CN108345585A - A kind of automatic question-answering method based on deep learning - Google Patents
A kind of automatic question-answering method based on deep learning Download PDFInfo
- Publication number
- CN108345585A CN108345585A CN201810026979.8A CN201810026979A CN108345585A CN 108345585 A CN108345585 A CN 108345585A CN 201810026979 A CN201810026979 A CN 201810026979A CN 108345585 A CN108345585 A CN 108345585A
- Authority
- CN
- China
- Prior art keywords
- question
- answer
- vector
- word
- text
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- 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
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Databases & Information Systems (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 a kind of automatic question-answering methods based on deep learning, it is intended to provide based on algorithm, full automatic question answering scheme to the user.The present invention uses the question and answer crawled from website to that as data source, can answer form complex problem.The present invention is on the basis of traditional Similar Problems retrieval, the content of text of problem is expressed as vector using BOW models, TFIDF models and Word2Vec models, Similar Problems are resequenced and screened by calculating the similarity between vector, semantic knowledge can be introduced, it solves the problems, such as the semantic gap in traditional problem retrieving, improves the validity of candidate answers.In addition, based on deep learning, the neural network model that the present invention is obtained using training carries out matching marking to problem and candidate answers, the high-rise matching characteristic between question and answer can be extracted automatically and automatically to the answer gone wrong, the accuracy of automatically request-answering system can be promoted, manual intervention is reduced simultaneously, reduces system development costs.
Description
Technical field
The present invention relates to the information retrieval in natural language processing field, document representation method, text similarities to calculate, certainly
A kind of dynamic question and answer field, and in particular to automatic question-answering method based on deep learning.
Background technology
With the rapid development of Internet, occurring a large amount of electronic document on network, user is in one problem of lookup
When answer, traditional information retrieval can not directly give answer, can only provide thousands of web page interlinkage, therefore can be automatic
The automatic question answering technology for providing optimum answer has been to be concerned by more and more people.The research of automatic question-answering method is broadly divided into template
The method of matched method, method for information retrieval and deep learning.
Method based on template matches needs a large amount of template of manual compiling, and cost is very high, and to the adaptation of new data
Property it is bad, once but successful match, the quality of answer are relatively high, the expert system of early stage mainly uses the side of template matches
Formula.
Mode based on information retrieval mainly studies Text similarity computing method, and converting problem to mathematical computations asks
Topic calculates problem most like therewith to customer problem and obtains candidate answers, then design answer feature by design feature
Choose optimum answer.When problem is retrieved, the similarity calculating methods such as most-often used BM25, TFIDF based on character registration,
But there are problems that semantic gap, the semanteme of text can not be got a real idea of.
Pass through a large amount of text datas, the automatic high-level semantics features for capturing text, in recent years based on the method for deep learning
It quickly grows, a large amount of nerve nets using convolutional neural networks CNN, shot and long term memory network LSTM, Recognition with Recurrent Neural Network RNN
Network is proposed that modelling effect has obtained large increase successively.
Invention content
The purpose of the present invention is using document representation method to resequence the Similar Problems of selection, similar ask is chosen
The answer of topic carries out matching marking as candidate answers, using neural network model to customer problem and candidate answers, to
To optimum answer.
To achieve the above object, the present invention adopts the following technical scheme that:
A kind of automatic question-answering method based on deep learning, includes the following steps:
1) the question and answer data that related field is crawled from internet obtain question and answer pair by surface cleaning, and relationship is arrived in storage
In type database, and build full article retrieval.
2) Chinese word segmentation tool is used, all question and answer data are segmented, including addition User Defined dictionary, text
It segments and stop words is gone to handle.
3) Word2Vec models are generated to the question and answer data training after participle in step 2), is obtained using Word2Vec models
The related term of the term vector of each word and each word.
4) BOW bag of words are built to the question and answer data after participle in step 2), uses BOW models the table of One-hot
The mode of showing obtains the BOW vectors of each word, meanwhile, TFIDF models are built to the data question and answer data after participle, are obtained each
Word corresponding TFIDF values in each question text.
5) use TFIDF, BOW vector and Word2Vec related terms to all problems structure text representation vector.
6) to user the problem of carries out full-text search and obtains Similar Problems, then carries out Similar Problems based in step 5)
The cosine similarity of text representation vector calculates, and after rearrangement, obtains the candidate answers of problem.
7) question and answer Matching Model of the training based on neural network, gives a mark to the matching degree of problem and candidate answers.
8) it when user puts question to, is segmented using step 2), using the text representation vector of step 5) Construct question, is used
Step 6) obtains candidate answers, and the highest candidate answers of matching score are obtained as final answer using step 7).
More specifically, the training parameter of the used Word2Vec models of term vector structure of question and answer data is arranged
For:Using Skip-gram algorithms, output term vector dimension is 200, and training window size is 5, and the minimum frequency of occurrences of word is 5, is adopted
Sample threshold value is 10-4。
Using TFIDF, BOW vector and Word2Vec related terms all problems are built with the specific method of text representation vector
For:First, it to each word in each problem, is multiplied by TFIDF using its BOW vector and is worth to vectorial W1, meanwhile, it uses
Word2Vec obtains maximally related 10 words of the word, and the BOW vectors of 10 correlation words are multiplied by inlet coefficient a=respectively
It 0.1 and sums and obtains W2, sum W1 and W2 to obtain term vector W3.Then, for each problem, using each word word to
Amount W3 sums, and obtains text representation vector.
The problem of being proposed for user first uses full-text search to obtain 500 Similar Problems, then using based on text table
The Similarity measures shown obtain the answer of the problem of 100 before ranking as candidate answers, finally use asking based on deep learning
It answers the Matching Model acquisition highest answer of matching score and returns to user.
The Chinese word segmentation tool uses the Jieba tools of Python;Word2Vec models use Python's
Gensim tools;Neural network model structure uses Tensorflow tools
Neural network structure for question and answer Matching Model is defined as follows shown in table.
Network depth | Title | Convolution kernel size/step-length | Convolution nuclear volume | Full articulamentum neuron number |
1 | Embedding | / | / | / |
2 | BiLSTM-1 | / | / | / |
3 | Conv-1 | 3x3/1 | 128 | / |
4 | Conv-2 | 3x3/1 | 256 | / |
5 | Conv-3 | 3x3/1 | 256 | / |
6 | BiLSTM-2 | / | / | / |
7 | FC-1 | / | / | 4096 |
8 | FC-2 | / | / | 4096 |
The present invention having the beneficial effect that compared with the existing technology:
1) it when the content of text to problem builds text representation vector, introduces related term and carries out joint expression, gained
The text representation vector arrived has semantic knowledge, can solve the problems, such as semantic gap.
2) on the basis of traditional Similar Problems retrieval, the step that reorders with semantic knowledge, Neng Gouti are added to
The degree of correlation of high Similar Problems, and reduce the range of candidate answers collection, it is possible to reduce the calculation amount of subsequent processing, raising are asked automatically
Answer the accuracy and efficiency of system;
3) due to the use of question and answer to as data source, therefore the problem of complex form can be answered, and with data set
Expansion, model performance can get a promotion.
4) question and answer Matching Model is had trained using the method based on deep learning, problem and candidate answers can be extracted automatically
Between matching characteristic and provide answer automatically, without manual intervention, reduce system development costs.
5) institute's procedures set forth of the present invention has generality, is applicable to the automatically request-answering system of structure different field.
Description of the drawings
Fig. 1 is a kind of overall flow figure of the automatic question-answering method based on deep learning;
Fig. 2 is that the exemplary plot of the Word2Vec term vectors generated is utilized in embodiment.
Fig. 3 is system effect figure in embodiment.
Specific implementation mode
Below in conjunction with specific example and attached drawing, invention is further described in detail.
As shown in Figure 1, bold portion is the structure stage of system, dotted portion corresponds to the service stage of user.
The system structure stage is divided into two pieces, is that deep learning is retrieved and be based on to the Similar Problems based on text representation respectively
Question and answer Matching Model, this two pieces of steps are in no particular order.
The structure of Similar Problems retrieval based on text representation is described as:
1) reconciliation phase separation is crawled from internet using the Requests tools of Python and BeautifulSoup tools to close
The question and answer data in field.Surface cleaning is carried out to question and answer data first, including deletes unusual character, the format letter in question and answer data
Breath, attaching metadata, limitation text size and code conversion and etc..Then the SimHash tools of Python is used to carry out
Data deduplication.Finally by obtained question and answer to storing into MySQL database, and examined using Elasticsearch structure full text
Rope service.
2) Jieba Chinese word segmentation tools are used, the User Defined dictionary of acquiescence is added, configure stop words dictionary and are enabled
Stop words function is gone, Chinese word segmentation is carried out to all question and answer data.
3) the Gensim tools of Python is used to generate Word2Vec models to the question and answer data training after participle in step 2),
It is 200 that Skip-gram algorithms, output term vector dimension are used in trained process, and training window size is 5, the minimum appearance of word
Frequency is 5, sampling threshold 10-4.200 dimension term vectors of each word, Yi Jiyu are exported using obtained Word2Vec models
Maximally related 10 words of each word.
4) dictionary is built to the question and answer data after participle in step 2), the word that frequency of occurrence is less than 50 is ignored when constructing dictionary
Language establishes the BOW vectors using One-hot representations using dictionary construction BOW bag of words, and for each word.
5) it uses TFIDF, BOW vector and Word2Vec related terms to all problems structure text representation vector, specifically does
Method is:To each word in each problem, it is multiplied by TFIDF using its BOW vector and is worth to vectorial W1, meanwhile, it uses
Word2Vec obtains maximally related 10 words of the word, and the BOW vectors of 10 correlation words are multiplied by inlet coefficient a=respectively
It 0.1 and sums and obtains W2, sum W1 and W2 to obtain term vector W3.For each problem, by the term vector W3 of each word into
Row summation is to get vectorial to the text representation of problem.
6) full-text search is carried out using Elasticsearch to customer problem and obtains 500 Similar Problems, to Similar Problems
Text representation vector is converted to by step 5), by the cosine phase for calculating separately the vector between Similar Problems and customer problem
The corresponding answer of the problem of like degree, carrying out sequencing of similarity, choosing 100 before ranking is as candidate answers.
The structure of question and answer Matching Model based on deep learning is described as:
1) Q & A database and full article retrieval is utilized to build question and answer data set:It is corresponding to answer to each problem
Case is positive sample, randomly selects other 299 answers as negative sample using Elasticsearch, constitutes question and answer data set.
2) question and answer data set is used to train the question and answer Matching Model based on neural network, the matching between question and answer
Degree is given a mark.
Complete the structure of entire model above, i.e. bold portion flow in Fig. 1.
In the service stage of user, 1 dotted portion flow of corresponding diagram can be divided mainly into following steps:
1) the problem of being proposed to user obtains similar the asking of 500 problems compositions using Elasticsearch full-text searches
Topic collection.
1) the problem of proposing user carries out Chinese word segmentation using Jieba participle tools, and Jieba still needs to be added user
Custom Dictionaries go the operations such as stop words processing.
2) BOW vectors, TFIDF the and Word2Vec model constructions customer problem obtained using the structure stage is asked to similar
Inscribe the text representation vector of collection.
3) it by calculating the cosine similarity between vector, resequences, selects to customer problem and candidate question set
The corresponding answer of the problem of taking 100 before ranking is as candidate answers.
4) the question and answer Matching Model obtained using the structure stage is carried out matching between customer problem and candidate answers and be beaten
Point, the answer for choosing highest scoring returns to user.
Embodiment
When user proposes problem " ginger can hair growth ", the process flow of system is described as follows:
1) similar 500 problems, institute are chosen using Elasticsearch full article retrievals from the database of system
500 obtained problems are all the problem of enquirement with user contain common word.
2) it uses Jieba participle tools to carry out Chinese word segmentation, User Defined dictionary is set in Jieba tools, enables and goes
Stop words, the problem after participle are " ginger | treatment | alopecia ".
3) TFIDF of customer problem and each word in Similar Problems is calculated.
4) BOW vectors, TFIDF the and Word2Vec models for utilizing the structure stage, to customer problem and Similar Problems structure
Build the text representation vector of 200 dimensions, wherein the term vector obtained according to Word2Vec models is as shown in Fig. 2, each word is turned
It is changed to the vector of 200 dimensions.
5) by calculate text representation vector between cosine similarity, choose 500 Similar Problems in customer problem
100 most like problems, and obtain the corresponding answer of this 100 problems and constitute candidate answers collection.
6) the question and answer Matching Model based on deep learning generated using the structure stage to problem and each candidate answers into
Row matching marking chooses the highest answer of matching score and returns to user, i.e. answer is that " hello, and ginger cannot cure seborrheica
Dermatitis, it is proposed that oral vitamin b, cystine, zinc gluconate, externally used compound ketoconazole shampoo clean scalp curing.", such as
Shown in Fig. 3.
Claims (5)
1. a kind of automatic question-answering method based on deep learning, it is characterised in that include the following steps:
1) the question and answer data that related field is crawled from internet obtain question and answer pair by surface cleaning, and relationship type number is arrived in storage
According in library, and build full article retrieval;
2) Chinese word segmentation tool is used, all question and answer data are segmented, including addition User Defined dictionary, text participle
With go stop words to handle;
3) Word2Vec models are generated to the question and answer data training after participle in step 2), is obtained using Word2Vec models each
The related term of the term vector of word and each word;
4) BOW bag of words are built to the question and answer data after participle in step 2), uses BOW models the expression side of One-hot
Formula obtains the BOW vectors of each word, meanwhile, TFIDF models are built to the data question and answer data after participle, obtain each word
The corresponding TFIDF values in each question text;
5) use TFIDF, BOW vector and Word2Vec related terms to all problems structure text representation vector;
6) to user the problem of carries out full-text search and obtains Similar Problems, then carries out Similar Problems based on text in step 5)
It indicates that the cosine similarity of vector calculates, after rearrangement, obtains the candidate answers of problem;
7) question and answer Matching Model of the training based on neural network, gives a mark to the matching degree of problem and candidate answers;
8) it when user puts question to, is segmented using step 2), using the text representation vector of step 5) Construct question, uses step
6) candidate answers are obtained, the highest candidate answers of matching score are obtained as final answer using step 7).
2. a kind of automatic question-answering method based on deep learning according to claim 1, it is characterised in that the Chinese
Participle tool uses the Jieba tools of Python;Word2Vec models use the Gensim tools of Python;Neural network model
Structure uses Tensorflow tools.
3. a kind of automatic question-answering method based on deep learning according to claim 1, it is characterised in that the step
3) parameter of training Word2Vec models is specifically configured in:
Using Skip-gram algorithms, output term vector dimension is 200, and training window size is 5, and the minimum frequency of occurrences of word is 5,
Sampling threshold is 10-4。
4. a kind of automatic question-answering method based on deep learning according to claim 1, it is characterised in that the step
5) it is to the specific method of all problems structure text representation vector with Word2Vec related terms using TFIDF, BOW vector in:
4.1) it to each word in each problem, is multiplied by TFIDF using its BOW vector and is worth to vectorial W1, meanwhile, it uses
Word2Vec obtains maximally related 10 words of the word, and the BOW vectors of 10 correlation words are multiplied by inlet coefficient a=respectively
It 0.1 and sums and obtains W2, sum W1 and W2 to obtain term vector W3;
4.2) it for each problem, is summed using the term vector W3 of each word in step 4.2), obtains text representation
Vector.
5. a kind of automatic question-answering method based on deep learning according to claim 1, it is characterised in that the step
8) it is specially:
5.1) full-text search acquisition is used to put question to similar m problem as Similar Problems with user;
5.2) text representation vector is generated using step 5) with m Similar Problems to customer problem, uses cosine similarity distance
It calculates similitude and sorts, the answer of n problem is as candidate answers before being screened from m problem;
5.3) it uses the neural network model in step 7) to obtain the matching score between problem and candidate answers, is chosen after sequence
The answer to rank the first returns to user.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810026979.8A CN108345585A (en) | 2018-01-11 | 2018-01-11 | A kind of automatic question-answering method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810026979.8A CN108345585A (en) | 2018-01-11 | 2018-01-11 | A kind of automatic question-answering method based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108345585A true CN108345585A (en) | 2018-07-31 |
Family
ID=62961080
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810026979.8A Pending CN108345585A (en) | 2018-01-11 | 2018-01-11 | A kind of automatic question-answering method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108345585A (en) |
Cited By (37)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108932350A (en) * | 2018-08-17 | 2018-12-04 | 沈阳农业大学 | Based on shifty rice pest intelligent answer method |
CN109271497A (en) * | 2018-08-31 | 2019-01-25 | 华南理工大学 | A kind of event-driven service matching method based on term vector |
CN109359182A (en) * | 2018-10-08 | 2019-02-19 | 网宿科技股份有限公司 | A kind of answer method and device |
CN109408678A (en) * | 2018-09-21 | 2019-03-01 | 广州神马移动信息科技有限公司 | Question and answer exchange method and its device |
CN109471980A (en) * | 2018-12-20 | 2019-03-15 | 厦门笨鸟电子商务有限公司 | A kind of search recommendation relatedness computation method based on text matches |
CN109739958A (en) * | 2018-11-22 | 2019-05-10 | 普强信息技术(北京)有限公司 | A kind of specification handbook answering method and system |
CN109766421A (en) * | 2018-12-28 | 2019-05-17 | 上海汇付数据服务有限公司 | Intelligent Answer System and method |
CN109933652A (en) * | 2019-01-17 | 2019-06-25 | 深圳壹账通智能科技有限公司 | Intelligent answer method, apparatus, computer equipment and storage medium |
CN109947928A (en) * | 2019-04-16 | 2019-06-28 | 上海海事大学 | A kind of retrieval type artificial intelligence question and answer robot development approach |
CN110032635A (en) * | 2019-04-22 | 2019-07-19 | 齐鲁工业大学 | One kind being based on the problem of depth characteristic fused neural network to matching process and device |
CN110083692A (en) * | 2019-04-22 | 2019-08-02 | 齐鲁工业大学 | A kind of the text interaction matching process and device of finance knowledge question |
CN110096699A (en) * | 2019-03-20 | 2019-08-06 | 华南师范大学 | Semantic-based machine reads the candidate answers screening technique understood and system |
CN110222163A (en) * | 2019-06-10 | 2019-09-10 | 福州大学 | A kind of intelligent answer method and system merging CNN and two-way LSTM |
CN110502620A (en) * | 2019-07-02 | 2019-11-26 | 平安科技(深圳)有限公司 | Hospital guide's Similar Problems are to generation method, system and computer equipment |
CN110516057A (en) * | 2019-08-23 | 2019-11-29 | 深圳前海微众银行股份有限公司 | A kind of petition letter problem answer method and device |
CN110597971A (en) * | 2019-08-22 | 2019-12-20 | 卓尔智联(武汉)研究院有限公司 | Automatic question answering device and method based on neural network and readable storage medium |
CN110619042A (en) * | 2019-03-13 | 2019-12-27 | 北京航空航天大学 | Neural network-based teaching question and answer system and method |
CN110737839A (en) * | 2019-10-22 | 2020-01-31 | 京东数字科技控股有限公司 | Short text recommendation method, device, medium and electronic equipment |
CN110750632A (en) * | 2019-10-21 | 2020-02-04 | 闽江学院 | Improved Chinese ALICE intelligent question-answering method and system |
CN110825857A (en) * | 2019-09-24 | 2020-02-21 | 平安科技(深圳)有限公司 | Multi-turn question and answer identification method and device, computer equipment and storage medium |
CN111090735A (en) * | 2019-12-25 | 2020-05-01 | 成都航天科工大数据研究院有限公司 | Intelligent question-answering method based on knowledge graph and performance evaluation method thereof |
CN111125295A (en) * | 2019-11-14 | 2020-05-08 | 中国农业大学 | Method and system for obtaining food safety question answers based on LSTM |
CN111190997A (en) * | 2018-10-26 | 2020-05-22 | 南京大学 | Question-answering system implementation method using neural network and machine learning sequencing algorithm |
CN111368058A (en) * | 2020-03-09 | 2020-07-03 | 昆明理工大学 | Question-answer matching method based on transfer learning |
CN111382250A (en) * | 2018-12-29 | 2020-07-07 | 深圳市优必选科技有限公司 | Question text matching method and device, computer equipment and storage medium |
CN111401052A (en) * | 2020-04-24 | 2020-07-10 | 南京莱科智能工程研究院有限公司 | Semantic understanding-based multilingual text matching method and system |
WO2020232877A1 (en) * | 2019-05-21 | 2020-11-26 | 平安科技(深圳)有限公司 | Question answer selection method and apparatus, computer device, and storage medium |
CN112131354A (en) * | 2020-11-26 | 2020-12-25 | 广州华多网络科技有限公司 | Answer screening method and device, terminal equipment and computer readable storage medium |
WO2021003813A1 (en) * | 2019-07-05 | 2021-01-14 | 平安科技(深圳)有限公司 | Answer generation method based on neural network model, and related device |
CN112507097A (en) * | 2020-12-17 | 2021-03-16 | 神思电子技术股份有限公司 | Method for improving generalization capability of question-answering system |
CN112836027A (en) * | 2019-11-25 | 2021-05-25 | 京东方科技集团股份有限公司 | Method for determining text similarity, question answering method and question answering system |
EP3842988A1 (en) * | 2019-12-27 | 2021-06-30 | Beijing Baidu Netcom Science And Technology Co. Ltd. | Method and apparatus for processing questions and answers, electronic device and storage medium |
CN113111168A (en) * | 2020-02-13 | 2021-07-13 | 北京明亿科技有限公司 | Alarm receiving and processing text household registration information extraction method and device based on deep learning model |
WO2021159877A1 (en) * | 2020-02-14 | 2021-08-19 | 华为技术有限公司 | Question answering method and apparatus |
WO2021169263A1 (en) * | 2020-02-26 | 2021-09-02 | 平安科技(深圳)有限公司 | Semantic matching method and device based on internal adversarial mechanism, and storage medium |
CN113342950A (en) * | 2021-06-04 | 2021-09-03 | 北京信息科技大学 | Answer selection method and system based on semantic union |
CN114117022A (en) * | 2022-01-26 | 2022-03-01 | 杭州远传新业科技有限公司 | FAQ similarity problem generation method and system |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102637192A (en) * | 2012-02-17 | 2012-08-15 | 清华大学 | Method for answering with natural language |
US20150227519A1 (en) * | 2014-02-11 | 2015-08-13 | International Business Machines Corporation | Candidate answers for speculative questions in a deep question answering system |
CN107368547A (en) * | 2017-06-28 | 2017-11-21 | 西安交通大学 | A kind of intelligent medical automatic question-answering method based on deep learning |
-
2018
- 2018-01-11 CN CN201810026979.8A patent/CN108345585A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102637192A (en) * | 2012-02-17 | 2012-08-15 | 清华大学 | Method for answering with natural language |
US20150227519A1 (en) * | 2014-02-11 | 2015-08-13 | International Business Machines Corporation | Candidate answers for speculative questions in a deep question answering system |
CN107368547A (en) * | 2017-06-28 | 2017-11-21 | 西安交通大学 | A kind of intelligent medical automatic question-answering method based on deep learning |
Non-Patent Citations (2)
Title |
---|
张扬扬: "自动问答技术在中草药系统中的研究应用", 《中国优秀硕士论文电子期刊网》 * |
悟乙已: "自然语言处理︱简述四大类文本分析中的"词向量"(文本词特征提取)", 《HTTPS://BLOG.CSDN.NET/SINAT_26917383/ARTICLE/DETAILS/52162589》 * |
Cited By (53)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108932350A (en) * | 2018-08-17 | 2018-12-04 | 沈阳农业大学 | Based on shifty rice pest intelligent answer method |
CN109271497B (en) * | 2018-08-31 | 2021-10-26 | 华南理工大学 | Event-driven service matching method based on word vector |
CN109271497A (en) * | 2018-08-31 | 2019-01-25 | 华南理工大学 | A kind of event-driven service matching method based on term vector |
CN109408678A (en) * | 2018-09-21 | 2019-03-01 | 广州神马移动信息科技有限公司 | Question and answer exchange method and its device |
CN109359182A (en) * | 2018-10-08 | 2019-02-19 | 网宿科技股份有限公司 | A kind of answer method and device |
CN111190997B (en) * | 2018-10-26 | 2024-01-05 | 南京大学 | Question-answering system implementation method using neural network and machine learning ordering algorithm |
CN111190997A (en) * | 2018-10-26 | 2020-05-22 | 南京大学 | Question-answering system implementation method using neural network and machine learning sequencing algorithm |
CN109739958A (en) * | 2018-11-22 | 2019-05-10 | 普强信息技术(北京)有限公司 | A kind of specification handbook answering method and system |
CN109471980A (en) * | 2018-12-20 | 2019-03-15 | 厦门笨鸟电子商务有限公司 | A kind of search recommendation relatedness computation method based on text matches |
CN109766421A (en) * | 2018-12-28 | 2019-05-17 | 上海汇付数据服务有限公司 | Intelligent Answer System and method |
CN111382250A (en) * | 2018-12-29 | 2020-07-07 | 深圳市优必选科技有限公司 | Question text matching method and device, computer equipment and storage medium |
CN109933652A (en) * | 2019-01-17 | 2019-06-25 | 深圳壹账通智能科技有限公司 | Intelligent answer method, apparatus, computer equipment and storage medium |
CN110619042A (en) * | 2019-03-13 | 2019-12-27 | 北京航空航天大学 | Neural network-based teaching question and answer system and method |
CN110096699A (en) * | 2019-03-20 | 2019-08-06 | 华南师范大学 | Semantic-based machine reads the candidate answers screening technique understood and system |
CN110096699B (en) * | 2019-03-20 | 2023-06-09 | 华南师范大学 | Candidate answer screening method and system based on semantic machine reading understanding |
CN109947928A (en) * | 2019-04-16 | 2019-06-28 | 上海海事大学 | A kind of retrieval type artificial intelligence question and answer robot development approach |
CN110032635B (en) * | 2019-04-22 | 2023-01-20 | 齐鲁工业大学 | Problem pair matching method and device based on depth feature fusion neural network |
CN110032635A (en) * | 2019-04-22 | 2019-07-19 | 齐鲁工业大学 | One kind being based on the problem of depth characteristic fused neural network to matching process and device |
CN110083692A (en) * | 2019-04-22 | 2019-08-02 | 齐鲁工业大学 | A kind of the text interaction matching process and device of finance knowledge question |
WO2020232877A1 (en) * | 2019-05-21 | 2020-11-26 | 平安科技(深圳)有限公司 | Question answer selection method and apparatus, computer device, and storage medium |
CN110222163B (en) * | 2019-06-10 | 2022-10-04 | 福州大学 | Intelligent question-answering method and system integrating CNN and bidirectional LSTM |
CN110222163A (en) * | 2019-06-10 | 2019-09-10 | 福州大学 | A kind of intelligent answer method and system merging CNN and two-way LSTM |
CN110502620B (en) * | 2019-07-02 | 2023-05-30 | 平安科技(深圳)有限公司 | Method, system and computer equipment for generating guide diagnosis similar problem pairs |
CN110502620A (en) * | 2019-07-02 | 2019-11-26 | 平安科技(深圳)有限公司 | Hospital guide's Similar Problems are to generation method, system and computer equipment |
WO2021003813A1 (en) * | 2019-07-05 | 2021-01-14 | 平安科技(深圳)有限公司 | Answer generation method based on neural network model, and related device |
CN110597971B (en) * | 2019-08-22 | 2022-04-29 | 卓尔智联(武汉)研究院有限公司 | Automatic question answering device and method based on neural network and readable storage medium |
CN110597971A (en) * | 2019-08-22 | 2019-12-20 | 卓尔智联(武汉)研究院有限公司 | Automatic question answering device and method based on neural network and readable storage medium |
CN110516057A (en) * | 2019-08-23 | 2019-11-29 | 深圳前海微众银行股份有限公司 | A kind of petition letter problem answer method and device |
CN110516057B (en) * | 2019-08-23 | 2022-10-28 | 深圳前海微众银行股份有限公司 | Petition question answering method and device |
CN110825857A (en) * | 2019-09-24 | 2020-02-21 | 平安科技(深圳)有限公司 | Multi-turn question and answer identification method and device, computer equipment and storage medium |
CN110825857B (en) * | 2019-09-24 | 2023-07-21 | 平安科技(深圳)有限公司 | Multi-round question and answer identification method and device, computer equipment and storage medium |
CN110750632B (en) * | 2019-10-21 | 2022-09-09 | 闽江学院 | Improved Chinese ALICE intelligent question-answering method and system |
CN110750632A (en) * | 2019-10-21 | 2020-02-04 | 闽江学院 | Improved Chinese ALICE intelligent question-answering method and system |
CN110737839A (en) * | 2019-10-22 | 2020-01-31 | 京东数字科技控股有限公司 | Short text recommendation method, device, medium and electronic equipment |
CN111125295A (en) * | 2019-11-14 | 2020-05-08 | 中国农业大学 | Method and system for obtaining food safety question answers based on LSTM |
CN111125295B (en) * | 2019-11-14 | 2023-11-24 | 中国农业大学 | LSTM-based method and system for obtaining answers to food safety questions |
CN112836027A (en) * | 2019-11-25 | 2021-05-25 | 京东方科技集团股份有限公司 | Method for determining text similarity, question answering method and question answering system |
CN111090735B (en) * | 2019-12-25 | 2023-03-10 | 成都航天科工大数据研究院有限公司 | Performance evaluation method of intelligent question-answering method based on knowledge graph |
CN111090735A (en) * | 2019-12-25 | 2020-05-01 | 成都航天科工大数据研究院有限公司 | Intelligent question-answering method based on knowledge graph and performance evaluation method thereof |
EP3842988A1 (en) * | 2019-12-27 | 2021-06-30 | Beijing Baidu Netcom Science And Technology Co. Ltd. | Method and apparatus for processing questions and answers, electronic device and storage medium |
US11461556B2 (en) | 2019-12-27 | 2022-10-04 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method and apparatus for processing questions and answers, electronic device and storage medium |
CN113111168A (en) * | 2020-02-13 | 2021-07-13 | 北京明亿科技有限公司 | Alarm receiving and processing text household registration information extraction method and device based on deep learning model |
WO2021159877A1 (en) * | 2020-02-14 | 2021-08-19 | 华为技术有限公司 | Question answering method and apparatus |
WO2021169263A1 (en) * | 2020-02-26 | 2021-09-02 | 平安科技(深圳)有限公司 | Semantic matching method and device based on internal adversarial mechanism, and storage medium |
CN111368058A (en) * | 2020-03-09 | 2020-07-03 | 昆明理工大学 | Question-answer matching method based on transfer learning |
CN111401052A (en) * | 2020-04-24 | 2020-07-10 | 南京莱科智能工程研究院有限公司 | Semantic understanding-based multilingual text matching method and system |
CN112131354B (en) * | 2020-11-26 | 2021-04-16 | 广州华多网络科技有限公司 | Answer screening method and device, terminal equipment and computer readable storage medium |
CN112131354A (en) * | 2020-11-26 | 2020-12-25 | 广州华多网络科技有限公司 | Answer screening method and device, terminal equipment and computer readable storage medium |
CN112507097B (en) * | 2020-12-17 | 2022-11-18 | 神思电子技术股份有限公司 | Method for improving generalization capability of question-answering system |
CN112507097A (en) * | 2020-12-17 | 2021-03-16 | 神思电子技术股份有限公司 | Method for improving generalization capability of question-answering system |
CN113342950A (en) * | 2021-06-04 | 2021-09-03 | 北京信息科技大学 | Answer selection method and system based on semantic union |
CN114117022B (en) * | 2022-01-26 | 2022-05-06 | 杭州远传新业科技有限公司 | FAQ similarity problem generation method and system |
CN114117022A (en) * | 2022-01-26 | 2022-03-01 | 杭州远传新业科技有限公司 | FAQ similarity problem generation method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108345585A (en) | A kind of automatic question-answering method based on deep learning | |
CN109271505B (en) | Question-answering system implementation method based on question-answer pairs | |
CN106484664B (en) | Similarity calculating method between a kind of short text | |
CN105589844B (en) | It is a kind of to be used to take turns the method for lacking semantic supplement in question answering system more | |
CN104765769B (en) | The short text query expansion and search method of a kind of word-based vector | |
CN104615767B (en) | Training method, search processing method and the device of searching order model | |
WO2019196314A1 (en) | Text information similarity matching method and apparatus, computer device, and storage medium | |
CN109271506A (en) | A kind of construction method of the field of power communication knowledge mapping question answering system based on deep learning | |
CN105528437B (en) | A kind of question answering system construction method extracted based on structured text knowledge | |
CN109960786A (en) | Chinese Measurement of word similarity based on convergence strategy | |
CN102637192A (en) | Method for answering with natural language | |
CN106126619A (en) | A kind of video retrieval method based on video content and system | |
CN110134946A (en) | A kind of machine reading understanding method for complex data | |
CN110675859A (en) | Multi-emotion recognition method, system, medium, and apparatus combining speech and text | |
CN110348024A (en) | Intelligent identifying system based on legal knowledge map | |
CN110765755A (en) | Semantic similarity feature extraction method based on double selection gates | |
CN108509521A (en) | A kind of image search method automatically generating text index | |
CN104008187A (en) | Semi-structured text matching method based on the minimum edit distance | |
CN109783825A (en) | A kind of ancient Chinese prose interpretation method neural network based | |
CN111694927A (en) | Automatic document review method based on improved word-shifting distance algorithm | |
CN113988053A (en) | Hot word extraction method and device | |
CN108520038B (en) | Biomedical literature retrieval method based on sequencing learning algorithm | |
CN112100212A (en) | Case scenario extraction method based on machine learning and rule matching | |
CN111460147A (en) | Title short text classification method based on semantic enhancement | |
CN110347812B (en) | Search ordering method and system for judicial texts |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180731 |
|
RJ01 | Rejection of invention patent application after publication |