WO2020147395A1

WO2020147395A1 - Emotion-based text classification method and device, and computer apparatus

Info

Publication number: WO2020147395A1
Application number: PCT/CN2019/117161
Authority: WO
Inventors: 金戈; 徐亮
Original assignee: 平安科技（深圳）有限公司
Priority date: 2019-01-17
Filing date: 2019-11-11
Publication date: 2020-07-23
Also published as: CN109960725A

Abstract

An emotion-based text classification method comprises: acquiring an emotion classification task comprising an identifier of a text to be classified; acquiring a corresponding text according to the identifier, the text comprising multiple questions and answers corresponding to the questions; calling a corresponding emotion classification model according to the emotion classification task, the emotion classification model comprising multiple submodels corresponding to the questions; inputting the answers corresponding to the questions into the submodels corresponding to the questions, performing operations by means of the submodels, and outputting emotion scores corresponding to the answers; and recognizing an emotion category corresponding to the text according to the emotion scores corresponding to the multiple answers.

Description

Emotion-based text classification processing method, device and computer equipment

This application claims the priority of the Chinese patent application filed to the Chinese Patent Office on January 17, 2019. The application number is 2019100428375, and the application name is "Sentiment-based text classification processing methods, devices and computer equipment". The reference is incorporated in this application.

Technical field

This application relates to an emotion-based text classification processing method, device, computer equipment and storage medium.

Background technique

With the development of computer technology, computer technology is gradually applied to the processing of emotion classification of text. For example, in an AI (Artificial Intelligence) interview, the interviewer’s responses are classified by emotional classification to obtain the interviewer’s emotional attitude, so that the company can screen the interviewer according to the needs. For another example, in a questionnaire survey, emotional classification processing is performed on the answers of the respondents to understand the emotional tendencies of the respondents.

In the traditional method, sentiment classification is usually performed on all the collected texts uniformly. However, the inventor realizes that in many cases, text content corresponds to different problems, and different problems may reflect different emotions. The unified sentiment classification of all text content cannot accurately reflect the sentiment expression under different problems, resulting in inaccurate results of sentiment classification. Therefore, how to improve the accuracy of sentiment classification has become a technical problem that needs to be solved at present.

Summary of the invention

According to various embodiments disclosed in the present application, an emotion-based text classification processing method, device, computer equipment, and storage medium are provided.

An emotion-based text classification processing method includes:

Acquiring an emotion classification task, where the emotion classification task includes a text identifier to be classified;

Obtaining a corresponding text to be classified according to the text identifier to be classified, where the text to be classified includes a plurality of questions and question answers corresponding to the questions;

Calling a corresponding sentiment classification model according to the sentiment classification task, the sentiment classification model including a plurality of sub-models corresponding to the question;

Input the question answer corresponding to the question into the sub-model corresponding to the question, and output the emotional score corresponding to the question answer through the sub-model calculation; and

Identify the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the answers to the multiple questions.

An emotion-based text classification processing device includes:

The task acquisition module is used to acquire the sentiment classification task, and the sentiment classification task includes the text identification to be classified;

A text obtaining module, configured to obtain a corresponding text to be classified according to the text identifier to be classified, where the text to be classified includes multiple questions and question answers corresponding to the questions;

The model calling module is used to call the corresponding emotion classification model according to the emotion classification task, the emotion classification model includes a plurality of sub-models corresponding to the question; the question answer corresponding to the question is input to the question corresponding The sub-model of, through the sub-model operation, output the emotional score corresponding to the question answer; and

The emotion type recognition module is used to recognize the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the multiple question answers.

A computer device includes a memory and one or more processors. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the one or more processors are executed The following steps:

One or more non-volatile computer-readable storage media storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors perform the following steps:

The details of one or more embodiments of the application are set forth in the drawings and description below. Other features and advantages of this application will become apparent from the description, drawings, and claims.

BRIEF DESCRIPTION

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings that need to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. A person of ordinary skill in the art can obtain other drawings based on these drawings without creative work.

Fig. 1 is an application environment diagram of an emotion-based text classification processing method according to one or more embodiments.

Fig. 2 is a schematic flowchart of an emotion-based text classification processing method according to one or more embodiments.

Fig. 3 is a schematic flowchart of the steps of extracting subject information according to emotion types in one or more embodiments.

Fig. 4 is a block diagram of an emotion-based text classification processing apparatus according to one or more embodiments.

Figure 5 is a block diagram of a computer device in accordance with one or more embodiments.

detailed description

In order to make the technical solutions and advantages of the present application more clear, the following describes the present application in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

The emotion-based text classification processing method provided in this application can be applied to the application environment shown in FIG. 1. The terminal 102 communicates with the server 104 through the network through the network. The server 104 obtains the sentiment classification task initiated by the terminal 102. The sentiment classification task includes the text identification to be classified. The server 104 obtains the corresponding text to be classified according to the text identification to be classified, and calls the corresponding sentiment classification model according to the obtained sentiment classification task. The server 104 inputs the question and question answer in the text to be classified into the sub-model of the emotion classification model, and outputs the emotion score corresponding to the question through the sub-model calculation. The server 104 recognizes the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the multiple questions. The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices. The server 104 may be implemented by an independent server or a server cluster composed of multiple servers.

In one of the embodiments, as shown in FIG. 2, an emotion-based text classification processing method is provided. Taking the method applied to the server 104 in FIG. 1 as an example for description, the method includes the following steps:

Step 202: Obtain an emotion classification task, and the emotion classification task includes a text identifier to be classified.

The server obtains the emotion classification task, analyzes the obtained emotion classification task, and obtains the text identification to be classified included in the emotion classification task. The emotion classification task can be created by the terminal and uploaded to the server. The sentiment classification task includes the text identification to be classified, and the text identification to be classified points to the corresponding text to be classified. The text identifiers to be classified can include multiple types. In one of the embodiments, the text identifier to be classified may be the file name of the text to be classified. The server can perform sentiment classification tasks, and perform sentiment classification processing on the text to be classified corresponding to the text identifier to be classified.

Step 204: Obtain the corresponding text to be classified according to the identifier of the text to be classified. The text to be classified includes multiple questions and question answers corresponding to the questions.

The server may obtain the corresponding text to be classified according to the identifier of the text to be classified. Specifically, the server may obtain the mapping relationship between the text identification to be classified and the text to be classified, and use the mapping relationship to obtain the text to be classified corresponding to the text identification to be classified. The text to be classified includes multiple questions and corresponding question answers.

The server can ask different questions for different users, or ask different users the same question. The question answer is the answer made by the user for each question. For example, in the process of an artificial intelligence interview, different questions can be asked for different interviewers, and the interviewer's answer can be received as the answer to the question corresponding to the question, so as to avoid the interviewer's private communication and more truly reflect the interviewer's adaptability. There are many ways to ask questions. The server can randomly select questions from preset questions to ask questions, or select related questions to ask questions based on the interviewer's answers. For another example, in the process of a questionnaire survey, the server may ask the same questions for different respondents, and receive answers to the questions from the respondents, so as to understand the respondent's emotional tendency toward these questions.

Step 206: Invoke a corresponding emotion classification model according to the emotion classification task. The emotion classification model includes a plurality of sub-models corresponding to the question.

The server can call the corresponding sentiment classification model according to the sentiment classification task. The sentiment classification model is a model obtained through training for sentiment classification of questions and question answer texts. The sentiment classification model can be trained by a variety of classification models, such as the Fasttext model. The sentiment classification model includes multiple sub-models, and different questions can correspond to different sub-models. The server can call the sub-model corresponding to the question in the sentiment classification model according to the sentiment classification task to perform sentiment classification processing on the response to the question.

Step 208: Input the question answer corresponding to the question into the sub-model corresponding to the question, and output the emotional score corresponding to the question answer through the sub-model calculation.

The server inputs the question answer corresponding to the question in the text to be classified into the sub-model corresponding to the question, and calls the sub-model to perform sentiment classification on the question answer. Through the operation of the sub-model, the sentiment score corresponding to the answer to the question is output. The sentiment score can reflect the emotional tendency of the question answer to the question. Specifically, the server obtains the probability of each emotion label corresponding to the question answer by calling the sub-model for calculation. Calculate according to the probability of the corresponding sentiment label, obtain the sentiment score corresponding to the question answer, and output the sentiment score corresponding to the question answer. For example, using the Fasttext model to perform sentiment classification on the text to be classified. The server inputs the question answers in the text to be classified into the sub-models of the Fasttext model corresponding to the questions, and the sub-models perform sentiment classification processing on the question responses to obtain the probabilities corresponding to multiple sentiment labels. For example, the emotion label may include positive and negative, and the result of the sentiment classification of the answer to the question is: a positive probability of 68%, and a negative probability of 32%. According to the sentiment label probability of the question answer, the sentiment score corresponding to the corresponding question answer is calculated. For example, when the full score is 10 points, the sentiment score is 7 points.

Step 210: Identify the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the multiple question answers.

The server synthesizes the sentiment scores corresponding to the multiple question answers, recognizes the sentiment type corresponding to the text to be classified according to the sentiment scores corresponding to the multiple question responses, and completes the sentiment classification task of emotional classification of the text to be classified.

In this embodiment, after obtaining the text identification to be classified in the sentiment classification task, multiple questions corresponding to the text to be classified and the question answer corresponding to each question are obtained according to the text identification to be classified, and the response in the sentiment classification model is called Multiple sub-models classify the sentiment of each question answer one by one, and identify the sentiment type corresponding to the text to be classified according to the obtained sentiment score corresponding to each question answer. By calling the corresponding sub-model, according to the specific situation of each question, the sentiment classification of the question answer is carried out one by one, and then the sentiment score corresponding to each question answer is synthesized to identify the sentiment type corresponding to the text to be classified. Compared with the unified sentiment classification of all texts in the traditional way, it fully combines the questions and carries out sentiment classification to the corresponding question answers, which effectively improves the accuracy of sentiment classification of the text.

In one of the embodiments, identifying the sentiment type corresponding to the text to be classified according to the sentiment scores corresponding to multiple question answers includes: obtaining the weight corresponding to each question; calculating according to the weight and the sentiment score corresponding to the question answer to obtain Modified score; add up the modified scores corresponding to the answers to multiple questions to obtain the total emotional score; identify the emotional type corresponding to the text to be classified according to the emotional total score.

The server obtains the weight corresponding to each question. Since each question has a different importance to the degree of classification of the classified text, in order to balance the impact of each question on the degree of sentiment classification, a weight is set for each question. The user can preset the corresponding importance for each question, and the server sets the corresponding weight of each question according to the importance of each question in all the questions. The weight of important questions can be higher, and the weight of simple questions can be lower. The server calculates according to the weight and the emotional score corresponding to each question answer, and obtains the modified score corresponding to each question answer. The revised score can objectively reflect the emotion of the corresponding question answer in the text to be classified. For example, in the interview process, different types of questions may appear, such as questions about identity, questions about personal skills, and questions about job search intentions. Among the above three types of questions, the weight of the question about identity can be less than the weight of the question about personal skills and job search intentions. Therefore, according to the sentiment scores of the question answers of the three types of questions, it is necessary to recalculate the corresponding weights of the questions to obtain the revised scores corresponding to the question answers.

The server accumulates the modified scores corresponding to the multiple question answers calculated to obtain the total emotional score. The total sentiment score is the total sentiment score of the text to be classified into the question. The server recognizes the emotion type corresponding to the text to be classified according to the total emotional score, and completes the emotion classification task of text classification of the text to be classified.

In this embodiment, the server obtains the weight corresponding to each question, calculates the revised score corresponding to each question answer, and accumulates the revised scores corresponding to the multiple question answers, to obtain the total emotion corresponding to the text to be classified. The score is used to identify the emotional type corresponding to the text to be classified through the emotional total score. By combining the weight corresponding to each question, the influence of the importance of the question on the final emotion type is solved, and the accuracy of emotion classification for the text to be classified is effectively improved.

In one of the embodiments, before obtaining the emotion classification task, the above emotion-based text classification processing method further includes: obtaining reply data corresponding to the question; identifying the data type corresponding to the reply data; when the data type is a voice type, The reply data of voice type is converted into reply data of text type; the text to be classified is generated according to the reply data of text type and the question.

The user's response data to the question can include multiple data types. For example, it can include text type and voice type. The server can obtain the reply data uploaded by the user in response to the question. After obtaining the reply data corresponding to the question, the server can determine the data type of the reply data. If the data type is a text type, the text type response data is directly used as the question response, and the text to be classified is generated by combining multiple questions and the corresponding question responses of the questions. If the reply data is voice type data, the server can convert the voice data into text type reply data, use the reply data as a question reply, and generate a text to be classified.

In this embodiment, in addition to receiving text-type reply data, the server may also receive voice-type reply data. By converting voice type response data into text type response data, the text to be classified is generated based on the question and text type response data, which enriches the data types of compatible question responses and facilitates users to answer questions in multiple ways.

In one of the embodiments, the question and the corresponding question answer are input to the corresponding sub-model, and the sub-model calculation is used to output the sentiment score corresponding to the question, including: segmenting the question answer to obtain multiple words; calling according to the question Corresponding sub-models; use the sub-models to classify the emotions of words and obtain the probability values corresponding to multiple emotional expressions; determine the emotional scores corresponding to the question answers according to the probability values.

The server can use multiple methods to segment the answers to the questions in the classified text to obtain multiple words. Specifically, the server may use one or a combination of string matching, comprehension, statistics and other methods to segment the answer to the question. The string matching method refers to matching the question answer with the entry in the preset dictionary. If the string is found in the dictionary, it is considered that the match is successful, that is, a word is recognized. The way of string matching can include forward string matching and reverse string matching. The way of understanding means that the server performs semantic analysis and syntactic analysis when cutting the question answer, and uses semantic information and syntactic information to process the ambiguity that occurs when the words are cut. The statistical method means that the server counts the frequency of the combination of adjacent co-occurring words in the question answer, and performs word segmentation based on their co-occurrence information. In one of the embodiments, the server may also introduce N-gram vectors when segmenting words. The obtained vector features are used to determine the word sequence after word segmentation of the question answer, which more accurately reflects the content of the question answer, and effectively improves the accuracy of sentiment classification of the classified text.

Each question corresponds to a sub-model. The server obtains the mapping relationship between the problem and the sub-model, and calls the sub-model corresponding to the problem according to the mapping relationship. The server inputs the multiple words obtained by word segmentation into the sub-model, and uses the sub-model to classify the words corresponding to the question answer, and obtains the probability values corresponding to the multiple preset emotion labels. The probability value indicates the probability that the emotion of the answer to the question belongs to the corresponding emotion label. The server may calculate the emotion score corresponding to the question answer according to the probability value corresponding to each of the multiple emotion tags.

In this embodiment, by segmenting the question answer, the sub-model corresponding to the question is called to classify the obtained words, and the sentiment score corresponding to the question answer is obtained. The corresponding sub-model is called for the specific content of each question, and the sentiment score corresponding to the question answer is calculated, which effectively improves the accuracy of sentiment classification for the text to be classified.

In one of the embodiments, after obtaining the text to be classified, the server cleans the text to be classified. For example, delete the text to be classified that does not meet the preset rules. Then perform sentiment classification on the text to be classified after cleaning. Specifically, the server divides the question answer into words, and after obtaining multiple words, calls the corresponding sub-model according to the question. According to the needs of the sub-model, multiple words are sorted into the data form required by the sub-model. For example, the server can sort the words into a table form, and then input the words in the table form into the sub-model, and use the sub-model to classify the sentiment of the question answer to obtain the sentiment score corresponding to the question answer.

In this embodiment, after the text to be classified is obtained, the text to be classified is cleaned up, the illegal text to be classified is eliminated, the words obtained by word segmentation are sorted, and the data form corresponding to the sub-model is obtained and input into the sub-model. Effectively improve the efficiency of sentiment classification for classified text.

In one of the embodiments, after identifying the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the multiple question answers, the above method further includes: extracting subject information according to the emotion type. As shown in Figure 3, this step specifically includes:

Step 302: Acquire preset types of emotional needs.

Step 304: Match the classified emotion type with the emotion demand type to obtain the corresponding first matching degree.

Step 306: Filter out target emotion types that meet preset conditions according to the first matching degree.

Step 308: Extract subject information corresponding to the target emotion type.

The server can obtain the type of emotional demand preset by the user. The type of emotional demand refers to the emotional type that meets the needs of the user, and can be preset according to the actual needs of the user. The server may use multiple methods to match the emotion type obtained after emotion classification of the text to be classified with the emotion demand type. Specifically, the server may match the emotional types obtained after classification with the emotional demand types one by one, and may also call multiple threads to match the emotional type and the emotional demand type in parallel. After the server matches the emotion type with the emotion demand type, the first matching degree corresponding to the emotion type is obtained. The server screens the first matching degrees corresponding to multiple emotion types, and screens out target emotion types that meet preset conditions. The preset condition may be a condition preset by the user, such as filtering out the emotion types whose first matching degree is greater than a threshold. The server extracts subject information corresponding to the target emotion type that meets the preset conditions to obtain subject information that meets the emotional demand type.

For example, in a smart interview, the server classifies the text to be classified by the interviewer and obtains the emotion type corresponding to the interviewer. The server can obtain the emotion demand type corresponding to the job position, and compare the emotion type of the interviewee with the position Match the types of emotional needs and get the first degree of matching between multiple interviewers and the position. The server screens according to the first matching degree, and screens out the target emotion type whose first matching degree meets the preset conditions, and the target emotion type is the type that meets the emotional needs of the post. The server extracts the subject information corresponding to the target emotion type that meets the conditions, and obtains the subject information that meets the job conditions. According to the emotional types of multiple interviewers, the target emotional type that matches the emotional needs of the job is selected, and the interviewers who meet the needs of the recruitment position are initially selected from multiple interviewers according to the emotional type, which effectively improves the efficiency and accuracy of the interview Sex.

In this embodiment, by matching the emotion type with the emotion demand type, the target emotion type that meets the preset condition is screened according to the first matching degree, and the emotion type is effectively used for screening to obtain the subject information that meets the conditions. Enriched target screening methods and improved screening efficiency.

In one of the embodiments, after identifying the sentiment type corresponding to the text to be classified according to the sentiment scores corresponding to the multiple question answers, the above method further includes: obtaining a variety of product information, the product information includes product types; The emotion type corresponding to the classified text is matched to obtain the second matching degree; when the second matching degree is greater than the predetermined value, the product information corresponding to the product type is marked as the target product information; the terminal identification corresponding to the text to be classified is extracted, and the target product The information is pushed to the terminal corresponding to the terminal identifier.

The server obtains product information corresponding to multiple products, and the product information includes the product type corresponding to the product. Multiple products can include products of different product types. For example, the fund products include high-risk and high-yield products as well as low-risk, low-yield products. The server can use multiple methods to match the product types of multiple products with the emotional types corresponding to the text to be classified. Specifically, the server may sequentially match product types corresponding to multiple products with emotion types, or call multiple threads to match the product types and emotion types in parallel to obtain the second matching degree between the product types and emotion types. When the second matching degree is greater than the predetermined value, the product information corresponding to the product type is marked as the target product information. The predetermined value is a value preset by the user. When the second matching degree is greater than the predetermined value, it means that the product corresponding to the product type matches the user corresponding to the emotion type. For example, high-risk and high-yield products are more likely to match users who tend to be adventurous, and low-risk and low-yield products are more likely to match users who tend to be conservative. The server extracts the terminal identifier corresponding to the terminal uploading the text to be classified, pushes the target product information to the terminal corresponding to the terminal identifier, and completes the push of product information that matches the emotional type.

In this embodiment, by matching the product type with the user's emotion type, target product information that matches the emotion type is pushed to the user, which effectively improves the accuracy of corresponding information push.

It is understandable that, in one of the embodiments, before obtaining various product information and matching product categories with emotional types, emotional demand types can also be acquired, and the classified emotional types can be matched with emotional demand types to obtain Corresponding to the first matching degree, the emotion type that meets the preset condition is filtered out according to the first matching degree.

The server then matches multiple product types with the selected emotion types to obtain the corresponding second matching degree, marks the product information corresponding to the product type with the second matching degree greater than the predetermined value as target product information, and pushes the target product information To the terminal corresponding to the emotion type. The server can use the emotion type to screen users, and screen out users who meet the type of emotional needs. For example, the server can filter out users who have a positive intention to purchase products according to their emotion types, and then match the product types with the selected user emotion types, and filter out product types that match the user's emotion types from multiple product types. Reach the effect of pushing product information that matches the user's emotional type to users with purchase intentions.

In this embodiment, the combination of screening of emotion types and screening of product types effectively improves the efficiency and accuracy of product information push.

In one of the embodiments, the above method further includes: establishing a general sentiment classification model, the general sentiment classification model includes a general sub-model; acquiring a training data set, the training data set includes multiple training texts and standard sentiment scores corresponding to the training texts; Call the general sub-model to perform operations on the training text to obtain the training emotional score; compare the training emotional score with the standard emotional score; adjust the general sub-model according to the comparison result to obtain the target sub-model.

The server can establish a general sentiment classification model, and the general sentiment classification model can adopt the Fasttext model. The server receives the training data set, and the training data set may include multiple training texts and standard sentiment scores corresponding to the training texts. The training text includes standard questions and answers to questions corresponding to the standard questions. Each standard question can correspond to multiple question answers, and the corresponding standard sentiment scores for multiple question answers. The server can use the training data set to train the general sentiment classification model. Specifically, the server invokes the general sub-model in the general sentiment classification model corresponding to the standard question, inputs the corresponding question answer into the general sub-model, and outputs the training sentiment score through the general sub-model calculation. The server may compare the training sentiment score with the standard sentiment score, and adjust the corresponding general sub-model according to the comparison result. After adjustment, repeatedly input the question answer into the corresponding general sub-model, and adjust the general sub-model until the training sentiment score is successfully compared with the standard sentiment score, and the target sub-model corresponding to the standard question is obtained. The server trains multiple general sub-models of the general sentiment classification model by using standard questions, question answers and standard sentiment scores in multiple training texts to obtain the target sentiment classification model.

In this embodiment, the server uses the training data set to train the general sentiment classification model to obtain a target sentiment classification model including multiple target sub-models, which effectively improves the accuracy of the target sentiment classification model for classifying text based on sentiment treatment .

It should be understood that, although the various steps in the flowchart of FIGS. 2-3 are displayed in sequence as indicated by the arrows, these steps are not necessarily performed in sequence in the order indicated by the arrows. Unless clearly stated in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least some of the steps in Figure 2-3 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. These sub-steps or stages The execution order of is not necessarily performed sequentially, but may be performed alternately or alternately with at least a part of other steps or sub-steps or stages of other steps.

In one of the embodiments, as shown in FIG. 4, an emotion-based text classification processing device is provided, which includes: a task acquisition module 402, a text acquisition module 404, a model calling module 406, and an emotion type recognition module 408, wherein:

The task acquisition module 402 is used to acquire the sentiment classification task, and the sentiment classification task includes the text identification to be classified.

The text obtaining module 404 is configured to obtain the corresponding text to be classified according to the identifier of the text to be classified. The text to be classified includes multiple questions and question answers corresponding to the questions.

The model calling module 406 is used to call the corresponding sentiment classification model according to the sentiment classification task. The sentiment classification model includes a plurality of sub-models corresponding to the question; the question answer corresponding to the question is input into the corresponding sub-model of the question, and the sub-model calculation Output the emotional score corresponding to the answer to the question.

The emotion type recognition module 408 is configured to recognize the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the multiple question answers.

In one of the embodiments, the above-mentioned model calling module 406 is also used to segment the answer to the question to obtain multiple words; call the corresponding sub-model according to the question; use the sub-model to classify the words and obtain the corresponding emotion tags Probability value: Determine the sentiment score corresponding to the question answer according to the probability value.

In one of the embodiments, the above-mentioned device further includes a need type matching module for obtaining preset emotional need types; matching the classified emotional type with the emotional need type to obtain the first matching degree corresponding to the emotional type; The first matching degree screens out the target emotion type that meets the preset conditions; extracts the subject information corresponding to the target emotion type.

In one of the embodiments, the above-mentioned device further includes a product push module for obtaining various product information, the product information includes the product type; the product type is matched with the emotion type corresponding to the text to be classified to obtain the second matching degree; When the second matching degree is greater than the predetermined value, the product information corresponding to the product type is marked as the target product information; the terminal identification corresponding to the text to be classified is extracted, and the target product information is pushed to the terminal corresponding to the terminal identification.

In one of the embodiments, the above-mentioned emotion type recognition module 408 is also used to obtain the weight corresponding to each question; calculate according to the weight and the emotion score corresponding to the question answer to obtain the modified score; correspond to multiple question answers The modified scores of is accumulated to obtain the total emotional score; the emotional type corresponding to the text to be classified is identified according to the total emotional score.

In one of the embodiments, the above-mentioned task acquisition module 402 is also used to acquire the reply data corresponding to the question; identify the data type corresponding to the reply data; when the data type is a voice type, convert the reply data of the voice type into a text type. Reply data; generate text to be classified according to the reply data of the text type and the question.

In one of the embodiments, the above-mentioned device further includes a model training module for establishing a general sentiment classification model, the general sentiment classification model includes a general sub-model; the training data set is obtained, and the training data set includes multiple training texts and corresponding training texts. Standard sentiment score; call the general sub-model to calculate the training text to obtain the training sentiment score; compare the training sentiment score with the standard sentiment score; adjust the general sub-model according to the comparison result to obtain the target sub-model .

For the specific limitation of the emotion-based text classification processing device, please refer to the above limitation of the emotion-based text classification processing method, which will not be repeated here. The various modules in the above emotion-based text classification processing device can be implemented in whole or in part by software, hardware, and combinations thereof. The above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. 5. The computer device includes a processor, memory, network interface, and database connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer-readable instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium. The database of the computer equipment is used to store emotion-based text classification processing data. The network interface of the computer device is used to communicate with external terminals through a network connection. The computer-readable instructions are executed by the processor to realize an emotion-based text classification processing method.

Those skilled in the art may understand that the structure shown in FIG. 5 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may Include more or less components than shown in the figure, or combine certain components, or have a different arrangement of components.

A computer device includes a memory and one or more processors. The memory stores computer readable instructions. When the computer readable instructions are executed by the processor, the one or more processors execute the above method. step.

One or more non-volatile computer-readable storage media storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute A step of.

A person of ordinary skill in the art may understand that all or part of the process in the method of the above embodiments may be completed by instructing relevant hardware through computer-readable instructions, and the computer-readable instructions may be stored in a non-volatile computer-readable In the storage medium, when the computer-readable instructions are executed, they may include the processes of the foregoing method embodiments. Wherein, any reference to the memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and/or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not a limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should It is considered as the range described in this specification.

The above-mentioned embodiments only express several implementation manners of the present application, and their descriptions are more specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that, for those of ordinary skill in the art, without departing from the concept of the present application, a number of modifications and improvements can also be made, which all fall within the protection scope of the present application. Therefore, the protection scope of the patent of this application shall be subject to the appended claims.

Claims

An emotion-based text classification processing method includes:

Acquiring an emotion classification task, where the emotion classification task includes a text identifier to be classified;

Obtaining a corresponding text to be classified according to the text identifier to be classified, where the text to be classified includes a plurality of questions and question answers corresponding to the questions;

Calling a corresponding sentiment classification model according to the sentiment classification task, the sentiment classification model including a plurality of sub-models corresponding to the question;

Input the question answer corresponding to the question into the sub-model corresponding to the question, and output the emotional score corresponding to the question answer through the sub-model calculation; and

Identify the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the answers to the multiple questions.
The method according to claim 1, wherein the inputting the question and the corresponding question answer to the corresponding sub-model, and outputting the sentiment score corresponding to the question through the sub-model calculation, comprises:

Segment the answer to the question and get multiple words;

Call the corresponding sub-model according to the problem;

Use the sub-model to perform emotion classification on the words to obtain probability values corresponding to multiple emotion labels; and

Determine the sentiment score corresponding to the question answer according to the probability value.
The method according to claim 1, characterized in that, after the identifying the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the multiple questions, the method further comprises:

Obtain preset types of emotional needs;

Matching the classified emotion type with the emotion demand type to obtain the first matching degree corresponding to the emotion type;

Filter out target emotion types that meet preset conditions according to the first matching degree; and

Extract the subject information corresponding to the target emotion type.
The method according to any one of claims 1 to 3, wherein after the identifying the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the multiple question answers, the method further comprises:

Obtain a variety of product information, where the product information includes product types;

Matching the product type with the emotion type corresponding to the text to be classified to obtain a second matching degree;

When the second degree of matching is greater than a predetermined value, mark the product information corresponding to the product type as target product information; and

The terminal identifier corresponding to the text to be classified is extracted, and the target product information is pushed to the terminal corresponding to the terminal identifier.
The method according to claim 1, wherein the identifying the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the multiple question answers comprises:

Get the weight corresponding to each question;

Calculate according to the weight and the sentiment score corresponding to the question answer to obtain the revised score;

Accumulate the revised scores corresponding to the answers to multiple questions to get the total emotional score; and

Identify the emotion type corresponding to the text to be classified according to the total emotion score.
The method according to claim 1, characterized in that, before said obtaining the emotion classification task, the method further comprises:

Obtaining response data corresponding to the question;

Identify the data type corresponding to the reply data;

When the data type is a voice type, converting the voice type reply data into text type reply data; and

The text to be classified is generated according to the reply data of the text type and the question.
The method according to claim 1, wherein the method further comprises:

Establishing a general sentiment classification model, the general sentiment classification model including general sub-models;

Acquiring a training data set, the training data set including a plurality of training texts and standard sentiment scores corresponding to the training texts;

Calling the general sub-model to perform operations on the training text to obtain training emotion scores;

Comparing the training emotion score with the standard emotion score; and

The general sub-model is adjusted according to the comparison result to obtain the target sub-model.
An emotion-based text classification processing device, the device comprising:

The task acquisition module is used to acquire the sentiment classification task, and the sentiment classification task includes the text identification to be classified;

A text obtaining module, configured to obtain a corresponding text to be classified according to the text identifier to be classified, where the text to be classified includes multiple questions and question answers corresponding to the questions;

The model calling module is used to call the corresponding emotion classification model according to the emotion classification task, the emotion classification model includes a plurality of sub-models corresponding to the question; the question answer corresponding to the question is input to the question corresponding The sub-model of, through the sub-model operation, output the emotional score corresponding to the question answer; and

The emotion type recognition module is used to recognize the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the multiple question answers.
The device according to claim 8, wherein the model calling module is further configured to segment the answer to the question to obtain multiple words; call the corresponding sub-model according to the question; and use the sub-model to pair Emotion classification is performed on the words to obtain probability values corresponding to multiple emotional tags; and the emotional score corresponding to the question answer is determined according to the probability value.
A computer device includes a memory and one or more processors. The memory stores at least one computer-readable instruction. When the computer-readable instruction is executed by the one or more processors, the one or Multiple processors load and execute the following steps:

Acquiring an emotion classification task, where the emotion classification task includes a text identifier to be classified;

Obtaining a corresponding text to be classified according to the text identifier to be classified, where the text to be classified includes a plurality of questions and question answers corresponding to the questions;

Calling a corresponding sentiment classification model according to the sentiment classification task, the sentiment classification model including a plurality of sub-models corresponding to the question;

Input the question answer corresponding to the question into the sub-model corresponding to the question, and output the emotional score corresponding to the question answer through the sub-model calculation; and

Identify the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the answers to the multiple questions.
The computer device according to claim 10, wherein the processor further executes the following steps when executing the computer-readable instructions: segmenting the answer to the question to obtain a plurality of words; calling the corresponding child according to the question Model; using the sub-model to perform emotional classification on the words to obtain probability values corresponding to multiple emotional tags; and determining the emotional score corresponding to the question answer according to the probability value.
The computer device according to claim 10, wherein the processor further executes the following steps when executing computer-readable instructions: obtaining a preset type of emotional need; comparing the classified emotional type with the emotional need type Matching to obtain a first matching degree corresponding to the emotion type; filtering out target emotion types that meet preset conditions according to the first matching degree; and extracting subject information corresponding to the target emotion type.
The computer device according to any one of claims 10-12, wherein the processor further executes the following steps when executing computer-readable instructions: acquiring multiple product information, the product information including product types; Matching the product type with the emotion type corresponding to the text to be classified to obtain a second matching degree; when the second matching degree is greater than a predetermined value, marking the product information corresponding to the product type as target product information; And extract the terminal identifier corresponding to the text to be classified, and push the target product information to the terminal corresponding to the terminal identifier.
The computer device according to claim 10, wherein the processor further executes the following steps when executing the computer-readable instructions: obtaining the weight corresponding to each question; according to the weight and the emotional score corresponding to the question answer Calculate the value to obtain the modified score; add up the modified scores corresponding to multiple question answers to obtain the total emotional score; and identify the emotional type corresponding to the text to be classified according to the emotional total score .
The computer device according to claim 10, wherein the processor further executes the following steps when executing the computer-readable instructions: obtaining response data corresponding to the question; identifying the data type corresponding to the response data; When the data type is a voice type, converting the voice type response data into text type response data; and generating a text to be classified according to the text type response data and the question.
One or more non-volatile computer-readable storage media storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to perform the following steps:

Acquiring an emotion classification task, where the emotion classification task includes a text identifier to be classified;

Obtaining a corresponding text to be classified according to the text identifier to be classified, where the text to be classified includes a plurality of questions and question answers corresponding to the questions;

Calling a corresponding sentiment classification model according to the sentiment classification task, the sentiment classification model including a plurality of sub-models corresponding to the question;

Input the question answer corresponding to the question into the sub-model corresponding to the question, and output the emotional score corresponding to the question answer through the sub-model calculation; and

Identify the emotion type corresponding to the text to be classified according to the emotion scores corresponding to the answers to the multiple questions.
The storage medium according to claim 16, wherein when the computer-readable instructions are executed by the processor, the following steps are further executed: word segmentation is performed on the answer to the question to obtain a plurality of words; call according to the question Corresponding sub-model; using the sub-model to perform emotion classification on the words to obtain probability values corresponding to multiple emotion labels; and determining the emotion score corresponding to the question answer according to the probability value.
The storage medium according to claim 16, wherein when the computer-readable instructions are executed by the processor, the following steps are further executed: obtaining a preset type of emotional demand; comparing the classified emotion type with the emotion The need type is matched to obtain the first matching degree corresponding to the emotion type; the target emotion type that meets the preset condition is screened according to the first matching degree; and the subject information corresponding to the target emotion type is extracted.
The storage medium according to any one of claims 16-18, wherein when the computer-readable instructions are executed by the processor, the following steps are further executed: obtaining information of a variety of products, the product information including products Type; match the product type with the emotion type corresponding to the text to be classified to obtain a second matching degree; when the second matching degree is greater than a predetermined value, mark the product information corresponding to the product type as a target Product information; and extracting the terminal identifier corresponding to the text to be classified, and pushing the target product information to the terminal corresponding to the terminal identifier.
The storage medium according to claim 16, wherein when the computer-readable instructions are executed by the processor, the following steps are further performed: obtaining a weight corresponding to each question; corresponding to the weight and the question answer Calculate the sentiment score of, to obtain the modified score; add up the modified scores corresponding to multiple question answers to obtain the sentiment total score; and identify the corresponding text to be classified according to the sentiment total score Type of emotion.