polarity prediction model and the difference of emotion polarity labels of historical conversation sentences in the training set; respectively providing a plurality of historical conversation sentences in the verification set to the emotion polarity prediction model after parameter adjustment; and correcting the emotion polarity labels of the historical conversation sentences in the training set according to the emotion polarity prediction result output by the emotion polarity prediction model after the parameters are adjusted and the difference between the emotion polarity labels of the historical conversation sentences in the verification set, and adjusting the hyper-parameters of the emotion polarity prediction model.

In another embodiment of the present disclosure, the modifying the emotion polarity labels of the historical conversational sentences in the training set according to the difference between the emotion polarity prediction result output by the emotion polarity prediction model after the parameter adjustment and the emotion polarity labels of the historical conversational sentences in the verification set includes: setting a regular expression according to the emotion polarity prediction result output by the emotion polarity prediction model after the parameter adjustment and the difference of the emotion polarity labels of the historical conversation sentences in the verification set; and modifying the emotion polarity labels of the historical conversation sentences in the training set according to the regular expression.

In another embodiment of the present disclosure, the training process of the emotion polarity prediction model further includes: respectively providing a plurality of historical conversation sentences with emotion polarity labels in the test set to the emotion polarity prediction model after parameter adjustment; judging whether the emotion polarity prediction model is successfully trained or not according to the emotion polarity prediction result output by the emotion polarity prediction model after the parameter adjustment and the difference of the emotion polarity labels of the historical conversation sentences in the test set; and if the judgment result is that the training is not completed successfully, continuing to train the emotion polarity prediction model by using a plurality of historical conversation sentences in a training set.

In another embodiment of the present disclosure, the determining the quality parameter of each of the plurality of conversations according to the emotion information of the conversational sentence includes: for any conversation, determining the quality parameter of the conversation according to the number of the conversation sentences with the same emotion information in the conversation and the number of the conversation sentences of the first conversation party in the conversation.

In still another embodiment of the present disclosure, the emotion information includes: positive, negative, and neutral; the determining the quality parameter of the conversation according to the number of the conversation sentences with the same emotion information in the conversation and the number of the conversation sentences of the first conversation party in the conversation includes: determining the number of conversational sentences of which the emotional information is positive, the number of conversational sentences of which the emotional information is negative and the number of conversational sentences of which the emotional information is neutral, and obtaining a first ratio, a second ratio and a third ratio by respectively comparing the ratios of the numbers of the conversational sentences of a first conversational party in the conversation; and predicting the quality parameter of the session according to the first ratio, the second ratio and the third ratio.

In another embodiment of the present disclosure, the predicting the quality parameter of the session according to the first ratio, the second ratio, and the third ratio includes: providing the first ratio, the second ratio and the third ratio to a quality parameter prediction model; and obtaining the quality parameters of the session according to the information output by the quality parameter prediction model.

In another embodiment of the present disclosure, the quality parameter prediction model is obtained by training using a historical conversational sentence, and the training process of the quality parameter prediction model includes: obtaining emotional information of a history conversation statement of a first conversation party in a plurality of history conversations; for any historical conversation, determining the number of historical conversation sentences of which the emotion information is positive, the number of historical conversation sentences of which the emotion information is negative and the number of historical conversation sentences of which the emotion information is neutral in the historical conversation, and respectively obtaining a fourth ratio, a fifth ratio and a sixth ratio from the ratios of the numbers of the historical conversation sentences of a first conversation party in the historical conversation; and determining the model parameters of the quality parameter prediction model according to the fourth ratio, the fifth ratio and the sixth ratio of the plurality of historical conversations and the service achievement state labels of the plurality of historical conversations.

In another embodiment of the present disclosure, the determining, according to the quality parameter of each of the plurality of sessions, a quality class to which each of the plurality of sessions belongs includes: according to a preset quality parameter value interval, taking a session with a quality parameter positioned in the value interval as a first quality session class, and taking a session with a quality parameter positioned outside the value interval as a second quality session class; wherein the session quality of the first quality session class is higher than the session quality of the second quality session class.

In yet another embodiment of the present disclosure, the method further comprises: and determining the quality parameter value intervals according to the data distribution of the quality parameters which are respectively predicted and obtained by the quality parameter prediction model aiming at the plurality of historical conversations and based on the service achievement state labels of the historical conversations.

In still another embodiment of the present disclosure, the determining the conversation feature distribution information of a predetermined conversation party according to emotion information of a conversation sentence of the predetermined conversation party in each conversation belonging to the same quality class includes: determining distribution information of positive conversational sentences, negative conversational sentences and neutral conversational sentences according to the emotional information of the conversational sentences of a preset conversational party in each conversation belonging to the first quality conversational category; and determining the distribution information of the positive direction conversational sentence, the negative direction conversational sentence and the neutral conversational sentence according to the emotional information of the conversational sentence of the preset conversational party in each conversation belonging to the second conversational class.

According to another aspect of the embodiments of the present disclosure, there is provided an apparatus for improving session quality, the apparatus including: the emotion information acquisition module is used for acquiring emotion information of a conversation statement of a preset conversation party in a plurality of conversations; a quality parameter determining module, configured to determine quality parameters of the sessions according to emotion information of the session statements; a quality class determining module, configured to determine, according to respective quality parameters of the multiple sessions, quality classes to which the multiple sessions belong; and the determining characteristic distribution module is used for determining the conversation characteristic distribution information of the predetermined conversation party according to the emotion information of the conversation sentences of the predetermined conversation party in all the conversations belonging to the same quality class.

In an embodiment of the present disclosure, the module for obtaining emotion information is further configured to: providing the conversation sentences of a predetermined conversation party in the plurality of conversations to an emotion polarity prediction model respectively; obtaining the emotion information of each conversation statement according to the information output by the emotion polarity prediction model; wherein, the emotion information of the conversational sentence comprises: at least one of positive, negative, and neutral.

In yet another embodiment of the present disclosure, the apparatus further comprises: a first training module, the first training module comprising: the first submodule is used for respectively providing a plurality of historical conversation sentences in the training set to the emotion polarity prediction model; the second submodule is used for adjusting model parameters of the emotion polarity prediction model according to the emotion polarity prediction result output by the emotion polarity prediction model and the difference of the emotion polarity labels of historical conversation sentences in the training set; the third sub-module is used for respectively providing a plurality of historical conversation sentences in the verification set to the emotion polarity prediction model after parameter adjustment; and the fourth sub-module is used for correcting the emotion polarity labels of the historical conversation sentences in the training set according to the emotion polarity prediction result output by the emotion polarity prediction model after the parameter adjustment and the difference of the emotion polarity labels of the historical conversation sentences in the verification set, and adjusting the hyper-parameters of the emotion polarity prediction model.

In yet another embodiment of the present disclosure, the fourth sub-module is further configured to: setting a regular expression according to the emotion polarity prediction result output by the emotion polarity prediction model after the parameter adjustment and the difference of the emotion polarity labels of the historical conversation sentences in the verification set; and modifying the emotion polarity labels of the historical conversation sentences in the training set according to the regular expression.

In yet another embodiment of the present disclosure, the first training module further comprises: the fifth sub-module is used for respectively providing a plurality of historical conversation sentences with emotion polarity labels in the test set to the emotion polarity prediction model after parameter adjustment; a sixth sub-module, configured to determine whether the emotion polarity prediction model is successfully trained according to an emotion polarity prediction result output by the emotion polarity prediction model after parameter adjustment and a difference between emotion polarity labels of historical conversation sentences in the test set; and if the judgment result is that the training is not completed successfully, triggering a first training module to continue to train the emotion polarity prediction model by using a plurality of historical conversation sentences in a training set.

In another embodiment of the present disclosure, the module for determining quality parameters includes: and the seventh submodule is used for determining the quality parameter of any conversation according to the number of the conversation sentences with the same emotion information in the conversation and the number of the conversation sentences of the first conversation party in the conversation.

In still another embodiment of the present disclosure, the emotion information includes: positive, negative, and neutral; the seventh sub-module includes: a first unit, configured to determine the number of conversational sentences of which the emotion information is positive, the number of conversational sentences of which the emotion information is negative, and the number of conversational sentences of which the emotion information is neutral, and obtain a first ratio, a second ratio, and a third ratio, by using ratios of the numbers of conversational sentences of a first conversational party in the conversation, respectively; and the second unit is used for predicting the quality parameter of the session according to the first ratio, the second ratio and the third ratio.

In yet another embodiment of the present disclosure, the second unit is further configured to: providing the first ratio, the second ratio and the third ratio to a quality parameter prediction model; and obtaining the quality parameters of the session according to the information output by the quality parameter prediction model.

In yet another embodiment of the present disclosure, the apparatus further includes: a second training module to: obtaining emotional information of a history conversation statement of a first conversation party in a plurality of history conversations; for any historical conversation, determining the number of historical conversation sentences of which the emotion information is positive, the number of historical conversation sentences of which the emotion information is negative and the number of historical conversation sentences of which the emotion information is neutral in the historical conversation, and respectively obtaining a fourth ratio, a fifth ratio and a sixth ratio from the ratios of the numbers of the historical conversation sentences of a first conversation party in the historical conversation; and determining the model parameters of the quality parameter prediction model according to the fourth ratio, the fifth ratio and the sixth ratio of the plurality of historical conversations and the service achievement state labels of the plurality of historical conversations.

In yet another embodiment of the present disclosure, the determine quality category module is further configured to: according to a preset quality parameter value interval, taking a session with a quality parameter positioned in the value interval as a first quality session class, and taking a session with a quality parameter positioned outside the value interval as a second quality session class; wherein the session quality of the first quality session class is higher than the session quality of the second quality session class.

In yet another embodiment of the present disclosure, the apparatus further includes: and the quality parameter value interval determining module is used for determining the quality parameter value interval according to the data distribution of the service achievement state labels of the historical sessions, wherein the quality parameters are respectively predicted and obtained by the quality parameter prediction model aiming at the historical sessions.

In yet another embodiment of the present disclosure, the determining the feature distribution module is further configured to: determining distribution information of positive conversational sentences, negative conversational sentences and neutral conversational sentences according to the emotional information of the conversational sentences of a preset conversational party in each conversation belonging to the first quality conversational category; and determining the distribution information of the positive direction conversational sentence, the negative direction conversational sentence and the neutral conversational sentence according to the emotional information of the conversational sentence of the preset conversational party in each conversation belonging to the second conversational class.

According to still another aspect of the embodiments of the present disclosure, there is provided a computer-readable storage medium storing a computer program for executing the above-mentioned method for improving session quality.

According to still another aspect of the embodiments of the present disclosure, there is provided an electronic apparatus including: a processor; a memory for storing the processor-executable instructions; the processor is used for reading the executable instructions from the memory and executing the instructions to realize the method for improving the session quality.

Based on the method and the device for improving the session quality provided by the embodiment of the disclosure, the quality categories to which the sessions belong are determined by using the emotion information of the session sentences, and the session feature distribution information of the predetermined session party is determined by using the emotion information of the session sentences of the predetermined session party in the sessions belonging to the same quality category, so that the session mode and the session features of the predetermined session party in the sessions can be objectively shown, and the predetermined session party has global overall cognition on the session mode, the session features and the like of the predetermined session party. Therefore, the technical scheme provided by the disclosure is beneficial to improving the session quality of the predetermined session party, thereby being beneficial to promoting the service to achieve the purpose.

The technical solution of the present disclosure is further described in detail by the accompanying drawings and examples.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the description, serve to explain the principles of the disclosure.

The present disclosure may be more clearly understood from the following detailed description, taken with reference to the accompanying drawings, in which:

FIG. 1 is a schematic diagram of one embodiment of a suitable scenario for use with the present disclosure;

FIG. 2 is a flow diagram of one embodiment of a method of the present disclosure for improving session quality;

FIG. 3 is a flow diagram of one embodiment of the present disclosure for training an emotion polarity prediction model using historical conversational utterances;

FIG. 4 is a flow diagram of one embodiment of training a quality parameter prediction model using historical conversational utterances according to the present disclosure;

FIG. 5 is a schematic diagram of one embodiment of a quality parameter profile obtained based on a session verification set according to the present disclosure;

FIG. 6 is a schematic diagram of one embodiment of a quality parameter profile obtained based on a session test set according to the present disclosure;

FIG. 7 is a schematic diagram of one embodiment of a quality parameter profile obtained based on a session training set according to the present disclosure;

FIG. 8 is a schematic block diagram illustrating an embodiment of an apparatus for improving session quality according to the present disclosure;

fig. 9 is a block diagram of an electronic device provided in an exemplary embodiment of the present disclosure.

Detailed Description

Example embodiments according to the present disclosure will be described in detail below with reference to the accompanying drawings. It is to be understood that the described embodiments are merely a subset of the embodiments of the present disclosure and not all embodiments of the present disclosure, with the understanding that the present disclosure is not limited to the example embodiments described herein.

It should be noted that: the relative arrangement of the components and steps, the numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.

It will be understood by those of skill in the art that the terms "first," "second," and the like in the embodiments of the present disclosure are used merely to distinguish one element from another, and are not intended to imply any particular technical meaning, nor is the necessary logical order between them. It is also understood that in embodiments of the present disclosure, "a plurality" may refer to two or more than two and "at least one" may refer to one, two or more than two.

It is also to be understood that any reference to any component, data, or structure in the embodiments of the disclosure, may be generally understood as one or more, unless explicitly defined otherwise or stated otherwise.

In addition, the term "and/or" in the present disclosure is only one kind of association relationship describing the associated object, and means that there may be three kinds of relationships, such as a and/or B, and may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" in the present disclosure generally indicates that the former and latter associated objects are in an "or" relationship. It should also be understood that the description of the various embodiments of the present disclosure emphasizes the differences between the various embodiments, and the same or similar parts may be referred to each other, so that the descriptions thereof are omitted for brevity.

Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description. The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application, or uses.

Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate.

It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.

Embodiments of the present disclosure may be implemented in electronic devices such as terminal devices, computer systems, servers, etc., which are operational with numerous other general purpose or special purpose computing system environments or configurations. Examples of well known terminal devices, computing systems, environments, and/or configurations that may be suitable for use with an electronic device, such as a terminal device, computer system, or server, include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, hand-held or laptop devices, microprocessor-based systems, set top boxes, programmable consumer electronics, network pcs, minicomputer systems, mainframe computer systems, distributed cloud computing environments that include any of the above, and the like.

Electronic devices such as terminal devices, computer systems, servers, etc. may be described in the general context of computer system-executable instructions, such as program modules, being executed by a computer system. Generally, program modules may include routines, programs, objects, components, logic, data structures, etc. that perform particular tasks or implement particular abstract data types. The computer system/server may be implemented in a distributed cloud computing environment. In a distributed cloud computing environment, tasks may be performed by remote processing devices that are linked through a communications network. In a distributed cloud computing environment, program modules may be located in both local and remote computer system storage media including memory storage devices.

Summary of the disclosure

In carrying out the present disclosure, the inventors discovered that for some businesses, a conversation of a worker with a user is often used for the purpose of implementing the business. For example, for a property rental service, a session between a property broker and a user is often used to prompt the user to delegate the property broker to help it complete the property rental program.

At present, the session quality is usually improved by using a standardized session phrase and the like, so as to make the service achieve the final purpose as much as possible. However, in the conversation process with the user, the conversation sentences used by the user are often diversified, and the standardized conversation phrases are not favorable for flexibly coping with the conversation sentences in various forms of the user, so that the business is not favorable for achieving the final purpose, and the conversation experience of the user is not favorable for improving.

Brief description of the drawings

One example of an application scenario of the techniques for improving session quality provided by the present disclosure is shown in fig. 1.

In fig. 1, an IM (instant messaging) tool (such as a WeChat or QQ) is installed in each of the smart phone 100 of the user 101, the smart phone 102-1 of the user 103-1, the smart phones 102-2 and … … of the user 103-2, and the smart phone 102-n of the user 103-n, and the user 101 can perform a conversation with the user 103-1, the users 103-2 and … …, and the user 103-n by using the IM tool.

Assume that user 101 is a house broker. User 103-1, users 103-2, … …, and user 103-n are all users who have a rental need. User 103-1, users 103-2, … …, and user 103-n each use the IM tool to enter text or input speech, and each send conversational sentences to user 101 to express their house-renting needs. Similarly, user 101 enters text or speech using the IM tool to form conversational utterances for replying to user 103-1, users 103-2, … …, and user 103-n, respectively.

According to the method and the device, the emotional information of each conversation sentence of the user 101 in each conversation is determined in the process that the user 101 carries out conversation with the user 103-1, the users 103-2, … … and the user 103-n respectively, and the conversation feature distribution information of the user 101 is obtained based on the emotional information of each conversation sentence, so that the conversation features of the user 101 can be reflected more objectively. The user 101 can improve the future session according to the session characteristics thereof, so as to improve the session quality thereof, and further facilitate to improve the delegation rate or transfer delegation rate thereof, and enable the user who has a session with the user to obtain better session experience.

Exemplary method

Fig. 2 is a flowchart illustrating an embodiment of a method for improving session quality according to the present disclosure.

As shown in fig. 2, the method of this embodiment includes the steps of: s200, S201, S202, and S203. The following describes each step.

S200, obtaining emotion information of a conversation sentence of a predetermined conversation party in a plurality of conversations.

A conversation in this disclosure may refer to a conversation between at least two parties to the conversation. For example, a conversation formed by two parties or more parties using an IM facility. All conversation parties of each conversation in the plurality of conversations of the present disclosure include a predetermined conversation party, for example, all conversation parties of each conversation include a worker numbered with a predetermined value, and the like. The present disclosure is not limited to other conversation parties for each of the plurality of conversations. For example, a plurality of sessions in the present disclosure may be regarded as all sessions formed by a predetermined conversation party performing information interaction with a plurality of users, respectively.

One session in the present disclosure would typically include: at least one conversational sentence of a predetermined conversational party and at least one conversational sentence of other conversational parties with whom the predetermined conversational party is interacting in information. The party subscribing to the conversation may refer to a party wishing to improve the quality of the conversation. For example, the intended conversant may be a particular house broker, etc.

The conversational sentence in the disclosure may be a conversational sentence formed by a conversational party based on text interaction, or a conversational sentence formed by a conversational party based on voice interaction. The conversational sentence in the present disclosure is typically a textual conversational sentence. For audio-formatted conversational utterances, the present disclosure may utilize speech recognition techniques to convert the audio-formatted conversational utterances to textual-formatted conversational utterances. The emotion information of the conversational sentence in the present disclosure may refer to an emotional color carried by the conversational sentence. For example, the conversation sentence shows a positive emotion or a negative emotion, or the like.

S201, determining quality parameters of a plurality of conversations according to emotion information of conversation sentences.

The quality parameter of a conversation in the present disclosure is determined by using the emotion information of all conversation sentences of a predetermined conversation party in the conversation. The quality parameter of a session in the present disclosure may refer to a parameter for indicating whether the session is good or bad. The present disclosure may obtain the quality parameter of the session in a variety of ways. For example, the present disclosure may count emotion information of a conversational sentence in one conversation, and determine a quality parameter of the conversation according to the statistical result.

S202, determining the quality categories of the multiple sessions according to the quality parameters of the multiple sessions.

A quality class in the present disclosure may refer to at least two different classifications formed by classifying the quality of a session. The quality class may also be referred to as a quality class or quality level, etc. Since the quality parameter of the session may represent the quality of the session, the present disclosure may determine the quality class to which the session belongs using the quality parameter of the session. For example, the quality category to which the session belongs can be determined by judging the value interval to which the quality parameter of the session belongs.

S203, determining conversation characteristic distribution information of the predetermined conversation party according to the emotion information of the conversation sentences of the predetermined conversation party in the conversations belonging to the same quality class.

The session feature distribution information in the present disclosure may indicate a session manner of a predetermined session party, session characteristics, and the like. Such as the conversation mode and the conversation characteristics of the predetermined conversation parties in different quality categories.

The method determines the quality types of the multiple conversations to which the multiple conversations belong by utilizing the emotion information of the conversation sentences of the preset conversation parties in the multiple conversations, and determines the conversation characteristic distribution information of the preset conversation parties by utilizing the emotion information of the conversation sentences of the preset conversation parties in the conversations belonging to the same quality type, so that the conversation mode and the conversation characteristics of the preset conversation parties in the multiple conversations can be objectively shown, and the overall cognition of the preset conversation parties on the conversation mode, the characteristics and the like of the preset conversation parties is facilitated. Therefore, the technical scheme provided by the disclosure is beneficial to improving the session quality of the predetermined session party, thereby being beneficial to promoting the service to achieve the purpose.

In one alternative example, the emotion information of the conversational sentence in the present disclosure may be the emotion polarity of the conversational sentence. The emotional polarity may include: positive, negative, and neutral. The positive direction indicates that the emotional color of the conversational sentence is positive upward. Negative going indicates that the emotional color of the conversational sentence is negative. Neutral means that the conversational sentence does not have a positive emotional color, nor a negative emotional color. For example, "you can consider the house, close to the subway entrance, and the traffic is convenient", and the emotional polarity of the conversation sentence is positive. As another example, "budget 250 ten thousand" or so, the emotional polarity of this conversational sentence is neutral. As another example, "that counts bar", the emotion polarity of the conversation sentence is negative.

Alternatively, in the case where the emotion information is expressed using positive, negative, or neutral, the nature of determining the emotion information of a conversational sentence in the present disclosure is a three-classification process of a conversational sentence. The emotion information of the conversation sentences of the predetermined conversation party in the plurality of conversations can be obtained by utilizing the emotion polarity prediction model successfully trained in advance. Specifically, the present disclosure may provide conversational sentences of a predetermined conversational partner in a plurality of conversations to the emotion polarity prediction model, respectively, and obtain emotion information of each conversational sentence according to information output by the emotion polarity prediction model for each conversational sentence, respectively. The information output by the emotion polarity prediction model for a conversational sentence may include three probability values, namely, a probability value that the emotion polarity of the input conversational sentence is positive, a probability value that the emotion polarity is negative, and a probability value that the emotion polarity is neutral. The method and the device can select a maximum probability value from three probability values output by an emotion polarity prediction model aiming at an input conversational sentence, and use the emotion polarity corresponding to the maximum probability value as the emotion polarity of the input conversational sentence.

According to the method and the device, three classification processing is performed on the conversation sentences by using the emotion polarity prediction model, so that the emotion polarities of all the conversation sentences can be conveniently and accurately obtained.

In an alternative example, the emotion polarity prediction model in the present disclosure is obtained by training using historical conversational sentences. One process of training an emotion polarity prediction model using historical conversational sentences is shown in FIG. 3. The process shown in FIG. 3 includes the following steps:

and S300, training the emotion polarity prediction model.

S301, providing the plurality of historical conversation sentences in the training set to the emotion polarity prediction model respectively.

Alternatively, the historical conversational utterances in the training set of the present disclosure are generally independent of the predetermined conversational partner described above. A plurality of historical conversational sentences in the training set each have an emotional polarity label. According to the method, the history conversation sentences can be labeled by utilizing the open-source Chinese emotion analysis platform, so that the emotion polarity labels of the history conversation sentences are obtained. For the property field, the method can prepare historical conversational sentences of a plurality of property brokers and obtain the emotion polarity labels of each historical conversational sentence by using a Chinese emotion analysis platform.

Alternatively, the historical conversational sentences in the training set may be provided to the emotion polarity prediction model after being represented using the word vector. The dimensionality of the word vector is typically determined by the hyper-parameters of the emotion polarity prediction model.

Optionally, the emotion polarity labels of the historical conversational sentences in the training set may be inaccurate, for example, because the chinese emotion analysis platform usually has a certain degree of lack in the accuracy of labeling the conversational sentences in a specific field (e.g., a real estate field, etc.), the accuracy of the emotion polarity labels of the historical conversational sentences in the training set has a certain degree of lack. The method can be used for correcting the emotion polarity labels of the historical conversation sentences in the training set by utilizing the subsequent steps. The emotion polarity label before correction can be called a pre-labeled emotion polarity label or an emotion polarity initial label and the like.

Alternatively, the emotion polarity prediction model in the present disclosure may be a model based on the FastText classification algorithm. The emotion polarity prediction model may also be referred to as a fast text classifier or the like. The number of the historical conversational sentences provided to the emotion polarity prediction model at a time by the present disclosure may be determined according to a preset batch number. For example, one tenth of the historical conversational utterances is retrieved from the training set at a time. The history conversational sentences once supplied to the emotion polarity prediction model are respectively input as models, sequentially supplied to the emotion polarity prediction model, subjected to emotion polarity prediction processing for each input history conversational sentence through the emotion polarity prediction model, and output emotion polarity prediction results for each history conversational sentence.

S302, adjusting model parameters of the emotion polarity prediction model according to the emotion polarity prediction result output by the emotion polarity prediction model and the difference of emotion polarity labels of historical conversation sentences in a training set.

Optionally, the loss is calculated by using a loss function based on cross entropy (i.e., cross entropy is calculated) according to the difference between the emotion polarity prediction result output by the emotion polarity prediction model and the emotion polarity label of the corresponding historical conversational sentence in the training set, and the loss is propagated in the emotion polarity prediction model in a reverse direction, so that the model parameters of the emotion polarity prediction model are adjusted. The model parameters in this disclosure generally refer to: the hidden layer of the model is used to convert the input sparse vectors into a first matrix of dense vectors having a first predetermined dimension, and the output layer is used to convert the hidden output dense vectors into a second matrix of vectors having a second predetermined dimension. The second predetermined dimension is related to the number of classifications that the present disclosure classifies emotional polarity on the historical conversational utterances. For example, where the emotional polarity includes positive, negative, and neutral, the second predetermined dimension is 3. The model parameters in this disclosure do not include the hyper-parameters of the emotion polarity prediction model.

And S303, respectively providing the plurality of historical conversation sentences in the verification set to the emotion polarity prediction model after parameter adjustment.

Optionally, the historical conversational sentences in the verification set of the present disclosure are generally independent of the predetermined conversational party, and the historical conversational sentences in the verification set may be partially the same as the historical conversational sentences in the training set or completely different from the historical conversational sentences in the training set. And each historical conversation statement in the verification set has an emotional polarity label. The emotion polarity labels of the historical conversation sentences in the verification set in the present disclosure may be labels set in a manual labeling manner. For example, each historical conversational sentence in the verification set is provided to the public standard platform, so that the emotion polarity labels of each historical conversational sentence can be obtained from the public standard platform. The sentiment polarity tags of the historical conversational utterances in the verification set of the present disclosure may be considered accurate tags. The number of historical conversational utterances contained in the validation set is often less than the number of historical conversational utterances contained in the training set. For example, the verification set contains 3000 historical conversational utterances, while the training set contains 30000 historical conversational utterances.

Optionally, the number of the historical conversation sentences acquired from the verification set at a time in the present disclosure may be determined according to a preset batch processing number. For example, one fifth of the historical session statements are obtained from the authentication set at a time. All historical conversation sentences acquired from the verification set at a time are used as input of the emotion polarity prediction model after parameter adjustment and are sequentially provided for the emotion polarity prediction model after parameter adjustment, emotion polarity prediction processing is respectively carried out on each input historical conversation sentence through the emotion polarity prediction model after parameter adjustment, and emotion polarity prediction results are respectively output for each historical conversation sentence.

S304, according to the emotion polarity prediction result output by the emotion polarity prediction model after parameter adjustment and the difference of emotion polarity labels of historical conversation sentences in the verification set, the emotion polarity labels of the historical conversation sentences in the training set are corrected, and the hyper-parameters of the emotion polarity prediction model after parameter adjustment are adjusted.

Optionally, since the emotion polarity tag of the historical conversational sentence in the verification set of the present disclosure is considered to be an accurate tag, if the emotion polarity prediction result output by the parameter-adjusted emotion polarity prediction model for a historical conversational sentence is different from the emotion polarity tag of the historical conversational sentence in the verification set, the emotion polarity prediction result of the current parameter-adjusted emotion polarity prediction model may be considered to be incorrect. Reasons for this inaccuracy may include: and the emotion polarity labels of the historical conversation sentences in the training set are incorrect, so that the model parameters of the emotion polarity prediction model are not adjusted to proper values.

Optionally, the historical dialogue sentences with wrong emotion polarity labels in the training set can be determined according to the emotion polarity prediction result output by the emotion polarity prediction model after parameter adjustment and the difference between the emotion polarity labels of the historical dialogue sentences in the verification set, and the emotion polarity labels of the historical dialogue sentences are corrected.

Optionally, the present disclosure may determine a history session statement set according to a difference between an emotion polarity prediction result output by the emotion polarity prediction model after parameter adjustment and an emotion polarity tag of a corresponding history session statement in the verification set. For convenience of description, the historical session set will be referred to as a badcase set below. The history session statements contained in the badcase set are: and verifying the concentrated historical conversation sentences which are predicted to be wrong by the emotion polarity prediction model. According to the method and the device, the corresponding regular expression can be set according to the historical conversation sentences and the emotion polarity labels thereof in the badcase set, and the emotion polarity labels of the corresponding historical conversation sentences in the training set are corrected by utilizing the regular expression, so that the emotion polarity labels of the historical conversation sentences in the training set are gradually corrected. Specifically, the regular expression can be utilized to traverse all the historical conversational sentences in the training set, so that the emotion polarity labels of the historical conversational sentences matched with the regular expression in the training set are corrected. In addition, the number of history session statements in the badcase set is usually gradually reduced as the number of iterations is increased.

Optionally, the regular expression in the present disclosure may be a regular expression for a sentence, or may be a regular expression for a keyword. In addition, the regular expressions provided by the present disclosure may be one or more. The present disclosure does not limit the concrete manifestation of the regular expression.

Optionally, the super-parameters of the current emotion polarity prediction model can be adjusted according to the emotion polarity prediction result output by the emotion polarity prediction model after parameter adjustment and the difference between the emotion polarity labels of the corresponding historical conversation sentences in the verification set. The hyper-parameters of the emotion polarity prediction model of the present disclosure may include, but are not limited to: the number of times of training the emotion polarity prediction model by using the historical conversation sentences in the training set, the N-Gram (N-Gram) adopted by the emotion polarity prediction model, the initialized dimension of the vector output by the hidden layer of the emotion polarity prediction model and the like. In the one-cycle process shown in fig. 3, only one of the hyper-parameters is usually adjusted, and after a plurality of cycles, if the parameter is considered to be unnecessary to be adjusted, another one of the hyper-parameters is adjusted in the subsequent cycle. And so on until all the parameters in the super-parameters are considered to be unnecessary to be adjusted.

Optionally, an example of S304 adjusting the hyper-parameter of the current emotion polarity prediction model may be: judging whether the difference between the emotion polarity prediction result output by the emotion polarity prediction model after parameter adjustment and the emotion polarity labels of the historical conversation sentences in the verification set is approximately the same as the difference obtained in S304 in the previous cycle process (for example, the difference between the current difference and the previous difference belongs to a predetermined range), if so, determining that the ith parameter adjusted in the previous cycle process does not need to be adjusted again, and adjusting the value of the (i + 1) th parameter in the hyper-parameters at this time. For example, the value of the (i + 1) th parameter may be set to a smaller value. If not, it is assumed that the ith parameter adjusted during the previous cycle still needs to be adjusted. For example, in the previous cycle, when the setting of the ith parameter is small, the ith parameter may be increased by a little.

The emotion polarity labels of the historical conversation sentences in the training set are corrected by using the verification set, so that the phenomenon that each historical conversation sentence in the training set must be accurately marked in advance is avoided, and the training set is obtained quickly; since the number of historical conversational sentences in the verification set is usually much smaller than the number of historical conversational sentences in the training set, the present disclosure is advantageous for reducing the workload of manually labeling the historical conversational sentences. Finally, the method is beneficial to improving the training efficiency of the emotion polarity prediction model.

S305, respectively providing a plurality of historical conversation sentences with emotion polarity labels in the test set to the emotion polarity prediction model after parameter adjustment.

Optionally, the historical conversational utterances in the test set of the present disclosure are generally unrelated to the predetermined conversational party, and preferably, the historical conversational utterances in the test set are completely different from the historical conversational utterances in the verification set. Each historical conversational sentence in the test set has an emotional polarity tag. The emotion polarity labels of the historical conversation sentences in the test set in the disclosure may be labels set in a manual labeling manner. For example, each historical conversational sentence in the test set is provided to the public standard platform, so that the emotion polarity label of each historical conversational sentence can be obtained from the public standard platform. The sentiment polarity tags of the historical conversational utterances in the test set of the present disclosure may be considered accurate tags. The number of the historical conversational sentences contained in the test set is often smaller than the number of the historical conversational sentences contained in the training set. For example, the test set contains 3000 historical conversational utterances, while the training set contains 30000 historical conversational utterances.

Optionally, the number of the historical conversation sentences acquired from the test set at a time in the present disclosure may be determined according to a preset batch processing number. For example, one fifth of the historical conversational utterances are retrieved from the test set at a time. All historical conversation sentences acquired from the test set at a time are used as input of the emotion polarity prediction model after parameter adjustment and are sequentially provided for the emotion polarity prediction model after parameter adjustment, emotion polarity prediction processing is respectively carried out on each input historical conversation sentence through the emotion polarity prediction model after parameter adjustment, and emotion polarity prediction results are respectively output for each historical conversation sentence.

S306, determining the prediction accuracy of the current emotion polarity prediction model according to the emotion polarity prediction result output by the emotion polarity prediction model after parameter adjustment and the difference of the emotion polarity labels of the historical conversation sentences in the test set.

Optionally, because the emotion polarity tag of the history session statement in the test set of the present disclosure is considered to be an accurate tag, if the emotion polarity prediction result output by the parameter-adjusted emotion polarity prediction model for a history session statement is different from the emotion polarity tag of the history session statement in the test set, the emotion polarity prediction result of the current parameter-adjusted emotion polarity prediction model may be considered to be incorrect. Reasons for this inaccuracy may include: and the emotion polarity labels of the historical conversation sentences in the training set are incorrect, so that the model parameters of the emotion polarity prediction model are not adjusted to proper values.

Optionally, the number of history session sentences with wrong prediction by the emotion polarity prediction model after parameter adjustment is counted according to the emotion polarity prediction result output by the emotion polarity prediction model after parameter adjustment and the difference between emotion polarity labels of history session sentences in a test set, and the prediction accuracy of the current emotion polarity prediction model is determined based on the number. For example, the ratio of the counted number to the number of the historical conversation sentences acquired from the test set at this time is used as the prediction accuracy of the current emotion polarity prediction model.

S307, judging whether the prediction accuracy of the current emotion polarity prediction model reaches a preset accuracy, if so, determining that the emotion polarity prediction model is successfully trained, and going to S308; if the preset accuracy rate is not reached, the emotion polarity prediction model is determined to be unsuccessfully trained and the process returns to S301.

Optionally, the predetermined accuracy in the present disclosure is a preset value, such as 90%. And under the condition that the preset accuracy rate is not reached, ending the iterative training process and starting the next iterative training process. And if the iterative training times reach the preset iterative times, however, the prediction accuracy of the current emotion polarity prediction model still does not reach the preset accuracy, finishing the training of the emotion polarity prediction model.

According to the method, whether the emotion polarity prediction model is successfully trained or not is judged by utilizing the test set, so that the phenomenon of overfitting of the emotion polarity prediction model in the training process is avoided, and the prediction accuracy of the emotion polarity prediction model is improved.

And S308, finishing the training of the emotion polarity prediction model.

In an alternative example, the present disclosure may determine the quality parameter of the conversation by using the emotion information of the conversation sentence: for any conversation, determining the quality parameter of the conversation according to the number of the conversation sentences with the same emotional information in the conversation and the number of the conversation sentences of the first conversation party in the conversation. For example, the present disclosure may determine the quality parameter of the conversation according to the ratio of the number of conversational utterances with the same emotion information in the conversation to the number of conversational utterances of the first conversation party in the conversation. The first party in the present disclosure S307 may refer to a plurality of different predetermined parties. The first party may typically be a staff member of the party providing the service, e.g. for a house renting and selling service, the first party is typically a house broker. The distinction of the predetermined conversation party from the first conversation party in the present disclosure includes: the predetermined party is usually specific to an individual, and the first party is not specific to an individual but usually corresponds to a group. For example, the first party is not typically specific to a particular house broker, but rather to a group of house brokers, and the intended party is typically specific to a particular house broker.

In one alternative example, it is assumed that the emotion information in the present disclosure includes: positive, negative and neutral. In this assumed case, for a conversation, the present disclosure may first count the number of conversational sentences whose emotion information is positive, the number of conversational sentences whose emotion information is negative, and the number of conversational sentences whose emotion information is neutral in the conversation, and may also count the number of conversational sentences of a first conversational party (e.g., a property broker) in the conversation, so that the present disclosure obtains the first number, the second number, the third number, and the fourth number. Then, the ratio of the first number to the fourth number, the ratio of the second number to the fourth number, and the ratio of the third number to the fourth number are calculated, respectively, so as to obtain a first ratio, a second ratio, and a third ratio. Then, the present disclosure may predict the quality parameter of the session according to the first ratio, the second ratio, and the third ratio. For example, the present disclosure may perform corresponding calculations for the first ratio, the second ratio, and the third ratio, and use the calculated results as the quality parameter of the session.

In an alternative example, the present disclosure may obtain the quality parameter of the session using a preset quality parameter prediction model. For example, the present disclosure may provide the first ratio, the second ratio, and the third ratio obtained as described above to a quality parameter prediction model, where the quality parameter prediction model calculates the three input ratios, and the quality parameter prediction model outputs the calculation result, so that the present disclosure obtains the quality parameter of the session. The calculation output by the quality parameter prediction model is typically a value within a predetermined range of values, for example, a value within a range of values 0 and 1. The method and the device provide an implementation mode for obtaining the quality parameters of the session by utilizing the quality parameter prediction model, and are favorable for objectively and accurately obtaining the quality parameters of the session.

Optionally, the quality parameter prediction model of the present disclosure may be implemented by using an LR (Logistic Regression) model. The logistic regression model of the present disclosure may refer to a linear regression model normalized by Sigmoid function. That is, the logistic regression model can be regarded as applying a logistic function based on linear regression. The quality parameter prediction model of the present disclosure may be obtained using historical conversational sentence training. Training the logistic regression model using the historical conversational utterances may be considered as determining coefficients in the logistic regression algorithm using the historical conversational utterances. One process of the present disclosure for training a quality parameter prediction model using historical conversational utterances is illustrated in fig. 4. The process shown in FIG. 4 includes the following steps:

s400, obtaining the emotional information of the history conversation sentences of the first conversation party in the plurality of history conversations.

Alternatively, the historical sessions in S400 may be historical sessions in a session training set. That is, the session training set includes a plurality of historical sessions, each historical session includes a plurality of historical session statements, for example, each historical session includes: a plurality of historical conversational utterances of a user and a plurality of historical conversational utterances of a first party. The first party in S400 may refer to a plurality of different predetermined parties and may typically be a staff member of the party providing the service, e.g. for a house renting and selling service, the first party is typically a house broker. The difference between the predetermined conversation party and the first conversation party in the present disclosure is as described in the above embodiments, and the description is not repeated here.

Optionally, the emotion information of the history conversation sentence of the first conversation party in the history conversation in the conversation training set in the present disclosure may be obtained by using the emotion polarity prediction model successfully trained.

S401, for any history conversation, determining the number of history conversation sentences of which the emotion information is positive, the number of history conversation sentences of which the emotion information is negative and the number of history conversation sentences of which the emotion information is neutral in the history conversation, and obtaining a fourth ratio, a fifth ratio and a sixth ratio by respectively comparing the ratios of the numbers of history conversation sentences of the first conversation party in the history conversation.

Alternatively, it is assumed that a history session includes 10 history session statements, i.e., the 1 st history session statement, the 2 nd history session statement, … … and the 10 th history session statement. The 4 historical conversational sentences are the historical conversational sentences of the user, and are respectively the 1 st, the 3 rd, the 6 th and the 9 th. The 6 historical conversational sentences are the historical conversational sentences of the first conversational party, and are respectively the 2 nd, 4 th, 5 th, 7 th, 8 th and 10 th historical conversational sentences. If the emotional information of the 2 nd, 4 th, 5 th and 10 th historical conversational sentences is positive, and the emotional information of the 7 th and 8 th historical conversational sentences is neutral, the fourth ratio obtained by calculation in the disclosure is 2/3, the fifth ratio is 0, and the sixth ratio is 1/3.

S402, determining model parameters of the quality parameter prediction model according to the fourth ratio, the fifth ratio and the sixth ratio of the plurality of historical conversations and the service achievement state labels of the plurality of historical conversations.

Alternatively, the service achievement status label of the historical session in the present disclosure may be determined according to the historical session statement included in the historical session. For example, the traffic achievement status label of the history session may be determined by searching all history session statements of the user included in the history session for predetermined content. The predetermined content can be determined according to the actual condition of the service. For example, for a house renting and selling service in the house property field, the predetermined content may be a user contact (e.g., a user's phone number) or the like.

Optionally, if predetermined content is searched in all history session sentences of a user included in a history session (for example, a user contact address is searched), the service achievement state label of the history session may be set as a label (for example, 1) for representing successful achievement of the service; if the predetermined content is not searched in all the history session sentences of the user included in a history session, the service achievement status label of the history session may be set to a label (e.g., 0) for representing that the service is not successfully achieved.

Optionally, in the present disclosure, the fourth ratio, the fifth ratio, the sixth ratio, and the service achievement state labels of the multiple historical sessions, which correspond to the multiple historical sessions, may be respectively substituted into the quality parameter prediction model, so as to obtain multiple equations (e.g., multiple logistic regression equations), and by solving the multiple equations, values of each parameter in the logistic regression equations may be obtained, where the values of each parameter in the equations are the model parameters of the quality parameter prediction model. After the values of all the parameters in the equation are determined, the quality parameter prediction model is successfully trained. The present disclosure may predict the quality parameters of the conversational sentence in S200 using a successfully trained quality parameter prediction model.

Optionally, the quality parameter prediction model of the present disclosure associates the emotion information of the historical conversational sentence with the service achievement status label, that is, the present disclosure establishes a relationship between the emotion information of the historical conversational sentence and the service achievement status label. The relation established by the quality parameter prediction model can be verified by utilizing the session verification set and the session test set, and whether the relation exists between the emotion information of the historical session statement and the service achievement state label or not is determined according to the verification result. One specific validation process is as follows:

first, the present disclosure may obtain a fourth ratio, a fifth ratio, a sixth ratio, and service achievement state labels of each of a plurality of historical sessions from a session verification set, and provide the fourth ratio, the fifth ratio, and the sixth ratio of each of the historical sessions to a successfully trained quality parameter prediction model, so as to obtain a quality parameter of each of the historical sessions. The present disclosure may form a quality parameter profile for both types of quality parameters. An example of the quality parameter profile is shown in fig. 5. In fig. 5, the abscissa represents the quality parameter predicted by the quality parameter prediction model, and the ordinate represents the number of history sessions having the same quality parameter in the same class of quality parameters. The darker curve (i.e., curve 500) represents the distribution of the first type of quality parameter and the lighter curve (i.e., curve 501) represents the distribution of the second type of quality parameter.

Similarly, the present disclosure may obtain a fourth ratio, a fifth ratio, a sixth ratio, and a service achievement status label of each of the plurality of historical sessions from the session test set, and provide the fourth ratio, the fifth ratio, and the sixth ratio of each of the historical sessions to the successfully trained quality parameter prediction model, so as to obtain the quality parameter of each of the historical sessions. The present disclosure may form a quality parameter profile for both types of quality parameters. An example of the quality parameter profile is shown in fig. 6. In fig. 6, the abscissa represents the quality parameter predicted by the quality parameter prediction model, and the ordinate represents the number of history sessions having the same quality parameter in the same class of quality parameters. The darker curve (i.e., curve 600) represents the distribution of the first type of quality parameter and the lighter curve (i.e., curve 601) represents the distribution of the second type of quality parameter.

Secondly, by comparing the quality parameter distribution based on the session verification set (as shown in fig. 5) and the quality parameter distribution based on the test set (as shown in fig. 6), the quality parameter distributions of the two are basically the same, and therefore, the present disclosure can determine that there is indeed an association between the emotion information of the historical conversational sentence and the service achievement status label.

In addition, the present disclosure may also utilize a session training set to obtain a quality parameter profile. An example of the quality parameter distribution map is shown in fig. 7. In fig. 7, the abscissa represents the quality parameter predicted by the quality parameter prediction model, and the ordinate represents the number of history sessions having the same quality parameter in the same class of quality parameters. The darker curve (i.e., curve 700) represents the distribution of the first type of quality parameter and the lighter curve (i.e., curve 701) represents the distribution of the second type of quality parameter.

The quality parameter distribution conditions of the three are basically the same by comparing the quality parameter distribution condition based on the session verification set (as shown in fig. 5), the quality parameter distribution condition based on the test set (as shown in fig. 6) and the quality parameter distribution condition based on the session training set (as shown in fig. 7), so that the correlation between the emotion information of the historical session sentence and the service achievement state label can be further determined.

Optionally, the session verification set and the session test set each include a plurality of historical sessions, and each historical session includes a plurality of historical session statements. The historical sessions contained in the session training set, the session verification set and the session test set are usually different, so that the reliability of the verification result is ensured.

In an optional example, the present disclosure may classify the respective quality parameters of each session by using the quality parameter value intervals, so as to obtain the quality category to which each session belongs. For example, when the quality classes include a first quality session class and a second quality session class, the present disclosure may use, according to a preset quality parameter value interval, a session in which the quality parameter is located in the value interval as the first quality session class, and a session in which the quality parameter is located outside the value interval as the second quality session class. Since the session quality of the first quality session class is higher than the session quality of the second quality session class, the first quality session class may be referred to as a premium session class and the second quality session class may be referred to as a low quality session class. The quality type of each session can be conveniently determined by utilizing the quality parameter value interval, and the quality type of the session can be accurately obtained under the condition that the quality parameter value interval is reasonably set.

It should be particularly noted that, the above description is made by taking the example that the quality categories include two categories, and the quality categories in the present disclosure may also include three or more categories, as long as corresponding quality parameter value intervals are set for each quality category in advance, which is not limited in the present disclosure.

In an optional example, the present disclosure may obtain the quality parameter value interval during a process of training the quality parameter prediction model using a historical conversational sentence. For example, the present disclosure may utilize a session verification set and/or a session test set to obtain a quality parameter value interval. As a more specific example, the present disclosure uses the session verification set and the session test set to obtain the quality parameter distribution, as shown in fig. 5 and 6, and the present disclosure may determine a quality parameter value interval according to the curves shown in fig. 5 and 6, for example, the interval may be 0.42 to 0.58. If the quality parameter of a session is less than 0.42 or greater than 0.58, the session is considered to belong to a second quality session class; a session is considered to belong to the first quality session class if its quality parameter is neither less than 0.42 nor greater than 0.58.

In an alternative example, in a case that the emotion information of the conversational sentence of the present disclosure includes positive direction, negative direction, and neutral, and the quality category of the conversation includes a high quality conversation category and a low quality conversation category, the implementation manner of the present disclosure to determine the conversation feature distribution information of the predetermined conversation party according to the emotion information of the conversational sentence of the predetermined conversation party in each conversation belonging to the same quality category may include: determining distribution information of positive conversational sentences, negative conversational sentences and neutral conversational sentences according to the emotional information of the conversational sentences of the preset conversational party in each conversation belonging to the high-quality conversational category; and determining the distribution information of the positive conversational sentence, the negative conversational sentence and the neutral conversational sentence according to the emotional information of the conversational sentence of the predetermined conversational party in each conversation belonging to the low-quality conversational category. The distribution information of the forward conversational sentence may include, but is not limited to: the number of conversational sentences in the forward direction, the number of conversational sentences in the same forward direction, or the clustering information of the conversational sentences in the forward direction, etc. The distribution information of the negative conversational sentence may include, but is not limited to: the number of negative conversational sentences, the number of same negative conversational sentences, or the clustering information of negative conversational sentences, etc. The distribution information of the neutral conversational utterances may include, but is not limited to: the number of neutral conversational utterances, the number of same neutral conversational utterances, or the clustering information of neutral conversational utterances, and the like.

The method and the device can objectively and detailedly show the conversation mode and the conversation characteristics of the preset conversation party by providing the distribution information of the positive conversation sentences, the negative conversation sentences and the neutral conversation sentences of the preset conversation party, and are favorable for leading the preset conversation party to have clearer cognition on the conversation mode, the conversation characteristics and the like of the preset conversation party, thereby being favorable for helping the preset conversation party to improve the conversation quality.

Exemplary devices

Fig. 8 is a schematic structural diagram of an embodiment of an apparatus for improving session quality according to the present disclosure. The apparatus of this embodiment may be used to implement the method embodiments of the present disclosure described above.

As shown in fig. 8, the apparatus of this embodiment mainly includes: an emotion information acquisition module 800, a quality parameter determination module 801, a quality category determination module 802, and a feature distribution determination module 803. Optionally, the apparatus may further include: a first training module 804, a second training module 805, and a determine span module 806.

The obtain emotion information module 800 is used for obtaining emotion information of a conversation sentence of a predetermined conversation party in a plurality of conversations. Optionally, the emotion information of the conversational sentence in the present disclosure includes: at least one of positive, negative, and neutral. In this case, the module 800 for obtaining emotion information may provide the conversational sentence of a predetermined conversation party in the plurality of conversations to the emotion polarity prediction model, respectively, and obtain emotion information of each conversational sentence according to information output by the emotion polarity prediction model.

The determine quality parameter module 801 is configured to determine quality parameters of the sessions according to emotion information of the session sentence.

Optionally, the determining quality parameter module 801 may include: seventh sub-module 8011. The seventh sub-module 8011 may be configured to, for any conversation, determine a quality parameter of the conversation based on the number of conversational utterances in the conversation that have the same emotion information and the number of conversational utterances of the first conversational party in the conversation. In the case where the emotion information in the present disclosure includes positive, negative, and neutral, seventh sub-module 8011 may include: a first element 80111 and a second element 80112. The first unit 80111 is configured to determine the number of conversational utterances whose emotion information is positive, the number of conversational utterances whose emotion information is negative, and the number of conversational utterances whose emotion information is neutral in the conversation, and obtain a first ratio, a second ratio, and a third ratio, where the ratios of the numbers of conversational utterances of the first conversational party in the conversation are respectively obtained. The second unit 80112 is configured to predict a quality parameter of the session based on the first ratio, the second ratio and the third ratio. For example, the second unit 80112 may provide the first ratio, the second ratio and the third ratio to the quality parameter prediction model, and obtain the quality parameter of the session according to the information output by the quality parameter prediction model.

The quality class determining module 802 is configured to determine a quality class to which each of the plurality of sessions belongs according to a quality parameter of each of the plurality of sessions. For example, the quality class determining module 802 may determine, according to a preset quality parameter value interval, a session with a quality parameter located in the value interval as a first quality session class, and a session with a quality parameter located outside the value interval as a second quality session class; wherein the session quality of the first quality session class is higher than the session quality of the second quality session class.

The determining characteristic distribution module 803 is configured to determine conversation characteristic distribution information of a predetermined conversation party according to emotion information of a conversation sentence of the predetermined conversation party in each conversation belonging to the same quality class. For example, the determine feature distribution module 804 may determine distribution information of the positive direction conversational sentence, the negative direction conversational sentence, and the neutral conversational sentence according to the emotion information of the conversational sentence of the predetermined conversational partner in each conversation belonging to the first quality conversational class, and determine distribution information of the positive direction conversational sentence, the negative direction conversational sentence, and the neutral conversational sentence according to the emotion information of the conversational sentence of the predetermined conversational partner in each conversation belonging to the second conversational class.

The first training module 804 may include: a first sub-module 8041, a second sub-module 8042, a third sub-module 8043, a fourth sub-module 8044, a fifth sub-module 8045, and a sixth sub-module 8046. The first sub-module 8041 is configured to provide a plurality of historical conversational sentences in the training set to the emotion polarity prediction model respectively. The second sub-module 8042 is configured to adjust model parameters of the emotion polarity prediction model according to an emotion polarity prediction result output by the emotion polarity prediction model and a difference between emotion polarity labels of historical conversation sentences in the training set. The third sub-module 8043 is configured to provide the plurality of historical conversational sentences in the verification set to the parameter-adjusted emotion polarity prediction model respectively. The fourth sub-module 8044 is configured to correct the emotion polarity labels of the historical conversational sentences in the training set according to the emotion polarity prediction result output by the emotion polarity prediction model after parameter adjustment and the difference between the emotion polarity labels of the historical conversational sentences in the verification set. For example, the fourth sub-module 8044 may set a regular expression according to the emotion polarity prediction result output by the emotion polarity prediction model after parameter adjustment and the difference between the emotion polarity labels of the historical conversational sentences in the verification set, and modify the emotion polarity labels of the historical conversational sentences in the training set according to the regular expression. In addition, the fourth sub-module 8044 may also adjust the hyper-parameter of the current emotion polarity prediction model according to the emotion polarity prediction result output by the emotion polarity prediction model after parameter adjustment and the difference between the emotion polarity labels of the historical conversation sentences in the verification set. The fifth sub-module 8045 is configured to provide the plurality of historical conversational sentences with emotion polarity labels in the test set to the parameter-adjusted emotion polarity prediction model respectively. The sixth sub-module 8046 is configured to determine whether the emotion polarity prediction model is successfully trained according to the emotion polarity prediction result output by the parameter-adjusted emotion polarity prediction model and the difference between the emotion polarity labels of the historical session sentences in the test set; and if the judgment result is that the training is not completed successfully, triggering the first training module to continue to train the emotion polarity prediction model by using a plurality of historical conversation sentences in the training set.

The second training module 805 is configured to obtain emotion information of history session statements of a first session party in a plurality of history sessions, and for any history session, the second training module 805 determines ratios of the number of history session statements of which the emotion information is positive, the number of history session statements of which the emotion information is negative, and the number of history session statements of which the emotion information is neutral, to the number of history session statements of the first session party in the history session, respectively, and obtains a fourth ratio, a fifth ratio, and a sixth ratio; thereafter, the second training module 805 may determine the model parameters of the quality parameter prediction model according to the fourth ratio, the fifth ratio, the sixth ratio of the plurality of historical sessions, and the service achievement state labels of the plurality of historical sessions.

The value-taking interval determining module 806 is configured to determine a value-taking interval of the quality parameter according to data distribution of the service achievement state labels of the history sessions, where the quality parameters are obtained by respectively predicting the plurality of history sessions by the quality parameter prediction model.

The operations specifically executed by each module, each sub-module, and each unit included in the apparatus of the present disclosure may be referred to in the description of the above method embodiments, and are not described in detail here.

Exemplary electronic device

An electronic device according to an embodiment of the present disclosure is described below with reference to fig. 9. FIG. 9 shows a block diagram of an electronic device in accordance with an embodiment of the disclosure. As shown in fig. 9, the electronic device 91 includes one or more processors 911 and memory 912.

The processor 911 may be a Central Processing Unit (CPU) or other form of processing unit having capabilities for improving the quality of a session and/or instruction execution capabilities, and may control other components in the electronic device 91 to perform desired functions.

Memory 912 may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory, for example, may include: random Access Memory (RAM) and/or cache memory (cache), etc. The nonvolatile memory, for example, may include: read Only Memory (ROM), hard disk, flash memory, and the like. One or more computer program instructions may be stored on the computer-readable storage medium and executed by processor 911 to implement the methods for improving session quality and/or other desired functionality of the various embodiments of the present disclosure described above. Various contents such as an input signal, a signal component, a noise component, etc. may also be stored in the computer-readable storage medium.

In one example, the electronic device 91 may further include: an input device 913, and an output device 914, among others, which are interconnected by a bus system and/or other form of connection mechanism (not shown). The input device 913 may include, for example, a keyboard, a mouse, or the like. The output device 914 may output various information to the outside. The output devices 914 can include, for example, a display, speakers, a printer, and a communication network and remote output devices connected thereto, among others.

Of course, for simplicity, only some of the components of the electronic device 91 relevant to the present disclosure are shown in fig. 9, omitting components such as buses, input/output interfaces, and the like. In addition, the electronic device 91 may include any other suitable components depending on the particular application.

Exemplary computer program product and computer-readable storage Medium

In addition to the above-described methods and apparatus, embodiments of the present disclosure may also be a computer program product comprising computer program instructions that, when executed by a processor, cause the processor to perform the steps in the method for improving session quality according to various embodiments of the present disclosure described in the "exemplary methods" section above of this specification.

The computer program product may write program code for carrying out operations for embodiments of the present disclosure in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server. Furthermore, embodiments of the present disclosure may also be a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, cause the processor to perform steps in a method for improving session quality according to various embodiments of the present disclosure described in the "exemplary methods" section above in this specification.

The computer-readable storage medium may take any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may include, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium may include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The foregoing describes the general principles of the present disclosure in conjunction with specific embodiments, however, it is noted that the advantages, effects, etc. mentioned in the present disclosure are merely examples and are not limiting, and they should not be considered essential to the various embodiments of the present disclosure. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the disclosure is not intended to be limited to the specific details so described. In the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same or similar parts in the embodiments are referred to each other. For the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The block diagrams of devices, apparatuses, systems referred to in this disclosure are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, and systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," comprising, "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".

The methods and apparatus of the present disclosure may be implemented in a number of ways. For example, the methods and apparatus of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order for the steps of the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above unless specifically stated otherwise. Further, in some embodiments, the present disclosure may also be embodied as programs recorded in a recording medium, the programs including machine-readable instructions for implementing the methods according to the present disclosure. Thus, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

It is also noted that in the devices, apparatuses, and methods of the present disclosure, each component or step can be decomposed and/or recombined. These decompositions and/or recombinations are to be considered equivalents of the present disclosure.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects, and the like, will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the disclosure. Thus, the present disclosure is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the disclosure to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.

Claims

1. A method for improving session quality, comprising:

obtaining emotion information of a conversation statement of a predetermined conversation party in a plurality of conversations;

determining respective quality parameters of the plurality of conversations according to the emotional information of the conversation sentences;

determining quality categories to which the plurality of sessions belong according to respective quality parameters of the plurality of sessions;

and determining the conversation feature distribution information of a predetermined conversation party according to the emotion information of the conversation sentences of the predetermined conversation party in all the conversations belonging to the same quality class.

2. The method of claim 1, wherein the obtaining of emotion information of conversational utterances of predetermined conversational parties of the plurality of conversations comprises:

providing the conversation sentences of a predetermined conversation party in the plurality of conversations to an emotion polarity prediction model respectively;

obtaining the emotion information of each conversation statement according to the information output by the emotion polarity prediction model;

wherein, the emotion information of the conversational sentence comprises: at least one of positive, negative, and neutral.

3. The method of claim 2, wherein the emotion polarity prediction model is obtained by training with a historical conversational sentence, and the training process of the emotion polarity prediction model comprises:

respectively providing a plurality of historical conversation sentences in the training set to an emotion polarity prediction model;

adjusting model parameters of the emotion polarity prediction model according to the emotion polarity prediction result output by the emotion polarity prediction model and the difference of emotion polarity labels of historical conversation sentences in the training set;

respectively providing a plurality of historical conversation sentences in the verification set to the emotion polarity prediction model after parameter adjustment;

and correcting the emotion polarity labels of the historical conversation sentences in the training set according to the emotion polarity prediction result output by the emotion polarity prediction model after the parameters are adjusted and the difference between the emotion polarity labels of the historical conversation sentences in the verification set, and adjusting the hyper-parameters of the emotion polarity prediction model.

4. The method of claim 3, wherein the modifying the emotion polarity labels of the historical conversational sentences in the training set according to the difference between the emotion polarity prediction result output by the parameter-adjusted emotion polarity prediction model and the emotion polarity labels of the historical conversational sentences in the verification set comprises:

setting a regular expression according to the emotion polarity prediction result output by the emotion polarity prediction model after the parameter adjustment and the difference of the emotion polarity labels of the historical conversation sentences in the verification set;

and modifying the emotion polarity labels of the historical conversation sentences in the training set according to the regular expression.

5. The method of claim 3 or 4, wherein the training process of the emotion polarity prediction model further comprises:

respectively providing a plurality of historical conversation sentences with emotion polarity labels in the test set to the emotion polarity prediction model after parameter adjustment;

judging whether the emotion polarity prediction model is successfully trained or not according to the emotion polarity prediction result output by the emotion polarity prediction model after the parameter adjustment and the difference of the emotion polarity labels of the historical conversation sentences in the test set;

and if the judgment result is that the training is not completed successfully, continuing to train the emotion polarity prediction model by using a plurality of historical conversation sentences in a training set.

6. The method of any one of claims 1 to 5, wherein the determining quality parameters of each of the plurality of conversations from emotion information of the conversational sentence comprises:

for any conversation, determining the quality parameter of the conversation according to the number of the conversation sentences with the same emotion information in the conversation and the number of the conversation sentences of the first conversation party in the conversation.

7. The method of claim 6, wherein the affective information comprises: positive, negative, and neutral;

the determining the quality parameter of the conversation according to the number of the conversation sentences with the same emotion information in the conversation and the number of the conversation sentences of the first conversation party in the conversation includes:

determining the number of conversational sentences of which the emotional information is positive, the number of conversational sentences of which the emotional information is negative and the number of conversational sentences of which the emotional information is neutral, and obtaining a first ratio, a second ratio and a third ratio by respectively comparing the ratios of the numbers of the conversational sentences of a first conversational party in the conversation;

and predicting the quality parameter of the session according to the first ratio, the second ratio and the third ratio.

8. An apparatus for improving session quality, wherein the apparatus comprises:

the emotion information acquisition module is used for acquiring emotion information of a conversation statement of a preset conversation party in a plurality of conversations;

a quality parameter determining module, configured to determine quality parameters of the sessions according to emotion information of the session statements;

a quality class determining module, configured to determine, according to respective quality parameters of the multiple sessions, quality classes to which the multiple sessions belong;

and the determining characteristic distribution module is used for determining the conversation characteristic distribution information of the predetermined conversation party according to the emotion information of the conversation sentences of the predetermined conversation party in all the conversations belonging to the same quality class.

9. A computer-readable storage medium, the storage medium storing a computer program for performing the method of any of the preceding claims 1-7.

10. An electronic device, the electronic device comprising:

a processor;

a memory for storing the processor-executable instructions;

the processor is configured to read the executable instructions from the memory and execute the instructions to implement the method of any one of claims 1-7.