CN108153736A - A kind of relative mapping method based on vector space model - Google Patents
A kind of relative mapping method based on vector space model Download PDFInfo
- Publication number
- CN108153736A CN108153736A CN201711453984.9A CN201711453984A CN108153736A CN 108153736 A CN108153736 A CN 108153736A CN 201711453984 A CN201711453984 A CN 201711453984A CN 108153736 A CN108153736 A CN 108153736A
- Authority
- CN
- China
- Prior art keywords
- predicate
- candidate
- vertex
- relative
- entity
- 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.)
- Granted
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
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- 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
- G06F40/295—Named entity recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a kind of relative mapping methods based on vector space model, belong to natural language processing field.The basic step of the method for the present invention is as follows:First using the correspondence between relative and entity pair, by counting entity to the frequency of appearance and each relatival feature vector of specificity construction.Secondly using vector space model calculated relationship word and the similarity of corresponding predicate, the higher predicate of similarity value is chosen as relatival candidate.Finally by all relatival candidate sequences, choose the predicate candidate with highest confidence level and construct mapping dictionary.Relative predicate mapping method proposed by the present invention can provide a kind of effective solution for the automatic mapping of predicate in relative in natural language and RDF graph data, and reaching, which makes natural language be converted to diagram data, carries out corresponding matched purpose.
Description
Technical field
The invention belongs to natural language processing fields, and in particular to a kind of novel natural language relative and RDF graph number
The technology mapped according to middle predicate.
Background technology
With the development of computer technology, digital information is doubled and redoubled, and the data of magnanimity can be for people to use.However,
Increase a large amount of accumulation data to be formed in face of information explosion formula, text based search engine is merely able to be arranged according to keyword
Sequence index can not really answer the problem of user proposes.At the same time, more and more knowledge mappings start to occur, and compel
It cuts and wants to, directly by the information of structuring and RDF question and answer technologies, the problem of user be understood and answered.In RDF
It is particularly important to the processing of natural language in question answering system.Can natural language generally comprises the elements such as relative and entity,
These elements in natural language are correctly correctly converted into structural data, decide to obtain the correctness of answer.And
In these elements, relatival processing is the first step that can make natural language successful conversion.
It is mapped as that natural language is converted to structural data and question answering in natural language provides base as it can be seen that relatival
Plinth.However, due to the changeable form of natural language, tend not to determine corresponding mapping rule according to the detection of specific word
Then.If, with reference to a large amount of entity pair, relatival mapping relations can be automatically generated according to the relatival feature of natural language,
The accuracy rate of natural language structuring will be improved, so as to contribute to the question answering system of automation.
The research of relative predicate mapping, can meet natural language structuring accurately and rapidly active demand and is
Question answering in natural language is had laid a good foundation.At present, in natural language processing field, have a large amount of research and improve relative
The accuracy rate of predicate mapping, it is seen that the relatival predicate mapping of natural language has extensive future in engineering applications.
Invention content
Present invention aim to address natural language relatives can not be converted into RDF graph data predicate so as to generate
The problem of structured language, with reference to the computational methods of vector similarity in vector space model, proposes a kind of based on vector space
The relative mapping method of model.
Relative mapping method provided by the invention based on vector space model is as follows:
1st, using the path between entity pair in RDF graph data, determine each it is relatival may candidate predicate or
Predicate path;
The determining step in candidate predicate or predicate path is as follows:
The 1.1st, natural language is converted into the combination of relative and entity pair,
Define 1:Natural language sentence S can be converted into an entity pair and relatival combination, with triple G=(R,
E1, E2) represent, wherein:
1. .R is relative, relative is one section of plain text, represents the contact between two entities.
②.E1、E2For entity, two entities collectively form an entity pair.
1.2nd, entity is mapped as vertex in RDF graph data,
Define 2:According to the label of entity, each entity can correspond to an example in RDF graph data, each example
A vertex is represented, the corresponding vertex of the entity is:
Vi=F(Ei)
1.3rd, using the simple path between vertex as the corresponding candidate of a relative.
Define 3:Simple path refers to the path that vertex is not repeated in vertex sequence, closed loop be not present in path.
Define 4:It is relatival one semantic possible candidate in natural language to scheme the simple path between upper two vertex.
It enablesWithRelative R is represented respectivelyiTwo entities connectedWithCorresponding vertex on the diagram, then one between vertex
Simple path P is defined as:
Define 5:One relative RiAll path candidates, uniformly constitute relative RiCandidate collection, the Candidate Set
Conjunction is defined as:
2nd, by candidate predicate path, according to their vertex pair, construction feature is vectorial, to characterize each predicate
Candidate, while for calculating and relatival similarity;
The constitution step of the feature vector of predicate candidate is as follows;
2.1st, every dimension of predicate candidate feature vector is by vertex to forming, and vertex is to being that a predicate path connects
Two vertex connect, vertex expresses the feature vector collectively formed the feature of semanteme in predicate path.
2.2nd, feature vector is subjected to quantification treatment, calculates the inverse document frequency per dimension respectively.
Define 6:In RDF graph data, each predicate path candidateTwo vertex j and k are connected, j, k form one
Vertex pair, pathIt is also possible to connecting other nodes similar to j, k, all these vertex pair are considered, it willEven
The document that all vertex may be constructed a predicate candidate to combining is connect, is denoted asThe common structure of all documents
Into the document sets of predicate candidate.
2.2.1, the inverse document frequency for calculating every dimension.
Define 7:Each vertex pair can be used as a word in document, according to this vertex to going out in entire document sets
Existing number calculates the inverse document frequency on vertex pair, and the total document sets for enabling predicate candidate are CRDF, vertex is to for VPRDF, Mei Gewen
Shelves beThe inverse document frequency is defined as follows:
2.2.2, feature vector is generated by the inverse document frequency of every dimension.
Define 8:It enablesFor predicate candidateVertex to the inverse document frequency of m, then
Feature vector be defined as follows:
3rd, using the entity pair in natural language, can also the corresponding document sets of production Methods word, so as to tectonic relationship
The feature vector of word;
The constitution step of relatival feature vector is as follows;
3.1st, every dimension of relationship word feature vector expresses the feature vector formed by entity by entity to forming
The relatival feature of semanteme.
3.2nd, relationship word feature vector is subjected to quantification treatment, inverse document of the calculated relationship word feature vector per dimension
Frequency.
Define 9:Each natural language triple is all by relative R and two entity E1And E2Composition, different natural languages
In sentence, same relation word has multiple and different entities pair, considers all these entities pair, by all entities of R to combination
Get up to form a relatival feature documents, be denoted as RS (R), all documents together constitute relatival document sets.
3.2.1, the inverse document frequency for calculating every dimension.
Define 7:By each entity to for a word in document, according to time of this entity to occurring in entire document sets
Number, the inverse document frequency of computational entity pair, it is C to enable relatival total document setsREL, entity is to for EPREL, each relative document
For RS (R), which is defined as follows:
3.2.2, feature vector is generated by the inverse document frequency of every dimension.
Define 8:It enablesEntity for relative R is to the inverse document frequency of m, then the feature vector definition of relative R is such as
Under:
4th, each predicate relatival to one is candidate, walks to obtain feature vector according to second step, third, count respectively
Corresponding similarity is calculated, is finally ranked up according to obtaining result, chooses the higher predicate of similarity as candidate;
4.1st, each predicate is candidate in calculated relationship word R and its candidate collectionVector similarity.
Define 9:It enablesIt is that relative R and its predicate are candidateSimilarity, VRFeature for relative R
Vector,For predicate candidateFeature vector, thenIt is defined as follows:
4.2nd, similarity is normalized, generates confidence level.
Define 10:Since the cosine similarity value being calculated is smaller, obtained similarity is used and opens biquadratic
It is amplified, obtains final confidence level, confidence level is defined as follows:
4.3rd, a relative and the similarity calculation of its all possible predicate candidate are completed, obtain confidence level,
Then it is ranked up according to corresponding confidence measure, most confidence level is more than the candidate as rationally candidate, and pressing of σ=0.6 at last
It is ranked up from high to low according to confidence level, ultimately generates relatival corresponding predicate candidate dictionary.
The advantages of the present invention:
The present invention is by semantic and design feature relatival in natural language, proposing a kind of based on vector space model
Relative mapping method, can accurately by relative be mapped as predicate candidate, have a clear superiority in terms of accuracy.This hair
It is bright to meet relative mapping fast and accurately active demand in natural language processing, and can be the natural language in RDF graph data
Speech question and answer lay a good foundation.
Description of the drawings
Fig. 1 is the method for the present invention general flow chart;
Fig. 2 is RDF graph data of the present invention;
Specific embodiment
The process flow of the method for the present invention is as shown in Figure 1.
The specific embodiment of the method for the present invention is introduced with reference to example, table 1 show part natural language triple,
Including relative and entity pair.A part for RDF graph data is as shown in Fig. 2, the red path in wherein Fig. 2 is relative
A possible predicate of wasbornin is candidate.The specific step of the method for the present invention is introduced below in conjunction with RDF graph data shown in Fig. 2
Suddenly:
1 natural language triple of table
Step 1:Determine that relatival predicate is candidate
First by a relatival entity pair, according to label, their corresponding vertex are found in RDF graph data, so
The simple path between vertex is traversed afterwards, and simple path refers to the path that vertex is not repeated in path, i.e., in path not
There are closed loops, and for reasons of efficiency, the length in path is set as being no more than 3.Fig. 2 is a part for RDF graph data, according to table
Relative wasbornin and entity pair in 1<YaoMing,ShangHai>, can be using path BirthPlace as one
Possible predicate is candidate.
Determine that the realization pseudocode of predicate path candidate is as follows:
Input:Entity is to EPRELTwo correspondent entity EP1And EP2
Output:Predicate alternative path set
The algorithm description search relative may Semantic mapping predicate candidate method.First, according to the mark of entity pair
Label determine the entity with same label in RDF graph data, and function F is to determine function, along being connected between two entities
Side, search out different predicate paths, these predicate paths, often with certain Semantic, using these paths as closing
Copula is possible semantic candidate, the predicate candidate collection of production Methods word.
Step 2:Construction feature vector
2.1st, the feature vector of predicate candidate is constructed.By all vertex pair of each predicate candidate, regard a void as
Document by each vertex pair, regards a function word as, by calculating the inverse document frequency of each function word, generates feature vector.To scheme
For BirthPlace in 2, the vertex of BirthPlace to having respectively<Yao Ming,ShangHai>、<Obama,
American>With<Nelson Mandela,Mvezo>, these three components have together constituted with the dead letter shelves of BirthPlace, respectively
The inverse document frequency on three vertex pair is calculated, so as to generate the feature vector of predicate candidate BirthPlace.
2.2nd, the feature vector of tectonic relationship word.By each relatival all entities pair, regard a dead letter shelves as,
Using the same method in step 2.1, feature vector is generated.By taking the was born in table 1 as an example, the reality of was born in
Body to having respectively<Yao Ming,ShangHai>、<Obama,American>With<Nelson Mandela,Mvezo>, count respectively
Calculate the inverse document frequency of three components, the feature vector of production Methods word wasborn in.
Step 3:Calculate similarity
The similarity of the corresponding predicate candidate feature vector of each relationship word feature vector is calculated respectively, finally according to phase
Confidence level is obtained like degree.
The pseudocode of similarity calculation is as follows:
Input:Relatival dead letter shelves RS (R), the dead letter shelves of predicate candidate
Output:Relative and the similarity of predicate candidate
The method that the algorithm description is generated feature vector according to inverse document frequency and similarity is calculated using feature vector,
After the completion of calculating, predicate candidate of the confidence level more than σ=0.6 is chosen as rationally candidate, and being reversed according to confidence level, it is raw
Into mapping dictionary.
Claims (6)
1. a kind of relative mapping method based on vector space model, it is characterised in that this method includes:
1st, using the path between entity pair in RDF graph data, each relatival possible candidate predicate or predicate are determined
Path;
2nd, by candidate predicate path, according to their vertex pair, construction feature vector is waited to characterize each predicate
Choosing, while for calculating and relatival similarity;
3rd, using the entity pair in natural language, can also the corresponding document sets of production Methods word, so as to tectonic relationship word
Feature vector;
4th, each predicate relatival to one is candidate, and feature vector is obtained according to the 2nd step, the 3rd step, calculates correspond to respectively
Similarity, be finally ranked up according to obtaining result, choose the higher predicate of similarity as candidate.
2. the relative mapping method according to claim 1 based on vector space model, it is characterised in that in the 1st step:
The determining step in candidate predicate or predicate path is as follows:
The 1.1st, natural language is converted into the combination of relative and entity pair,
Define 1:Natural language sentence S can be converted into an entity pair and relatival combination, with triple G=(R, E1, E2)
It represents, wherein:
1. .R is relative, relative is one section of plain text, represents the contact between two entities;
②.E1、E2For entity, two entities collectively form an entity pair;
1.2nd, entity is mapped as vertex in RDF graph data,
Define 2:According to the label of entity, each entity can correspond to an example in RDF graph data, and each example represents
One vertex, the corresponding vertex of the entity are:
Vi=F (Ei)
1.3rd, using the simple path between vertex as the corresponding candidate of a relative;
Define 3:Simple path refers to the path that vertex is not repeated in vertex sequence, closed loop be not present in path;
Define 4:It is relatival one semantic possible candidate, order in natural language to scheme the simple path between upper two vertex
WithRelative R is represented respectivelyiTwo entities connectedWithCorresponding vertex on the diagram, then a simpied method between vertex
Diameter P is defined as:
Define 5:One relative RiAll path candidates, uniformly constitute relative RiCandidate collection, which determines
Justice is:
3. the relative mapping method according to claim 1 based on vector space model, it is characterised in that in the 2nd step:
The constitution step of the feature vector of predicate candidate is as follows;
2.1st, every dimension of predicate candidate feature vector is by vertex to forming, and vertex is to being that a predicate path is connected
Two vertex, vertex express the feature vector collectively formed the feature of semanteme in predicate path;
2.2nd, feature vector is subjected to quantification treatment, calculates the inverse document frequency per dimension respectively;
Define 6:In RDF graph data, each predicate path candidateTwo vertex j and k are connected, j, k form a vertex
It is right, pathIt is also possible to connecting other nodes similar to j, k, all these vertex pair are considered, it willConnection is all
Vertex may be constructed the document of a predicate candidate to combining, be denoted asAll documents together constitute meaning
The document sets of language candidate;
2.2.1, the inverse document frequency for calculating every dimension;
Define 7:Each vertex pair can be used as a word in document, according to this vertex to occurring in entire document sets
Number calculates the inverse document frequency on vertex pair, and the total document sets for enabling predicate candidate are CRDF, vertex is to for VPRDF, each document isThe inverse document frequency is defined as follows:
2.2.2, feature vector is generated by the inverse document frequency of every dimension;
Define 8:It enablesFor predicate candidateVertex to the inverse document frequency of m, thenFeature vector definition such as
Under:
4. the relative mapping method according to claim 1 based on vector space model, it is characterised in that in the 3rd step:
The constitution step of relatival feature vector is as follows;
3.1st, every dimension of relationship word feature vector expresses pass by entity by entity to forming to the feature vector formed
The feature of semanteme of copula;
3.2nd, relationship word feature vector is subjected to quantification treatment, inverse document frequency of the calculated relationship word feature vector per dimension
Rate;
Define 9:Each natural language triple is all by relative R and two entity E1And E2Composition, different natural language sentences
In, same relation word has multiple and different entities pair, considers all these entities pair, by all entities of R to combining
A relatival feature documents are formed, are denoted as RS (R), all documents together constitute relatival document sets;
Define 8:It enablesEntity for relative R is to the inverse document frequency of m, then the feature vector of relative R is defined as follows:
5. the relative mapping method according to claim 1 based on vector space model, it is characterised in that in the 4th step:
4.1st, each predicate is candidate in calculated relationship word R and its candidate collectionVector similarity;
Define 9:It enablesIt is that relative R and its predicate are candidateSimilarity, VRFor the feature vector of relative R,For predicate candidateFeature vector, thenIt is defined as follows:
4.2nd, similarity is normalized, generates confidence level;
Define 10:Since the cosine similarity value being calculated is smaller, obtained similarity is carried out using biquadratic is opened
Amplification, obtains final confidence level, confidence level is defined as follows:
4.3rd, a relative and the similarity calculation of its all possible predicate candidate are completed, obtains confidence level, then
It is ranked up according to corresponding confidence measure, most confidence level is more than the candidate as rationally candidate of σ=0.6 at last, and according to putting
Reliability is ranked up from high to low, ultimately generates relatival corresponding predicate candidate dictionary.
6. the predicate relative mapping method according to claim 4 based on vector similarity, it is characterised in that the 3.2nd step
In:
3.2.1, the inverse document frequency for calculating every dimension;
Define 7:By each entity to for a word in document, according to this entity to the number that occurs in entire document sets,
The inverse document frequency of computational entity pair, it is C to enable relatival total document setsREL, entity is to for EPREL, each relative document is
RS (R), the inverse document frequency are defined as follows:
3.2.2, feature vector is generated by the inverse document frequency of every dimension.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711453984.9A CN108153736B (en) | 2017-12-28 | 2017-12-28 | Relation word mapping method based on vector space model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711453984.9A CN108153736B (en) | 2017-12-28 | 2017-12-28 | Relation word mapping method based on vector space model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108153736A true CN108153736A (en) | 2018-06-12 |
CN108153736B CN108153736B (en) | 2021-07-02 |
Family
ID=62463597
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711453984.9A Expired - Fee Related CN108153736B (en) | 2017-12-28 | 2017-12-28 | Relation word mapping method based on vector space model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108153736B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109213844A (en) * | 2018-08-13 | 2019-01-15 | 腾讯科技(深圳)有限公司 | A kind of text handling method, device and relevant device |
CN109408527A (en) * | 2018-10-15 | 2019-03-01 | 广东顺德西安交通大学研究院 | A kind of RDF structuralized query method for auto constructing based on vector space |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110246407A1 (en) * | 2008-12-15 | 2011-10-06 | Korea Instititute of Science & Technology Informat | System and method for hybrid rete reasoning based on in-memory and dbms |
US20150127323A1 (en) * | 2013-11-04 | 2015-05-07 | Xerox Corporation | Refining inference rules with temporal event clustering |
CN105468605A (en) * | 2014-08-25 | 2016-04-06 | 济南中林信息科技有限公司 | Entity information map generation method and device |
CN106649266A (en) * | 2016-11-29 | 2017-05-10 | 北京科技大学 | Logical inference method for ontology knowledge |
CN106997399A (en) * | 2017-05-24 | 2017-08-01 | 海南大学 | A kind of classification question answering system design method that framework is associated based on data collection of illustrative plates, Information Atlas, knowledge mapping and wisdom collection of illustrative plates |
CN107506486A (en) * | 2017-09-21 | 2017-12-22 | 北京航空航天大学 | A kind of relation extending method based on entity link |
-
2017
- 2017-12-28 CN CN201711453984.9A patent/CN108153736B/en not_active Expired - Fee Related
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110246407A1 (en) * | 2008-12-15 | 2011-10-06 | Korea Instititute of Science & Technology Informat | System and method for hybrid rete reasoning based on in-memory and dbms |
US20150127323A1 (en) * | 2013-11-04 | 2015-05-07 | Xerox Corporation | Refining inference rules with temporal event clustering |
CN105468605A (en) * | 2014-08-25 | 2016-04-06 | 济南中林信息科技有限公司 | Entity information map generation method and device |
CN106649266A (en) * | 2016-11-29 | 2017-05-10 | 北京科技大学 | Logical inference method for ontology knowledge |
CN106997399A (en) * | 2017-05-24 | 2017-08-01 | 海南大学 | A kind of classification question answering system design method that framework is associated based on data collection of illustrative plates, Information Atlas, knowledge mapping and wisdom collection of illustrative plates |
CN107506486A (en) * | 2017-09-21 | 2017-12-22 | 北京航空航天大学 | A kind of relation extending method based on entity link |
Non-Patent Citations (2)
Title |
---|
YUAN XIAOJIE ET AL: "Question classification question answering based on real-world web data sets", 《JOURNAL OF SOUTHEAST UNIVERSITY(ENGLISH EDITION)》 * |
周寻: "基于语义的谓词标注方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109213844A (en) * | 2018-08-13 | 2019-01-15 | 腾讯科技(深圳)有限公司 | A kind of text handling method, device and relevant device |
CN109213844B (en) * | 2018-08-13 | 2023-03-21 | 腾讯科技(深圳)有限公司 | Text processing method and device and related equipment |
CN109408527A (en) * | 2018-10-15 | 2019-03-01 | 广东顺德西安交通大学研究院 | A kind of RDF structuralized query method for auto constructing based on vector space |
CN109408527B (en) * | 2018-10-15 | 2020-12-01 | 广东顺德西安交通大学研究院 | RDF structured query automatic construction method based on vector space |
Also Published As
Publication number | Publication date |
---|---|
CN108153736B (en) | 2021-07-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106295796B (en) | entity link method based on deep learning | |
Bhatia et al. | Automatic labelling of topics with neural embeddings | |
CN107480143B (en) | Method and system for segmenting conversation topics based on context correlation | |
WO2017101342A1 (en) | Sentiment classification method and apparatus | |
TW202020688A (en) | Method for determining address text similarity, address searching method, apparatus, and device | |
CN105528437B (en) | A kind of question answering system construction method extracted based on structured text knowledge | |
CN109190117A (en) | A kind of short text semantic similarity calculation method based on term vector | |
CN111143672B (en) | Knowledge graph-based professional speciality scholars recommendation method | |
CN107562831A (en) | A kind of accurate lookup method based on full-text search | |
CN105095433A (en) | Recommendation method and device for entities | |
JP2014112316A (en) | Question answering program using large amounts of comment sentences, server and method | |
CN110019732A (en) | A kind of intelligent answer method and relevant apparatus | |
WO2015032301A1 (en) | Method for detecting the similarity of the patent documents on the basis of new kernel function luke kernel | |
CN109145083B (en) | Candidate answer selecting method based on deep learning | |
CN105740448B (en) | More microblogging timing abstract methods towards topic | |
CN113283236B (en) | Entity disambiguation method in complex Chinese text | |
CN114780690B (en) | Patent text retrieval method and device based on multi-mode matrix vector representation | |
CN116680384A (en) | Knowledge question-answering method, device, equipment and storage medium | |
CN111160041B (en) | Semantic understanding method and device, electronic equipment and storage medium | |
WO2024099037A1 (en) | Data processing method and apparatus, entity linking method and apparatus, and computer device | |
CN113761890A (en) | BERT context sensing-based multi-level semantic information retrieval method | |
Zeng et al. | Entity linking on Chinese microblogs via deep neural network | |
CN116561264A (en) | Knowledge graph-based intelligent question-answering system construction method | |
CN102298589A (en) | Method and device for generating emotion tendentiousness template, and method and device for using emotion tendentiousness template | |
CN108153736A (en) | A kind of relative mapping method based on vector space model |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210702 Termination date: 20211228 |
|
CF01 | Termination of patent right due to non-payment of annual fee |