CN107506486A - A kind of relation extending method based on entity link - Google Patents
A kind of relation extending method based on entity link Download PDFInfo
- Publication number
- CN107506486A CN107506486A CN201710858346.9A CN201710858346A CN107506486A CN 107506486 A CN107506486 A CN 107506486A CN 201710858346 A CN201710858346 A CN 201710858346A CN 107506486 A CN107506486 A CN 107506486A
- Authority
- CN
- China
- Prior art keywords
- entity
- relation
- mrow
- candidate
- msub
- 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
- 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
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/288—Entity relationship models
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Computational Linguistics (AREA)
- Machine Translation (AREA)
Abstract
The present invention provides a kind of relation extending method based on entity link, the present invention uses the technological means of entity link, by the text link in natural language into knowledge base corresponding to physically, the Entity recognition not marked in message box property value is come out, according to candidate's entity type and the entity type matching degree at relation both ends, by the use of encyclopaedia entry tag system as the type of entity, obtain the relationship type at relation both ends by counting stipulations.Meanwhile present invention employs sequence of the method for Model Fusion to candidate's entity, by the blend of predominance linear model of nonlinear lifting (boosting) integrated model, model performance is improved, ensures the accuracy of relation.The entity not linked in message box can be efficiently identified out by the present invention, and solve the problems such as entity and entity alias of the same name, with target entity opening relationships, implementation relation expands.
Description
Technical field
The present invention relates to a kind of relation extending method, more particularly to a kind of relation extending method based on entity link.
Background technology
In recent years, in order to solve the matter of semantics of internet information, it is proposed that semantic web, i.e., it is each comprising a large amount of descriptions
The WWW of relation is enriched between kind entity and entity, the things on network is understood as an entity, possesses unique system
One resource identifier (URI), there is semantic interlink between these entities, machine can be allowed to understand text.Based on this, major search is drawn
Hold up company and all issued knowledge mapping, it is intended to meet the searching requirement that user increasingly improves, improve Consumer's Experience.Knowledge mapping
Claim knowledge base, its essence is exactly a kind of machine readable semantic net being made up of multiple elements such as concept, entity, attribute, relation
Network, entity and its relational organization are got up in a structured way.At present, established with the method for automation or semi-automation
Multiple large scale knowledge bases, for knowledge question, knowledge reasoning and knowledge reasoning etc..Well-known knowledge base such as Google's knowledge graph
Spectrum, DBpedia and Freebase etc..Wherein, encyclopaedia class website be build knowledge base a most important source, its structuring
Degree is high, knowledge coverage rate field is wide, information updating speed is fast, and the abundant information for being described the entity is contained in each entry,
In addition to non-structured basic content of text, the message box of description entity attribute is also included, is retouched in the form of attribute-value pair
Predicate bar entity attributes and relation, wherein property value may contain link and point to other entities, be establish knowledge base one
Important sources.Large scale knowledge base such as DBpedia is the relation extraction that entity is carried out from the message box of English Wiki, by property value
In the extraction with the sensing of hyperlink other entities be relation, then convert them into resource description framework (RDF) ternary
Group.
However, when knowledge is extracted from the message box of Chinese encyclopaedia by prior art, compared to English Wiki, its interior chain
Very imperfect, only very small part entity is marked out hyperlink, lost many semantic relations, so needing what completion lacked
Chain fetches Extended Relations.Such as in the message box of entity " Qinghai-Tibet Platean ", property value " Lee of " song original singer " this relation
Na ", still " Li Na " is only to occur in the form of a character string, not and is linked to that " singer Li Na " is corresponding real
On body.
The method that prior art solves this problem is broadly divided into two kinds:One kind is by string matching property value and entity
Title, it specifically, can be matched if there is the title of an entity with the property value, just establish a relation, it is this
Method is only to use string matching.But because the diversity and ambiguousness of natural language, same entity have different tables
Up to (deformation of physical name), and same physical name may correspond to different entities (disambiguation of entity of the same name), and this method may
Mistake can be caused;Another method is to utilize order models, extracts several features, and all candidate's entities are ranked up,
But the type information of entity is not accounted for, cause very poor for the entity result under same category.
The content of the invention
The present invention use entity link technological means, by the text link in natural language into knowledge base corresponding reality
On body, the Entity recognition not marked in message box property value is come out, according to candidate's entity type and the entity class at relation both ends
Type matching degree, by the use of encyclopaedia entry tag system as the type of entity, obtain the relation at relation both ends by counting stipulations
Type.Meanwhile present invention employs sequence of the method for Model Fusion to candidate's entity, by nonlinear lifting (boosting)
In the blend of predominance linear model of integrated model, model performance is improved, ensures the accuracy of relation.Can be effective by the present invention
Ground identifies the entity not linked in message box, and solves the problems such as entity and entity alias of the same name, establishes and closes with target entity
System, implementation relation expand.
Brief description of the drawings
Fig. 1 is the operational flow diagram of the present invention.
Fig. 2 is the Model Fusion figure of the embodiment of the present invention.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in each embodiment of invention described below
Conflict can is not formed each other to be mutually combined.
The invention provides a kind of relation extending method based on entity link technology.It is the frame of this method as shown in Figure 1
Figure, including three modules:Candidate generation module, feature extraction module and order module.
Give a subject entity ei, the attribute r of its message box (infobox)jProperty value in presence may point to it is a certain
The physical name m of entityj。Expression is possible to linked candidate's entity, and task is from all candidate's entitiesIn look for
To target entity really to be linked.
Candidate generation module realizes the fit generation of all Candidate Sets, and the generation of candidate's entity set is mainly based upon
Physical name mjObtained with the similarity of character string of physical name in knowledge base.In order to generate all candidate's entities that may be pointed to
The present invention extracts in encyclopaedia information to establish the mapping between physical name and entity, for example, entry in itself, the disambiguation page, entry text
" alias " attribute etc. in hyperlink Anchor Text and message box (infobox) in this, generates a physical name-entity dictionary D.Table
1 is that the entity word-entity maps dictionary D, each single item in dictionary<Key (key), it is worth (value), number (count)>Represent
The number that correspondent entity occurs under physical name, corresponding entity Candidate Set and the physical name.Wherein, number can embody entity
Popularity.After having physical name-entity dictionary, it is possible to identify not yet linked property value m, then found in dictionary
All candidate's entity E corresponding to itm。
Table 1
Each entry in encyclopaedia describes an entity, contains the various information of entity.Wherein, the title of entry
It is the name of the entity most standard, the present invention extracts the entitled dictionary D of entry page key (key), its described entity
Uniform Resource Identifier (URI) as value (value) be added in dictionary D.It is worth noting that, in encyclopaedia, if
Entity of the same name, in order to prevent ambiguity, identified by bracket.Such as entity " in sportsman Li Na " the entry page, mark
Topic is " Li Na (Chinese Women tennis star) ", such case, can be removed bracket as key, will " Li Na " be key, entity
" sportsman Li Na " Uniform Resource Identifier is as value.
Different entities in encyclopaedia might have identical name, and the disambiguation page of the candidate generation module is exactly to use
There are the different entities of identical physical name in differentiation.For example, occur 47 differences in the disambiguation page of physical name " Li Na "
Entity, include entity " sportsman Li Na " and " multiple entities such as singer Li Na ".
Hyperlinked information is generally comprised in the text of encyclopaedia entry the physical name occurred in text is linked to corresponding reality
On body, it is very useful that this information provides the number that signified stereotropic alias, the deformation of physical name and entity are mentioned etc.
Information;Meanwhile the number that correspondent entity occurs under the physical name also can be therefrom extracted, if a certain entity refers in the text
The number arrived is more, then illustrates that popularity is higher, such as, when mentioning entity " Li Na ", everybody is first it is envisioned that entity " fortune
Li Na " rather than other entities of the same name are mobilized, so number information can be very good to reflect entity popularity, this information has
Help judge that " " Li Na " entity more of the same name than other sportsman Li Na " more likely turns into linked object to entity.
In most information frame, can all there is entity " alias " attribute, to describe entity alias, abbreviation, outer literary fame etc.
The variation information of physical name, the variation title that this partial data can identifies physical name in physical name is extracted, improve relation
The recall rate of expansion.For example have " shark " in the alias of entity " Sha Kuier Ao Ni'ers ", this explanation, when mentioning " big shark
During this character string of fish ", it is possible to be linked to " Sha Kuier Ao Ni'ers ", and this is physically.
After physical name-entity dictionary D is established, m is retrieved from DjCorresponding candidate's entityBut multiple times be present
Entity is selected, i.e.,More than 1, therefore the sequence to each candidate's entity is most important.Extraction feature characterizes candidate's entity and mj、
eiAnd rjRelevance, then candidate's entity is given a mark using the order models of paired (pairwise), acquirement divides highest
As target entity, with eiOpening relationships, generate new knowledge.The present invention is according to the characteristics of message box (infobox), from candidate
Semanteme and textual association degree, relation-entity type these three angle design characteristic extracting modules of the entity in itself, between entity are next
Embody the matching degree of candidate's entity.
In order to formally represent characteristic function,Represent eiMessage box (infobox) in the entity that has linked,
Represent eiText in have the entity of hyperlink,Represent eiThere is the entity of hyperlink in summary info,Relation be present in expression
rjEntity.Table 2 is the characterizing definition of candidate's entity.
Table 2
Entity correlated characteristic:Including f1-f4, it is real with theme that this category feature depends only on candidate's entity and corresponding relation
Body.Entity popularity Popularity (ej) it is that can obtain entity word m from entity word-entity mapping dictionaryjTo entity ejBar
Part probability P (ej|mj), i.e. entity word mjRefer to entity ejPrior probability
Wherein, count (mj) refer to the m in hyperlink Anchor TextjThe total degree of appearance, count (mj, ej) refer to Anchor Text mjChain
It is connected to entity ejNumber.
Context relation feature:Including f5-f8, the semantic association of main code candidate entity and subject entity and entity word
Degree and text similarity.Specifically, come the semantic association degree between computational entity, given using based on encyclopaedia structure of hyperlinks
Two entity eiAnd ej, the mode for obtaining its semantic association degree is as follows:
Wherein,WithIt is chain respectively to eiAnd ejEntity sets, W is all entities in encyclopaedia.As can be seen that
Formula (1) is the deformation of Jie Kade (Jaccard) similarity factor, two entities to enter chain number jointly more, then semantic association degree
It is higher.Obtain candidate's entity ejWith entity sets TeiBetween semantic association degree mode it is as follows:
Similar, Ie can be usedi, AeiAnd ErjTo substitute TeiObtain the semantic association of candidate's entity and other entity setses
Degree, is correspondingly made available f6、f7And f10.Further, it is also possible to define the description information and subject entity ei description informations of candidate's entity
Text similarity, represent text using bag of words (Bag of Words), frequency-anti-document frequency (TF-IDF) represents
Weight vectors, the matching degree of vector is then weighed with cosine similarity.
Relation-entity type feature:Including f9-f11, when the type of candidate's entity is the same, above-mentioned feature is difficult to have
There is identification, so the present invention proposes the entity type of consideration relation the right and left, i.e.,<Subject entity type, relation, target are real
Body type>.For example a relation " representative works " is given, and if the type of subject entity is " singer ", linked entity
Type is likely to " song ";And if the type of subject entity is " performer ", that target entity type is more likely to belong to " video display
Works ".So if the type of relation both sides entity can be defined accurately, it is possible in given subject entity and relation,
Investigate whether each candidate's entity type matches.In order to define this feature, using the label information of encyclopaedia as reality in the present invention
Body type, then enumerate all entity types being connected with this relation and statistics number.Relation " masterpiece is illustrated in table 3
Entity type corresponding to product " and its right and left.
Table 3
So, it is assumed that Type (ei) it is entity eiType give a subject entity eiWith relation rj, obtain candidate's entity
ejObtained entity type score is as follows:
The present invention generates a vector to each candidate's entity and then defines scoring functions f (m, e)=(f1,
f2..., f11), marking s is carried out to each candidate's entityoF (m, e), it is real using the entity of highest scoring as the target to be linked
Body, with subject entity opening relationships.The parameter of order models, mould are learnt using paired (pairwise) training method herein
The training objective of type is correct entity is obtained the score higher than false entries.Use ei> ejPresentation-entity eiRanking it is higher,
It can so obtain training entity pair<ei, ej>, all entities can be obtained by with a partial order of candidate's entity to carrying out classification
Relation, so as to realize sequence.
The present invention introduces two kinds of models in sequence, lifts decision tree and logistic regression using gradient to train sequence mould
Type, the scheme of Model Fusion are as shown in Figure 2.Training obtains the ordering relation of candidate's inter-entity.Wherein, nonlinear gradient lifting
Decision-tree model (GBDT) can learn high dimensional feature, as the supplement of primitive character, finally enter in linear model, to institute
Some candidate's entities are given a mark, and acquirement divides target entity of the highest as opening relationships is wanted.
Gradient lifting decision-tree model is trained using primitive character, the leaf node of all trees of acquisition is exactly its generation
Feature space, when each sample point by gradient lifting decision-tree model each tree when, a leaf node can be fallen on, i.e.,
Produce an intermediate features.The gradient lifting decision-tree model is made up of more trees, and each iteration is all to reduce residual error
A decision tree is newly established on gradient direction, the result for all trees of finally adding up.All these features and primitive character intersect
It is input to together again in linear model logistic regression disaggregated model (LR) and obtains ordering relation.Gradient lifting decision-tree model can be with
Excavate out more added with the feature of discrimination, by two kinds of Model Fusions together, can effective lift scheme generalization ability.It is described
Use figure penalties function in gradient lifting decision tree, logistic regression disaggregated model by activation primitive (Sigmoid functions) come
Represent eiCompare ejThe higher probability of ranking is:
The training process that it is gradient lifting decision tree (MART) first that the training process of model parameter, which is, during each iteration
The parameter r being fitted on loss function negative gradient direction is removed in the generation newly setmi, obtain the value c being fitted on each leafmj.Generation ladder
After degree lifting decision-tree model tree-model, training data T ' corresponding to the feature after newly being changed, finally, cross entropy is made
Real label (label) is fitted for loss function, and the method declined using gradient goes to obtain model parameter W.
Finally it should be noted that:The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations;Although
The present invention is described in detail with reference to the foregoing embodiments, it will be understood by those within the art that:It still may be used
To be modified to the technical scheme described in previous embodiment, or equivalent substitution is carried out to which part technical characteristic;And
These modifications are replaced, and the essence of appropriate technical solution is departed from the spirit and model of various embodiments of the present invention technical scheme
Enclose.
Claims (6)
1. a kind of relation extending method based on entity link, it is characterised in that methods described includes to obtain in message box
Information, a physical name-entity dictionary is generated by candidate generation module, identified according to the physical name-entity dictionary
Not yet linked property value m, find all candidate's entity E corresponding to itm, it is real that candidate is then extracted by feature extraction module
The relevance of body and physical name, subject entity and attribute, gives a mark finally by order module to each candidate's entity, will
Point highest entity is as the target entity to be linked, by the target entity and subject entity opening relationships.
2. the method as described in claim 1, it is characterised in that information in encyclopaedia is extracted in the candidate generation module and is built
Vertical mapping between physical name and entity, described information include hyperlink anchor text of the entry, in the disambiguation page, entry text in itself
This and alias property etc., generate physical name-entity dictionary, each single item in the dictionary<Key (key), it is worth (value), number
(count)>The number that correspondent entity occurs under presentation-entity name, corresponding entity Candidate Set and the physical name, the number body
The popularity of real body, not yet linked property value m is identified by the physical name-entity dictionary, it is right then to find its
All candidate's entity E answeredm。
3. the method as described in claim 1, it is characterised in that the feature extraction module is according to message box (infobox)
Feature, extracted from semantic and three textual association degree, the relation-entity type angles of candidate's entity in itself, between entity
Extract to embody the matching degree of candidate's entity, described be extracted is characterized as entity correlated characteristic, context relation feature, closes
System-entity type feature.
4. method as claimed in claim 3, it is characterised in that the entity correlated characteristic includes entity ejPopularity, entity
eiWith entity ejCo-occurrence number, entity ejWhether in entity eiOccurred in text, entity eiWhether in ejOccurred in text,
Obtain entity word mjRefer to entity ejThe mode of prior probability be:
<mrow>
<msub>
<mi>f</mi>
<mn>1</mn>
</msub>
<mo>=</mo>
<mi>P</mi>
<mi>o</mi>
<mi>u</mi>
<mi>l</mi>
<mi>a</mi>
<mi>r</mi>
<mi>t</mi>
<mi>i</mi>
<mi>t</mi>
<mi>y</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>e</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<mi>c</mi>
<mi>o</mi>
<mi>u</mi>
<mi>n</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<mrow>
<msub>
<mi>m</mi>
<mi>j</mi>
</msub>
<mo>,</mo>
<msub>
<mi>e</mi>
<mi>j</mi>
</msub>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>c</mi>
<mi>o</mi>
<mi>u</mi>
<mi>n</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>m</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
Wherein, count (mj) it is m in hyperlink Anchor TextjThe total degree of appearance, count (mj, ej) it is Anchor Text mjIt is linked to reality
Body ejNumber;The context relation feature includes ejWithThe average semantic association degree of middle entity, ejWithMiddle entity is put down
Equal semantic association degree, ejWithThe average semantic association degree of middle entity, ejAnd eiText similarity, it is describedFor eiInformation
The entity linked in frame (infobox), it is describedFor eiText in have the entity of hyperlink, it is describedFor eiSummary letter
There is the entity of hyperlink in breath, obtain two entity eiAnd ejSemantic association degree SR mode is:
Wherein, it is describedAnd institute
StateIt is chain respectively to eiAnd ejEntity sets, the W is all entities in encyclopaedia;The relation-entity type feature
Including giving rjWhen, ejAnd rjThe entity type matching degree at both ends, ejWithThe average semantic association degree of middle entity, ejWithIn
The shared attribute number of entity, it is describedThe relation r be present in expressionjEntity, the extraction of the relation-entity type feature
It is using the label information of encyclopaedia as entity type, enumerates all entity types being connected with the relation of giving and statistics number.
5. the method as described in claim 1, it is characterised in that the order module learns to arrange using paired training method
The parameter of sequence model, the training objective of model make correct entity obtain the score higher than false entries, ei> ejPresentation-entity ei
Ranking it is higher, so as to the entity pair trained<ei, ej>, to all entities pair<ei, ej>Classification can is carried out to obtain
To a partial ordering relation of candidate's entity, so as to realize sequence.
6. method as claimed in claim 5, it is characterised in that the side merged during the order module sequence using multi-model
Method, decision tree (MART) and logistic regression classification (LR) training order models are lifted using gradient, the gradient lifts decision tree
It is made up of more trees, each iteration is all newly to establish a decision tree on the gradient direction for reducing residual error, is finally added up all
The result of tree, the result and primitive character of all trees of adding up, which intersect to be input to together in linear model logistic regression again, to be obtained
Ordering relation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710858346.9A CN107506486A (en) | 2017-09-21 | 2017-09-21 | A kind of relation extending method based on entity link |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710858346.9A CN107506486A (en) | 2017-09-21 | 2017-09-21 | A kind of relation extending method based on entity link |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107506486A true CN107506486A (en) | 2017-12-22 |
Family
ID=60698442
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710858346.9A Pending CN107506486A (en) | 2017-09-21 | 2017-09-21 | A kind of relation extending method based on entity link |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107506486A (en) |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108153736A (en) * | 2017-12-28 | 2018-06-12 | 南开大学 | A kind of relative mapping method based on vector space model |
CN108563710A (en) * | 2018-03-27 | 2018-09-21 | 腾讯科技(深圳)有限公司 | A kind of knowledge mapping construction method, device and storage medium |
CN108763333A (en) * | 2018-05-11 | 2018-11-06 | 北京航空航天大学 | A kind of event collection of illustrative plates construction method based on Social Media |
CN109408704A (en) * | 2018-09-03 | 2019-03-01 | 平安科技(深圳)有限公司 | Fund data correlating method, system, computer equipment and storage medium |
CN110134965A (en) * | 2019-05-21 | 2019-08-16 | 北京百度网讯科技有限公司 | Method, apparatus, equipment and computer readable storage medium for information processing |
CN110147401A (en) * | 2019-05-22 | 2019-08-20 | 苏州大学 | Merge the knowledge base abstracting method of priori knowledge and context-sensitive degree |
CN110245197A (en) * | 2019-05-20 | 2019-09-17 | 北京百度网讯科技有限公司 | A kind of the whole network entity associated method and system |
CN110555137A (en) * | 2018-03-31 | 2019-12-10 | 华为技术有限公司 | Label filling method and device |
CN110704613A (en) * | 2019-08-23 | 2020-01-17 | 上海科技发展有限公司 | Vocabulary database construction and query method, database system, equipment and medium |
CN110705301A (en) * | 2019-09-30 | 2020-01-17 | 京东城市(北京)数字科技有限公司 | Entity relationship extraction method and device, storage medium and electronic equipment |
CN110795565A (en) * | 2019-09-06 | 2020-02-14 | 腾讯科技(深圳)有限公司 | Semantic recognition-based alias mining method, device, medium and electronic equipment |
CN110852106A (en) * | 2019-11-06 | 2020-02-28 | 腾讯科技(深圳)有限公司 | Named entity processing method and device based on artificial intelligence and electronic equipment |
CN110968665A (en) * | 2019-11-08 | 2020-04-07 | 浙江工业大学 | Method for recognizing upper and lower level word relation based on gradient enhanced decision tree |
CN111259087A (en) * | 2020-01-10 | 2020-06-09 | 中国科学院软件研究所 | Computer network protocol entity linking method and system based on domain knowledge base |
CN111339313A (en) * | 2020-02-18 | 2020-06-26 | 北京航空航天大学 | Knowledge base construction method based on multi-mode fusion |
CN111401049A (en) * | 2020-03-12 | 2020-07-10 | 京东方科技集团股份有限公司 | Entity linking method and device |
CN111708891A (en) * | 2019-03-01 | 2020-09-25 | 九阳股份有限公司 | Food material entity linking method and device among multi-source food material data |
CN112883192A (en) * | 2021-02-09 | 2021-06-01 | 江苏名通信息科技有限公司 | Heterogeneous field user and resource association mining method and system |
CN113157946A (en) * | 2021-05-14 | 2021-07-23 | 咪咕文化科技有限公司 | Entity linking method and device, electronic equipment and storage medium |
CN113190154A (en) * | 2021-04-29 | 2021-07-30 | 北京百度网讯科技有限公司 | Model training method, entry classification method, device, apparatus, storage medium, and program |
CN113268987A (en) * | 2021-05-26 | 2021-08-17 | 北京百度网讯科技有限公司 | Entity name identification method and device, electronic equipment and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8874619B2 (en) * | 2011-06-03 | 2014-10-28 | Robert Mack | Method and apparatus for defining common entity relationships |
CN104933039A (en) * | 2015-06-04 | 2015-09-23 | 中国科学院新疆理化技术研究所 | Entity link system for language lacking resources |
CN105204827A (en) * | 2014-06-17 | 2015-12-30 | 索尼公司 | Information acquisition device and method and electronic equipment |
CN105335519A (en) * | 2015-11-18 | 2016-02-17 | 百度在线网络技术(北京)有限公司 | Model generation method and device as well as recommendation method and device |
CN106934020A (en) * | 2017-03-10 | 2017-07-07 | 东南大学 | A kind of entity link method based on multiple domain entity index |
CN107092605A (en) * | 2016-02-18 | 2017-08-25 | 北大方正集团有限公司 | A kind of entity link method and device |
-
2017
- 2017-09-21 CN CN201710858346.9A patent/CN107506486A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8874619B2 (en) * | 2011-06-03 | 2014-10-28 | Robert Mack | Method and apparatus for defining common entity relationships |
CN105204827A (en) * | 2014-06-17 | 2015-12-30 | 索尼公司 | Information acquisition device and method and electronic equipment |
CN104933039A (en) * | 2015-06-04 | 2015-09-23 | 中国科学院新疆理化技术研究所 | Entity link system for language lacking resources |
CN105335519A (en) * | 2015-11-18 | 2016-02-17 | 百度在线网络技术(北京)有限公司 | Model generation method and device as well as recommendation method and device |
CN107092605A (en) * | 2016-02-18 | 2017-08-25 | 北大方正集团有限公司 | A kind of entity link method and device |
CN106934020A (en) * | 2017-03-10 | 2017-07-07 | 东南大学 | A kind of entity link method based on multiple domain entity index |
Non-Patent Citations (2)
Title |
---|
曾宪宇 等: "用户在线购买预测:一种基于用户操作序列和选择模型的方法", 《计算机研究与发展》 * |
李禹恒 等: "基于微博文本的层次化实体链接方法", 《吉林大学学报(工学版)》 * |
Cited By (34)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108153736A (en) * | 2017-12-28 | 2018-06-12 | 南开大学 | A kind of relative mapping method based on vector space model |
CN108563710A (en) * | 2018-03-27 | 2018-09-21 | 腾讯科技(深圳)有限公司 | A kind of knowledge mapping construction method, device and storage medium |
CN108563710B (en) * | 2018-03-27 | 2021-02-02 | 腾讯科技(深圳)有限公司 | Knowledge graph construction method and device and storage medium |
CN110555137A (en) * | 2018-03-31 | 2019-12-10 | 华为技术有限公司 | Label filling method and device |
CN108763333A (en) * | 2018-05-11 | 2018-11-06 | 北京航空航天大学 | A kind of event collection of illustrative plates construction method based on Social Media |
CN108763333B (en) * | 2018-05-11 | 2022-05-17 | 北京航空航天大学 | Social media-based event map construction method |
CN109408704A (en) * | 2018-09-03 | 2019-03-01 | 平安科技(深圳)有限公司 | Fund data correlating method, system, computer equipment and storage medium |
CN109408704B (en) * | 2018-09-03 | 2023-05-30 | 平安科技(深圳)有限公司 | Fund data association method, system, computer device and storage medium |
CN111708891B (en) * | 2019-03-01 | 2023-12-08 | 九阳股份有限公司 | Food material entity linking method and device between multi-source food material data |
CN111708891A (en) * | 2019-03-01 | 2020-09-25 | 九阳股份有限公司 | Food material entity linking method and device among multi-source food material data |
CN110245197A (en) * | 2019-05-20 | 2019-09-17 | 北京百度网讯科技有限公司 | A kind of the whole network entity associated method and system |
CN110134965B (en) * | 2019-05-21 | 2023-08-18 | 北京百度网讯科技有限公司 | Method, apparatus, device and computer readable storage medium for information processing |
CN110134965A (en) * | 2019-05-21 | 2019-08-16 | 北京百度网讯科技有限公司 | Method, apparatus, equipment and computer readable storage medium for information processing |
CN110147401A (en) * | 2019-05-22 | 2019-08-20 | 苏州大学 | Merge the knowledge base abstracting method of priori knowledge and context-sensitive degree |
CN110704613A (en) * | 2019-08-23 | 2020-01-17 | 上海科技发展有限公司 | Vocabulary database construction and query method, database system, equipment and medium |
CN110795565A (en) * | 2019-09-06 | 2020-02-14 | 腾讯科技(深圳)有限公司 | Semantic recognition-based alias mining method, device, medium and electronic equipment |
CN110795565B (en) * | 2019-09-06 | 2023-10-27 | 腾讯科技(深圳)有限公司 | Alias mining method and device based on semantic recognition, medium and electronic equipment |
CN110705301B (en) * | 2019-09-30 | 2021-01-26 | 北京京东智能城市大数据研究院 | Entity relationship extraction method and device, storage medium and electronic equipment |
CN110705301A (en) * | 2019-09-30 | 2020-01-17 | 京东城市(北京)数字科技有限公司 | Entity relationship extraction method and device, storage medium and electronic equipment |
CN110852106B (en) * | 2019-11-06 | 2024-05-03 | 腾讯科技(深圳)有限公司 | Named entity processing method and device based on artificial intelligence and electronic equipment |
CN110852106A (en) * | 2019-11-06 | 2020-02-28 | 腾讯科技(深圳)有限公司 | Named entity processing method and device based on artificial intelligence and electronic equipment |
CN110968665A (en) * | 2019-11-08 | 2020-04-07 | 浙江工业大学 | Method for recognizing upper and lower level word relation based on gradient enhanced decision tree |
CN110968665B (en) * | 2019-11-08 | 2022-09-23 | 浙江工业大学 | Method for recognizing upper and lower level word relation based on gradient enhanced decision tree |
CN111259087A (en) * | 2020-01-10 | 2020-06-09 | 中国科学院软件研究所 | Computer network protocol entity linking method and system based on domain knowledge base |
CN111259087B (en) * | 2020-01-10 | 2022-10-14 | 中国科学院软件研究所 | Computer network protocol entity linking method and system based on domain knowledge base |
CN111339313A (en) * | 2020-02-18 | 2020-06-26 | 北京航空航天大学 | Knowledge base construction method based on multi-mode fusion |
CN111401049A (en) * | 2020-03-12 | 2020-07-10 | 京东方科技集团股份有限公司 | Entity linking method and device |
CN112883192A (en) * | 2021-02-09 | 2021-06-01 | 江苏名通信息科技有限公司 | Heterogeneous field user and resource association mining method and system |
CN112883192B (en) * | 2021-02-09 | 2023-09-05 | 江苏名通信息科技有限公司 | Heterogeneous domain user and resource association mining method and system |
CN113190154B (en) * | 2021-04-29 | 2023-10-13 | 北京百度网讯科技有限公司 | Model training and entry classification methods, apparatuses, devices, storage medium and program |
CN113190154A (en) * | 2021-04-29 | 2021-07-30 | 北京百度网讯科技有限公司 | Model training method, entry classification method, device, apparatus, storage medium, and program |
CN113157946A (en) * | 2021-05-14 | 2021-07-23 | 咪咕文化科技有限公司 | Entity linking method and device, electronic equipment and storage medium |
CN113268987B (en) * | 2021-05-26 | 2023-08-11 | 北京百度网讯科技有限公司 | Entity name recognition method and device, electronic equipment and storage medium |
CN113268987A (en) * | 2021-05-26 | 2021-08-17 | 北京百度网讯科技有限公司 | Entity name identification method and device, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107506486A (en) | A kind of relation extending method based on entity link | |
US10678816B2 (en) | Single-entity-single-relation question answering systems, and methods | |
CN108573411B (en) | Mixed recommendation method based on deep emotion analysis and multi-source recommendation view fusion of user comments | |
Bhagavatula et al. | Methods for exploring and mining tables on wikipedia | |
CN108415902B (en) | Named entity linking method based on search engine | |
CN106776711B (en) | Chinese medical knowledge map construction method based on deep learning | |
CN111950285B (en) | Medical knowledge graph intelligent automatic construction system and method with multi-mode data fusion | |
CN108763321B (en) | Related entity recommendation method based on large-scale related entity network | |
US9715493B2 (en) | Method and system for monitoring social media and analyzing text to automate classification of user posts using a facet based relevance assessment model | |
CN105893609B (en) | A kind of mobile APP recommended method based on weighted blend | |
CN105528437B (en) | A kind of question answering system construction method extracted based on structured text knowledge | |
CN110059271B (en) | Searching method and device applying tag knowledge network | |
CN107180045B (en) | Method for extracting geographic entity relation contained in internet text | |
Foley et al. | Learning to extract local events from the web | |
CN106960001B (en) | A kind of entity link method and system of term | |
CN110888991B (en) | Sectional type semantic annotation method under weak annotation environment | |
CN103336852B (en) | Across language ontology construction method and device | |
CN102254014A (en) | Adaptive information extraction method for webpage characteristics | |
CN106055675A (en) | Relation extracting method based on convolution neural network and distance supervision | |
Wu et al. | News filtering and summarization on the web | |
CN106934071A (en) | Recommendation method and device based on Heterogeneous Information network and Bayes's personalized ordering | |
CN111325018B (en) | Domain dictionary construction method based on web retrieval and new word discovery | |
CN108090223A (en) | A kind of opening scholar portrait method based on internet information | |
CN110209659A (en) | A kind of resume filter method, system and computer readable storage medium | |
CN110598787B (en) | Software bug classification method based on self-defined step length learning |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171222 |