CN110377713B - Method for improving context of question-answering system based on probability transition - Google Patents
Method for improving context of question-answering system based on probability transition Download PDFInfo
- Publication number
- CN110377713B CN110377713B CN201910641706.9A CN201910641706A CN110377713B CN 110377713 B CN110377713 B CN 110377713B CN 201910641706 A CN201910641706 A CN 201910641706A CN 110377713 B CN110377713 B CN 110377713B
- Authority
- CN
- China
- Prior art keywords
- data
- probability transition
- probability
- context
- transition matrix
- 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.)
- Active
Links
Classifications
-
- 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
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3346—Query execution using probabilistic model
-
- 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
- G06F40/205—Parsing
- G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0281—Customer communication at a business location, e.g. providing product or service information, consulting
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Finance (AREA)
- Mathematical Physics (AREA)
- Strategic Management (AREA)
- Artificial Intelligence (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Probability & Statistics with Applications (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Human Computer Interaction (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Machine Translation (AREA)
Abstract
A method for improving the context of a question-answering system based on probability transition belongs to the technical field of data processing methods, and adopts a classification algorithm and a probability transition matrix A to process user problem data, (1) preset labeling data of the system; (2) Receiving a user problem, and preprocessing to obtain processing data; (3) Training the labeling data through a classification algorithm to obtain an intention classification model; then, the labeling data are transmitted to a probability transition matrix A for training, and an initialized probability transition matrix A is obtained; (4) Predicting the processing data to obtain the distribution P of the prediction labels; the invention provides a context-combined intention recognition method, by which labeling data with context scenes can be unnecessary to prepare, and the labor cost is saved; through the self-learning capability of the probability transition matrix A, the accuracy of the whole system is higher and higher along with the change of the service time.
Description
Technical Field
The invention belongs to the technical field of data processing methods, and particularly relates to a method for improving the context of a question-answering system based on probability transfer.
Background
In the e-commerce field, when a user (i.e., a buyer) makes online shopping, a consultation behavior is generated to customer service. In many automated question-answering systems, what is needed is to identify and classify the intent of the buyer. It is common practice to classify intent recognition as short text, but this practice breaks the impact of the user's context (historical dialog) on intent classification. For example, the buyer speaks "160". The buyer may either express a confirmation price or be providing his height or weight. This requires confirmation of specific intent from the above. There are also some schemes to combine context intent recognition, such as inputting 5 questions of buyers in succession as input, sorting with hierarchical intent. Or the tags above are taken as a feature into the current sentence for calculation. However, since this approach requires continuous chat recording by the user and requires the labeling personnel to pay attention to the context of each piece of data when labeling, additional effort is incurred in labeling and labeling errors are also easily caused. Furthermore, another class of drawbacks of this approach is that the samples are highly unbalanced, requiring manual replenishment of the data. Another type of solution is to use a rule, i.e. to manually specify the rules that the context appears in, and it is obvious that this solution is time-consuming and labor-consuming, and it is difficult to guarantee that all the possibilities are enumerated.
Disclosure of Invention
The present invention aims to overcome the above-mentioned drawbacks and disadvantages and to provide a method for improving the context of a question-answering system based on probability transitions.
In order to solve the technical problems, the following technical scheme is adopted:
the method for improving the context of the question-answering system based on the probability transition realizes the improvement of the question-answering system by combining a classification algorithm and a probability transition matrix A, and comprises the following specific steps:
(1) Predicting a series of data for the system, and manually calibrating the data to obtain calibration data;
(2) Receiving user problems, preprocessing the user problems to obtain processing data, and facilitating subsequent link processing;
(3) The labeling data is processed, and the processing content is as follows:
training the obtained labeling data through a TEXT CNN model to obtain an intention classification model;
(3-2) inputting the labeling data into a probability transition matrix for training to obtain an initialized probability transition matrix A;
(4) Predicting the processing data in the step (2) through a classification algorithm to obtain the distribution P of the predicted tags;
(5) A series of calculations are performed on the distribution P of the predictive labels to screen out missing information Q i ，i＝1,2,3...n；
(6) Processing the processing data which is a complete session process in the step (2) by combining the probability transition matrix A obtained in the step (3-2) to obtain accurate missing sentences and corresponding context contents;
furthermore, the classification algorithms are TEXT CNN, LSTM, BERT and SVM, and particularly the classification algorithm with probability distribution of the prediction result is applicable to the system.
Further, the series of algorithms in the step (5) specifically includes:
(5-1) calculating an average number M of the distribution P by an average number formula,
wherein P is 1 、P 2 ...P i Representing a specific numerical value, i representing the number of the set of data;
(5-2) calculating by a variance formula, screening out the characteristics,
wherein i represents the number of data, M is an average number, s 2 Representing the variance, when the variance s 2 The smaller the value, the more difficult it is to judge the intention expressed for some of the processed data.
Further, the specific contents calculated in the step (6) in combination with the probability transition matrix a are as follows:
(6-1) combining the processed data, we have n dialogues for the sentence Q determined to be missing in step (4) i With a distribution P i Combining a random probability transition matrix A initialized by the system, and constructing an objective function: f= ||q i-1 -AQ i And the n pieces of processing data are data sets with complete dialogue scenes, so that the probability transition matrix A learns the prediction result of the current processing data, and the missing sentences and the corresponding context content are determined.
By adopting the scheme, the method has the following beneficial effects:
(1) The invention provides a context-combined intention recognition method, by which labeling data with context scenes can be unnecessary to prepare, and the labor cost is saved.
(2) Through the self-learning capability of the probability transition matrix A, the accuracy of the whole system is higher and higher along with the change of the service time.
Drawings
Fig. 1 is a flow chart of the entire system.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
The method for improving the context of the question-answering system based on the probability transition realizes the improvement of the question-answering system by combining a classification algorithm and a probability transition matrix A, and comprises the following specific steps:
(4) Predicting a series of data for the system, and manually calibrating the data to obtain calibration data;
(5) Receiving user problems, preprocessing the user problems to obtain processing data, and facilitating subsequent link processing;
(6) The labeling data is processed, and the processing content is as follows:
training the obtained labeling data through a TEXT CNN model to obtain an intention classification model;
(3-2) inputting the labeling data into a probability transition matrix for training to obtain an initialized probability transition matrix A;
(4) Predicting the processing data in the step (2) through a classification algorithm to obtain the distribution P of the predicted tags;
(5) A series of calculations are performed on the distribution P of the predictive labels to screen out missing information Q i I=1, 2, 3..n; the series of algorithms are specifically as follows:
(5-1) calculating an average number M of the distribution P by an average number formula,
wherein P is 1 、P 2 ...P i Representing a specific numerical value, i representing the number of the set of data;
(5-2) calculating by a variance formula, screening out the characteristics,
wherein i represents the number of data, M is an average number, s 2 Representing the variance, when the variance s 2 The smaller the value, the more difficult it is to judge the intention expressed for some of the processed data.
(6) Processing the processing data which is a complete session process in the step (2) by combining the probability transition matrix A obtained in the step (3-2) to obtain accurate missing sentences and corresponding context contents, wherein the specific contents calculated by combining the probability transition matrix A are as follows:
(6-1) combining the processed data, we have n dialogues for the sentence Q determined to be missing in step (4) i With a distribution P i Combining a random probability transition matrix A initialized by the system, and constructing an objective function: f= ||q i-1 -AQ i And (3) enabling the probability transition matrix A to learn the prediction result of the current processing data, so as to determine the missing statement and the corresponding context content.
Preferably, the classification algorithms are TEXT CNN, LSTM, BERT and SVM, and particularly, the classification algorithm with probability distribution of the prediction result is applicable to the system.
Preferably, the n pieces of processing data in the step (6-1) are data sets with complete dialogue scenes.
The working principle of the system is as follows: as shown in fig. 1, first, the TEXT CNN trains labeling data to obtain an intention classification model, then, the TEXT CNN predicts an intended label for a trained TEXT CNN for a received user problem, if the probability distribution of the predicted label is smooth, the predicted label is obtained, if the probability distribution of the predicted label is not smooth, unlabeled data (i.e., user problem data in a complete session) is trained and learned by an initialized probability transition matrix a, and a label probability distribution P obtained by the TEXT CNN for each sentence of user problem is combined, the obtained label probability distribution P and the probability transition matrix a are multiplied to obtain a new probability distribution, and an intended label corresponding to a new probability distribution top1 is output, thereby completing the whole system process.
The invention has been described in terms of embodiments, and the device can be modified and improved without departing from the principles of the invention. It should be noted that all technical solutions obtained by equivalent substitution or equivalent transformation fall within the protection scope of the present invention.
Claims (4)
1. A method for improving the context of a question-answering system based on probability transition adopts a classification algorithm and a probability transition matrix A to process user question data, and is characterized in that:
the specific processing steps are as follows:
(1) Presetting annotation data for a system;
(2) Receiving a user problem, and preprocessing to obtain processing data;
(3) Training the labeling data through a classification algorithm to obtain an intention classification model; then, the labeling data are transmitted to a probability transition matrix A for training, and an initialized probability transition matrix A is obtained;
(4) Predicting the processing data to obtain the distribution P of the prediction labels;
(5) A series of calculations are performed by predicting the distribution P of the tags, screening the missing information Q i ，i＝1，2，3...n；
(6) For a data set in a complete session process in the processed data, calculating by combining the initialized probability transfer matrix A to obtain accurate missing sentences and corresponding context contents; wherein, a series of calculations in the step (5) are specifically:
(5-1) calculating an average number M of the distribution P by an average number formula,wherein, P1, P2..pi represents a specific numerical value, i represents the number of the set of data;
(5-2) calculating by a variance formula,
wherein i represents the number of data, M is an average number, S 2 Representing the variance, when the variance S 2 The smaller the value, the harder it is to judge the intention expressed by some of the processed data;
the content calculated by combining the probability transition matrix A in the step (6) is as follows:
(6-1) combining the processed data, having n dialogues, with respect to the information Q selected as missing in step (5) i With a distribution P i Combining a random probability transition matrix A initialized by the system, and constructing an objective function: f= ||q i-1 -AQ i And (3) enabling the probability transition matrix A to learn the prediction result of the current processing data so as to determine the missing statement and the corresponding context.
2. A method for improving the context of a question-answering system based on probability transitions as set forth in claim 1, wherein: the labeling data is a series of problems of manual calibration.
3. A method for improving the context of a question-answering system based on probability transitions as set forth in claim 1, wherein: the classification algorithm is a classification algorithm with TEXT CNN, LSTM, BERT and SVM, and particularly has probability distribution of a prediction result, and the classification algorithm is applicable to the system.
4. A method for improving the context of a question-answering system based on probability transitions as set forth in claim 1, wherein: the n pieces of processing data combined in the step (6-1) are data sets with complete dialogue scenes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910641706.9A CN110377713B (en) | 2019-07-16 | 2019-07-16 | Method for improving context of question-answering system based on probability transition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910641706.9A CN110377713B (en) | 2019-07-16 | 2019-07-16 | Method for improving context of question-answering system based on probability transition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110377713A CN110377713A (en) | 2019-10-25 |
| CN110377713B true CN110377713B (en) | 2023-09-15 |
Family
ID=68253502
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910641706.9A Active CN110377713B (en) | 2019-07-16 | 2019-07-16 | Method for improving context of question-answering system based on probability transition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110377713B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111694952A (en) * | 2020-04-16 | 2020-09-22 | 国家计算机网络与信息安全管理中心 | Big data analysis model system based on microblog and implementation method thereof |
| CN115018656B (en) * | 2022-08-08 | 2023-01-10 | 太平金融科技服务(上海)有限公司深圳分公司 | Risk identification method, and training method, device and equipment of risk identification model |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106649694A (en) * | 2016-12-19 | 2017-05-10 | 北京云知声信息技术有限公司 | Method and device for identifying user's intention in voice interaction |
| CN107679234A (en) * | 2017-10-24 | 2018-02-09 | 上海携程国际旅行社有限公司 | Customer service information providing method, device, electronic equipment, storage medium |
| CN108829662A (en) * | 2018-05-10 | 2018-11-16 | 浙江大学 | A kind of conversation activity recognition methods and system based on condition random field structuring attention network |
| CN108897896A (en) * | 2018-07-13 | 2018-11-27 | 深圳追科技有限公司 | Keyword abstraction method based on intensified learning |
| WO2022095573A1 (en) * | 2020-11-09 | 2022-05-12 | 西安交通大学 | Community question answering website answer sorting method and system combined with active learning |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10331659B2 (en) * | 2016-09-06 | 2019-06-25 | International Business Machines Corporation | Automatic detection and cleansing of erroneous concepts in an aggregated knowledge base |
| KR20200123584A (en) * | 2019-04-22 | 2020-10-30 | 한국전자통신연구원 | Apparatus and method for predicting error of annotation |
-
2019
- 2019-07-16 CN CN201910641706.9A patent/CN110377713B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106649694A (en) * | 2016-12-19 | 2017-05-10 | 北京云知声信息技术有限公司 | Method and device for identifying user's intention in voice interaction |
| CN107679234A (en) * | 2017-10-24 | 2018-02-09 | 上海携程国际旅行社有限公司 | Customer service information providing method, device, electronic equipment, storage medium |
| CN108829662A (en) * | 2018-05-10 | 2018-11-16 | 浙江大学 | A kind of conversation activity recognition methods and system based on condition random field structuring attention network |
| CN108897896A (en) * | 2018-07-13 | 2018-11-27 | 深圳追科技有限公司 | Keyword abstraction method based on intensified learning |
| WO2022095573A1 (en) * | 2020-11-09 | 2022-05-12 | 西安交通大学 | Community question answering website answer sorting method and system combined with active learning |
Non-Patent Citations (1)
| Title |
|---|
| 周小强等.交互式问答的关系结构体系及标注.《中文信息学报》.2018,(第05期), * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110377713A (en) | 2019-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20190005027A1 (en) | System and Method For Domain-Independent Aspect Level Sentiment Detection | |
| US20200210526A1 (en) | Document classification using attention networks | |
| CN106611375A (en) | Text analysis-based credit risk assessment method and apparatus | |
| Ilmania et al. | Aspect detection and sentiment classification using deep neural network for Indonesian aspect-based sentiment analysis | |
| CN110377713B (en) | Method for improving context of question-answering system based on probability transition | |
| US20210117619A1 (en) | Cyberbullying detection method and system | |
| CN111966888B (en) | Aspect class-based interpretability recommendation method and system for fusing external data | |
| CN111738807B (en) | Method, computing device, and computer storage medium for recommending target objects | |
| CN111538841B (en) | Comment emotion analysis method, device and system based on knowledge mutual distillation | |
| CN113254592A (en) | Comment aspect detection method and system of multi-level attention model based on door mechanism | |
| Rao et al. | A first look: Towards explainable textvqa models via visual and textual explanations | |
| CN110704803A (en) | Target object evaluation value calculation method and device, storage medium and electronic device | |
| CN117390141B (en) | Agricultural socialization service quality user evaluation data analysis method | |
| CN114266241A (en) | Comment usefulness prediction method, device and medium based on text and emotion polarity | |
| CN111626331B (en) | Automatic industry classification device and working method thereof | |
| CN108460049A (en) | A kind of method and system of determining information category | |
| CN117112775A (en) | Technique for automatically filling in an input form to generate a list | |
| CN112313679A (en) | Information processing apparatus, information processing method, and program | |
| US20220404778A1 (en) | Intellectual quality management method, electronic device and computer readable storage medium | |
| Tran et al. | Efficient cnn models for beer bottle cap classification problem | |
| Wu et al. | The Impact of news sentiment on the stock market fluctuation: the case of selected energy sector | |
| CN113919906A (en) | Commodity comment data pushing method and device and storage medium | |
| CN111553726A (en) | HMM-based (hidden Markov model) -based system and method for predicting bill swiping | |
| CN114036288A (en) | Relation extraction method based on reinforcement learning | |
| KR20220118703A (en) | Machine Learning based Online Shopping Review Sentiment Prediction System and Method |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20230331 Address after: 104058, No. 2-10, No. 311 Huangpu Avenue Middle, Tianhe District, Guangzhou City, Guangdong Province, 510000 Applicant after: Guangzhou Tanyu Technology Co.,Ltd. Address before: 601-5, 1382 Wenyi West Road, Cangqian street, Yuhang District, Hangzhou City, Zhejiang Province, 310012 Applicant before: Hangzhou Weier Network Technology Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |