CN115186678A - Emotional tendency analysis method and system for question in intelligent question-answering system - Google Patents
Emotional tendency analysis method and system for question in intelligent question-answering system Download PDFInfo
- Publication number
- CN115186678A CN115186678A CN202210944030.2A CN202210944030A CN115186678A CN 115186678 A CN115186678 A CN 115186678A CN 202210944030 A CN202210944030 A CN 202210944030A CN 115186678 A CN115186678 A CN 115186678A
- Authority
- CN
- China
- Prior art keywords
- analysis
- question
- text
- intelligent
- answering
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/242—Dictionaries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Databases & Information Systems (AREA)
- Human Computer Interaction (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an emotional tendency analysis method and system for questions asked in an intelligent question-answering system, which comprises a task management module, an intelligent question-answering management module, a customer service intelligence selection module, a user characteristic analysis module and a service management module, wherein the user characteristic analysis module is a core module, the module mainly extracts characteristics of two aspects of real-time emotional polarity and user key phrases aiming at a customer chatting record, and the real-time emotional analysis task carries out emotional analysis aiming at current and historical chatting records generated by customers and customer service in an instant chatting process to obtain two types of positive or negative emotional polarities, so that the intelligent customer service system can not only solve the problems of the customers, but also can guide the customers to keep or change to positive emotions according to the emotional changes of the customers.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to an emotional tendency analysis method and system for questions asked in an intelligent question-answering system.
Background
With the rise of internet technology and artificial intelligence, online intelligent question-answering systems in the form of customer service robots and the like appear. The system processes the question which is described by natural language and is input by the user by using the techniques of knowledge representation, information retrieval, NLP and the like, and then returns the most relevant answer to the user.
The customer service system of the online intelligent question-answering system is not limited by customers or the number of customer services, can quickly and efficiently answer the questions of the users, can share knowledge, greatly reduces the investment of enterprises on human resources, and provides a knowledge storage channel for the enterprise to expand product recommendation and marketing services.
However, such customer service systems also have obvious disadvantages, and due to the limited knowledge base and the diversity of user questions, irrelevant answers are often pushed to users, and compared with manual customer service, an online automatic question-answering system has a great defect in friendliness, so that the user experience is poor.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a method and a system for analyzing emotional tendency of questions in an intelligent question-answering system, and aims to solve the problems in the prior art.
The technical purpose of the invention is realized by the following technical scheme:
an emotional tendency analysis system for questioning in an intelligent question-answering system comprises a task management module, an intelligent question-answering management module, a customer service intelligent selection module, a user characteristic analysis module and a service management module, and specifically comprises the following steps:
and the task management module is used for extracting task characteristics and classifying tasks.
The intelligent question-answering management module mainly comprises five parts, namely task analysis, task relevance analysis, question-answering information matching, question-answering satisfaction analysis and knowledge structure dynamic updating, and an intelligent question-answering module which can be continuously optimized according to user feedback is realized through the flow of the five parts.
And the customer service intelligent selection module consists of four parts, namely customer service characteristic evaluation, task-customer service characteristic matching, customer service polling decision and customer service performance evaluation, and distributes artificial customers of corresponding services through the characteristics of the questioning contents of the customers when the intelligent questioning and answering system cannot make feedback.
The system comprises a user characteristic analysis module, a real-time emotion analysis task and a query and answer analysis module, wherein the user characteristic analysis module is mainly used for extracting characteristics of two aspects of real-time emotion polarity and user key phrases aiming at a client chat record, and the real-time emotion analysis task carries out emotion analysis aiming at current and historical chat records generated in the question and answer process to obtain emotion polarity; and the extracted user key phrase is used for subsequent steps of intelligent customer service selection module, user characteristic analysis and the like.
And the service management module forms a set of complete auxiliary management module by using the knowledge structure information, the question and answer information, the user-customer service record and the task data, and provides necessary data for the whole system for analysis and training.
Further preferably, a short text sentiment classification model based on a long-short term memory network (LSTM) is constructed, a model diagram is shown in FIG. 4, the model integrates word vectors by using an LSTM neural network model to obtain sentence vectors, and information of adjacent positions is effectively integrated; and then, the sentence vectors are used for obtaining text feature vectors, so that semantic information among sentences in the text is effectively utilized.
More preferably, the real-time emotion analysis process is as follows:
1. training word vectors through a text corpus in advance to generate a word vector dictionary, taking the words as basic units forming sentences in the text emotion classification process, initializing the word vectors according to the dictionary, and initializing words which do not appear in the dictionary randomly;
2. after initializing word vectors, taking the word vectors as input of an LSTM model, and integrating the word vectors to obtain sentence vectors representing sentences;
3. taking the sentence vector as input, and obtaining a text vector integrating semantic information between sentences in the text by changing the training process of LSTM;
4. and the text vector is used for emotion analysis as text characteristic representation to obtain text emotion polarity.
Preferably, sentence vectors of context sentences are integrated to obtain chapter-level vectors, and short text emotion analysis tasks at sentence levels are converted into chapter-level tasks, so that the accuracy of text emotion classification is improved; sentence vectors with different lengths are used as input, document combination needs to generate a document vector with a fixed length as output, and simultaneously, because chat records are continuously changed and increased in instant chat, text time sequence is also a problem to be considered.
An electronic device, comprising: the intelligent question-answering system comprises a processor and a memory, wherein a computer program is stored in the memory, and is loaded and executed by the processor to realize the emotional tendency analysis method for questions in the intelligent question-answering system.
A computer-readable storage medium, wherein a computer program is stored in the storage medium, and the computer program is loaded and executed by a processor to implement the method for analyzing emotional tendency of questions in the above-mentioned intelligent question-answering system.
In summary, compared with the prior art, the beneficial effects of the invention are as follows: the intelligent question-answering system not only can accurately answer the problems of the client, but also can adjust the emotion of the user according to the instant emotion information of the user when the system is used, and the real-time emotion analysis task carries out emotion analysis on the current chat records and the historical chat records generated in the instant chat process of the client and the customer service to obtain two types of positive emotion polarities or negative emotion polarities, so that the intelligent customer service system not only can solve the problems of the client, but also can guide the client to keep or change to positive emotion according to the emotion change of the client.
Drawings
The above and other objects, features and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description read in conjunction with the accompanying drawings. Several embodiments of the present invention are illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which:
FIG. 1 is a diagram of the overall architecture of the system;
FIG. 2 is an emotional model classification model;
FIG. 3 is a corpus-to-information table;
FIG. 4 is a table of word vector training parameters;
FIG. 5 is a comparison of training speeds of the standard LSTM and the variant LSTM models;
FIG. 6 shows the real-time effect test results of the emotion analysis method.
Detailed Description
The principles and spirit of the present invention will be described with reference to several exemplary embodiments. It is understood that these embodiments are given solely for the purpose of enabling those skilled in the art to better understand and to practice the invention, and are not intended to limit the scope of the invention in any way. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art. An "embodiment" or "implementation" in the specification may mean either one embodiment or one implementation or a case of some embodiments or implementations.
As will be appreciated by one skilled in the art, embodiments of the present invention may be embodied as a system, apparatus, device, method, or computer program product. Accordingly, the present disclosure may be embodied in the form of: entirely hardware, entirely software (including firmware, resident software, micro-code, etc.), or a combination of hardware and software.
It is to be noted that any number of elements in the figures are provided by way of example and not limitation, and any nomenclature is used for distinction only and not in any limiting sense.
According to one embodiment of the invention, the emotional tendency analysis system for the questions asked in the intelligent question-answering system comprises a task management module, an intelligent question-answering management module, a customer service intelligent selection module, a user characteristic analysis module and a service management module, and specifically comprises the following modules:
the task management module has the main functions of task feature extraction and task classification, the weights of the same feature words in different types of questions are mostly different, the intelligent question answering accuracy is improved after classification, and the problems provided by customers are mostly short texts, so that feature extraction of short texts is needed before classification.
The intelligent question-answering management module mainly comprises five parts, namely task analysis, task association degree analysis, question-answering information matching, question-answering satisfaction degree analysis and knowledge structure dynamic updating, and realizes an intelligent question-answering module which can be continuously optimized according to user feedback through the five parts of processes.
The intelligent customer service selection module is formed by four parts, namely customer service characteristic evaluation, task-customer service characteristic matching, customer service polling decision and customer service performance evaluation.
The intelligent customer service system not only can solve the problems of the customers, but also can guide the customers to keep or change to positive emotions according to the emotion changes of the customers, so that the real-time emotion analysis in the instant chat of the customers belongs to one of the cores in the system; the extracted user key phrases are used as important factors in a product recommendation link and play an important role in subsequent customer service intelligent selection modules, user characteristic analysis and the like.
The business management module is used for managing the business data, aims to form a set of complete auxiliary management module by using knowledge structure information, question and answer information, user-customer service records and task data, provides necessary data for the whole system for analysis and training, and is an important intelligent component of the whole system.
The intelligent customer service system not only has knowledge, but also learns to 'look at face', the real-time emotion analysis is an important component for realizing the intellectualization of the customer service system, and the behavior of a user in instant chat can be analyzed, so that the customer can be guided to keep or change to positive emotion according to the emotion change of the customer, and therefore, the real-time emotion analysis method is provided and comprises the following steps:
constructing an LSTM-based short text sentiment classification model, wherein a model diagram is shown in FIG. 4, the model integrates word vectors by using an LSTM neural network model to obtain sentence vectors, and information of adjacent positions is effectively integrated; and then, the sentence vectors are used for obtaining text feature vectors, so that semantic information among sentences in the text is effectively utilized.
The real-time emotion analysis process is as follows:
1. training word vectors through a text corpus in advance to generate a word vector dictionary, taking the words as basic units forming sentences in the text emotion classification process, initializing the word vectors according to the dictionary, and initializing words which do not appear in the dictionary randomly;
2. after the word vector initialization work is carried out, the word vector is used as the input of an LSTM model, and a sentence vector representing a sentence is obtained by integrating the word vector;
3. taking the sentence vector as input, and obtaining a text vector integrating semantic information between sentences in the text by changing the training process of the LSTM;
4. and the text vector is used for emotion analysis as text characteristic representation to obtain text emotion polarity.
In the process of interacting with the user, the system background can automatically analyze the real-time emotion polarity of the user and display the real-time emotion polarity in the form of a change curve on the right side of the chat interface so as to assist the customer service to ask and answer.
In instant chat, some users habitually divide a paragraph containing complete semantic information into multiple sentences to be sent, the sent text content has strong correlation with the information sent by a chat user, if the information is ignored, a large error is brought to a real-time emotion analysis task, a sentence vector of a context sentence is integrated to obtain a chapter-level vector, the sentence-level short text emotion analysis task is converted into a chapter-level task, and therefore the accuracy of text emotion classification is improved. Given sentence vectors of different lengths as input, document combination needs to generate a document vector of a fixed length as output, and since chat records are continuously changed and increased in instant chat, text timing is also a problem to be considered.
According to another embodiment of the invention, a special experiment is designed for the proposed real-time emotion analysis method based on the LSTM neural network to test the effect, and the specific contents are as follows:
in order to simulate the situation of instant chat, corpora with different lengths need to be collected to respectively simulate the chat effect at different moments, so that a corpus is mined by the de-weighted and balanced Chinese emotion and serves as an experimental corpus, and each of 2000 corpora containing positive and negative comments in the fields of books, computers, hotels and the like has 12000; meanwhile, 7460 pieces of positive and negative corpora which supplement the three fields of books, computers and hotels are collected.
The corpus is divided into four subsets according to the average length and the average sentence number of a single text, and the specific situation is as shown in fig. 3. The present invention utilizes 19460 data collected and Gensimword2vec tool to perform word vector training with parameters as shown in FIG. 4, and finally obtains a dictionary containing 11450 words.
The chatting scenes at different moments are simulated by the corpora with different lengths, and the effectiveness of the method is indirectly verified by testing the effect of the method on the corpus with different text lengths. The method obtains sentence-level vector representation through word vectors, generates text-level vectors through the sentence vectors, and selects a variant LSTM model to test the effect of the method on optimizing the time sequence problem in order to accelerate the generation speed of the text vectors.
The effect of the method of the invention on training speed was tested in comparison with the standard LSTM method. In the emotion analysis in instant messaging, the speed of feature generation directly influences the performance of real-time emotion analysis, and in order to better verify that the variant LSTM model can effectively improve the training speed of the model, the invention also tests the effect on the IMBD movie comment corpus besides the corpus introduced in the preparation part before the experiment. Meanwhile, in order to accelerate the experiment progress, both tests are carried out in the GPU environment, the experiment result is shown in fig. 5, and it can be seen that although the LSTM variant selected by the method is lower than the standard LSTM in emotion classification accuracy, the loss precision is within an acceptable range, but the training speed is improved by about 40%.
As chatting progresses, the relevance of the current conversation is weaker and weaker with the initial chatting records, so the first 20 records of the current chatting message are intercepted in the system for real-time emotion analysis. Neglecting the influence of external factors such as network speed, network delay and the like, taking the time from the time when the system receives the information sent by the user from the WeChat end to the time when the single emotion analysis is completed as an evaluation standard, and respectively carrying out more than 300 conversations (each time the WeChat is used for sending a message is marked as a conversation), the result is shown in FIG. 6, and it is easy to see that five testers are used within one second for emotion analysis once (the system receives a WeChat end message), so that the real-time emotion analysis method provided by the invention can basically achieve the real-time effect in the customer service system. Meanwhile, the results show that the real-time emotion analysis time consumption tends to be average value at different moments for different testees, so that the method has universality in time.
According to another embodiment of the present invention, an electronic device is provided, which includes a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement the method for analyzing emotional tendency of questions in the above-mentioned intelligent question-answering system.
According to still another embodiment of the present invention, there is provided a computer-readable storage medium having a computer program stored therein, the computer program being loaded and executed by a processor to implement the method for analyzing emotional tendency of questions in the above-mentioned intelligent question-and-answer system.
The above description is intended to be illustrative of the present invention and not to limit the scope of the invention, which is defined by the claims appended hereto.
Claims (6)
1. An emotional tendency analysis system for questions in an intelligent question-answering system is characterized by comprising a task management module, an intelligent question-answering management module, a customer service intelligent selection module, a user characteristic analysis module and a service management module, and specifically comprising:
the task management module is used for extracting task characteristics and classifying tasks;
the intelligent question-answering management module mainly comprises five parts, namely task analysis, task relevance analysis, question-answering information matching, question-answering satisfaction analysis and knowledge structure dynamic updating, and realizes an intelligent question-answering module which can be continuously optimized according to user feedback through the flow of the five parts;
the customer service intelligent selection module consists of four parts, namely customer service characteristic evaluation, task-customer service characteristic matching, customer service polling decision and customer service performance evaluation, and distributes artificial customers of corresponding services through the characteristics of the questioning content of the customers when the intelligent questioning and answering system cannot make feedback;
the user characteristic analysis module is mainly used for extracting characteristics of two aspects of real-time emotion polarity and user key phrases aiming at the client chat records, and the real-time emotion analysis task carries out emotion analysis on current and historical chat records generated in the question and answer process to obtain emotion polarity; the extracted key phrases of the user are used for subsequent steps of intelligent customer service selection module, user characteristic analysis and the like;
and the service management module forms a set of complete auxiliary management module by the knowledge structure information, the question and answer information, the user-customer service record and the task data, and provides necessary data for the whole system for analysis and training.
2. The emotional tendency analysis method for questions in the intelligent question-answering system according to claim 1, wherein the emotional tendency analysis is based on a long-short term memory network to construct a short text emotional classification model, the model integrates word vectors to obtain sentence vectors by using a long-short term memory neural network model, and information of adjacent positions is effectively integrated; and then, the sentence vectors are used for obtaining text feature vectors, so that semantic information among sentences in the text is effectively utilized.
3. The method for analyzing emotional tendency of questions in the intelligent question-answering system according to claim 2, wherein the real-time emotional analysis process is as follows:
training word vectors through a text corpus in advance to generate a word vector dictionary, taking the words as basic units forming sentences in the text emotion classification process, initializing the word vectors according to the dictionary, and initializing words which do not appear in the dictionary randomly;
after initializing word vectors, taking the word vectors as input of a long-term and short-term memory network model, and integrating the word vectors to obtain sentence vectors representing sentences;
taking the sentence vector as input, and obtaining a text vector integrating semantic information between sentences in the text by changing the training process of LSTM;
the text vector is used for emotion analysis as text feature representation to obtain text emotion polarity.
4. The method for analyzing emotional tendency of questions in an intelligent question-answering system according to claim 1, wherein the user key phrases are extracted, sentence vectors of context sentences are integrated to obtain chapter-level vectors, and short text emotion analysis tasks at sentence levels are converted into chapter-level tasks, so that the accuracy of text emotion classification is improved; sentence vectors with different lengths are used as input, document combination needs to generate a document vector with a fixed length as output, and simultaneously, because chat records are continuously changed and increased in instant chat, text time sequence is also a problem to be considered.
5. An electronic device, comprising: comprising a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded by the processor and executed to realize the emotional tendency analysis method of the question in the intelligent question-answering system according to any one of claims 1 to 4.
6. A computer-readable storage medium, wherein a computer program is stored in the storage medium, and the computer program is loaded and executed by a processor to implement the method for analyzing emotional tendency of a question in the intelligent question-and-answer system according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210944030.2A CN115186678A (en) | 2022-08-08 | 2022-08-08 | Emotional tendency analysis method and system for question in intelligent question-answering system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210944030.2A CN115186678A (en) | 2022-08-08 | 2022-08-08 | Emotional tendency analysis method and system for question in intelligent question-answering system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115186678A true CN115186678A (en) | 2022-10-14 |
Family
ID=83523431
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210944030.2A Pending CN115186678A (en) | 2022-08-08 | 2022-08-08 | Emotional tendency analysis method and system for question in intelligent question-answering system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115186678A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117726301A (en) * | 2023-12-26 | 2024-03-19 | 重庆不贰科技(集团)有限公司 | Intelligent decision-making system based on production line management and Chat combined model |
-
2022
- 2022-08-08 CN CN202210944030.2A patent/CN115186678A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117726301A (en) * | 2023-12-26 | 2024-03-19 | 重庆不贰科技(集团)有限公司 | Intelligent decision-making system based on production line management and Chat combined model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Alva-Manchego et al. | ASSET: A dataset for tuning and evaluation of sentence simplification models with multiple rewriting transformations | |
| CN110442718B (en) | Statement processing method and device, server and storage medium | |
| CN110096698B (en) | Topic-considered machine reading understanding model generation method and system | |
| Smith | Computer–mediated negotiated interaction: An expanded model | |
| Mathew et al. | NLP-based personal learning assistant for school education | |
| WO2024011813A1 (en) | Text expansion method and apparatus, device, and medium | |
| Rankin | Variational learning in L2: The transfer of L1 syntax and parsing strategies in the interpretation of wh-questions by L1 German learners of L2 English | |
| Lertpiya et al. | A preliminary study on fundamental Thai NLP tasks for user-generated web content | |
| Hijjawi et al. | Arabic language challenges in text based conversational agents compared to the English language | |
| CN116821290A (en) | Multitasking dialogue-oriented large language model training method and interaction method | |
| CN115186678A (en) | Emotional tendency analysis method and system for question in intelligent question-answering system | |
| Nagender et al. | Whatsapp auto responder using natural language processing and AI | |
| Zhu et al. | YUN111@ Dravidian-CodeMix-FIRE2020: Sentiment Analysis of Dravidian Code Mixed Text. | |
| CN115017876A (en) | Method and terminal for automatically generating emotion text | |
| Liang et al. | Intelligent chat robot in digital campus based on deep learning | |
| KR102330970B1 (en) | Assessment system and method for education based on artificial intelligence | |
| Vajjala | Machine learning and applied linguistics | |
| CN113850087A (en) | Translation scoring method and related equipment thereof | |
| Srinivasan | Chatbot application for tourism using natural language tool kit | |
| Vanel et al. | A Survey of Socio-Emotional Strategies for Generation-Based Conversational Agents. | |
| O’Shea et al. | A novel approach for constructing conversational agents using sentence similarity measures | |
| CN113515932A (en) | Method, device, equipment and storage medium for processing question and answer information | |
| Liu et al. | Analysis of the influencing factors of online classes satisfaction based on text mining Take MOOC platform art education online classes as an example | |
| Pan et al. | How to Build Robust FAQ Chatbot with Controllable Question Generator? | |
| Provotar et al. | Features and problems of virtual communication by means of chatbots |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |