CN108415902A - A kind of name entity link method based on search engine - Google Patents

Abstract

The invention discloses a method for linking named entities in a search engine, comprising the following steps: segmenting input query text, identifying a group of named entities in the text that need to be linked to a knowledge base; To name an entity, search for a list of candidate entities in the Chinese entity knowledge base through entity reference; use the search engine to expand the context information of the entity reference; use the search engine to expand the context information of each candidate entity in the candidate entity list; calculate the entity reference and The matching degree between each candidate entity; the entity with the highest matching degree is selected for linking. This method retrieves entity references and candidate entities in a search engine, obtains information from search results to expand the context information of entity references and candidate entities, and provides additional information for improving the accuracy of entity links.

Description

A Method of Named Entity Linking Based on Search Engine

technical field

The invention relates to a network information processing method, in particular to a named entity linking method based on a search engine.

Background technique

With the continuous improvement of informatization, more and more people obtain information through the Internet, and traditional information dissemination media such as newspapers, magazines, and periodicals are gradually surpassed by portal websites, electronic libraries, and search engines. Most of the information obtained through the Internet is text information disseminated through platforms such as news websites, microblogs, and post bars. These texts contain a large number of named entities. growth of. How to efficiently and accurately link these large numbers of named entities to the corresponding entities in the knowledge base without ambiguity has become an urgent problem to be solved in the fields of information fusion, natural language processing, and information retrieval.

Named entity linking is mainly divided into two stages. The first stage is mainly to search for potential candidate entity sets in the knowledge base according to entity references. The second stage is mainly to disambiguate entities based on entity references and candidate entity sets. At present, the main work of the named entity linking method focuses on optimizing the entity disambiguation algorithm. The main methods of entity disambiguation are mainly divided into disambiguation methods based on entity popularity ranking, disambiguation methods based on context Disambiguation methods, graph-based disambiguation methods, etc.

The disambiguation method based on entity popularity ranking is the simplest and most direct entity disambiguation method, which returns the most popular entity according to the entity popularity judgment. For example, a person entity, the higher the popularity of the person, the greater the possibility of being linked. This popularity is usually measured by the number of links related to the entity, but the disadvantage of this method is that no matter whether the query entity is popular or not, it will returns the same result.

Disambiguation methods based on contextual similarity sorting use the similarity between the context of entity references and the context of candidate entities to determine which entities should be linked. This method usually expresses the context of the entity reference and the context of the candidate entity as a word bag vector or expresses some keywords such as named entities in the text as a word space vector, and then calculates the similarity between the vectors, such as cosine similarity. Its disadvantage is that the calculation of similarity depends on word co-occurrence information and requires a large amount of entity context information. When the context information is less, the accuracy of the judgment result is not high.

The classification-based disambiguation method regards the linking process as a classification process. During the training phase, the <entity designation, candidate entity> pair is extracted from the knowledge base and the entity in the knowledge base is linked as a positive instance, and no link is established. The pair of <entity designation, candidate entity> is used as a negative instance. Then, use the training set composed of these positive examples and negative examples to train a classifier. The common classification models used by the classifier include SVM, decision tree, naive Bayes, etc. In the named entity linking stage, the trained classifiers are used to classify the candidate entities referred to by the entities. However, this method is easy to get two or more linked entities, which needs to be sorted again by combining context similarity and other methods.

The graph-based disambiguation method uses entity references and candidate entities as graph nodes, and uses the relationship between entity references and candidate entities as graph edges to establish a graph structure. Usually, when calculating the relationship between an entity referring node and its candidate entity nodes, most methods utilize the context similarity feature between the two. Therefore, graph-based disambiguation methods also rely on word co-occurrence information to a certain extent, and there is also the problem that the accuracy of judgment results is not high when there is less context information.

To summarize the entity disambiguation methods mentioned above, most methods of named entity linking rely on the similarity feature between the entity reference context and the candidate entity context. However, the calculation of this feature requires a large amount of context information of the entity reference and its candidate entities. Overcome the problem of over-reliance on word co-occurrence information when calculating similarity. In order to obtain more context information of entity references and candidate entities, a method that can expand the context information of entity references and candidate entities is needed to improve the accuracy of calculation results.

Contents of the invention

The purpose of the present invention is to provide a method for linking named entities based on a search engine to solve the problems in the prior art.

In order to achieve the above object, the technical scheme adopted in the present invention is:

A method for linking named entities based on a search engine, characterized in that: comprising the following steps:

Step 1, segment the input query text, and identify a group of named entities in the text that need to be linked to the knowledge base;

Step 2, for each named entity, obtain a candidate entity list through entity reference;

Step 3, using the search engine to expand the contextual information of the entity reference;

Step 4, using a search engine to expand the context information of each candidate entity in the candidate entity list;

Step 5, calculating the matching degree between the entity reference and each candidate entity;

Step 6, sort the candidate entities according to their matching degree, and select the candidate entity with the highest matching degree as the target entity for linking.

The described method for linking named entities based on a search engine is characterized in that: in the step 1:

Named entities are person names, organization names, place names, and all other entities identified by names, and entities are objects or concepts in real life;

The query text is the text or text set entered by the user, including the entity reference and its context text, and its text length is a fixed length;

The knowledge base is a collection of several entity entries, including people, institutions, place names and all other entities identified by names;

Entity entry is the specific information stored in the knowledge base for each entity, including the context text of the entity, entity attributes and attribute values, the context information of the entity is the text describing the entity in detail, and the entity reference is the entity to be linked in the query text noun references.

The described method for linking named entities based on a search engine is characterized in that: in the step 2, the method for obtaining the candidate entity by referring to the entity is:

Take out each entity entry in the knowledge base in turn, traverse its name attribute value and alias attribute value, and calculate the string similarity between the entity reference in the query text and the name attribute value and alias attribute value of each entity entry; if the query If the similarity between the entity reference of the text and the name attribute value of the entity entry is greater than the preset threshold, the current entity entry is output to the candidate entity list. If the similarity between the entity reference of the query text and the alias attribute value of the entity entry is greater than the preset After the threshold is set, the entity entry corresponding to the alias attribute is output to the candidate entity list;

Obtain the initial candidate entity list through the above method, and filter the initial candidate entity list to obtain the candidate entity list, the method is:

Extract some entities in the knowledge base as the training set, use the marked <entity, type> pair set to train the entity-type classifier, and then use the entity-type classifier to identify the type of the entity in the candidate entity list, and compare it with the query text The entity to be linked belongs to the entity of the same type and is retained, and the entity of different type is deleted to obtain the final candidate entity list;

The type of the entity is the category attribute of the entity, including people, places, institutions, and time; the <entity, type> pair is a mapping between the entity and its type, and the correct <entity, type> pair is the training classifier A positive example, and a negative example on the contrary; the entity-type classifier is a multi-class classification model trained on a set of <entity, type>, through which the type of the entity can be judged.

The described method for linking named entities based on a search engine is characterized in that: in the step 3, using a search engine to expand the context information referred to by an entity is as follows:

Retrieve the entity references of a group of named entities in the query text in step 1 as seeds in the search engine to obtain the search result pages related to the entity references, and extract the titles and abstracts of the first m search results from the search result pages information to obtain the extended text of the entity reference, and the query text and the extended text together constitute the extended context information of the entity reference.

The described method for linking named entities based on a search engine is characterized in that: in the step 4, using a search engine to expand the context information of each candidate entity in the list of candidate entities is:

Segment the context text of each candidate entity in the knowledge base in the candidate entity list, perform named entity recognition on the text after segmentation to obtain a group of named entities related to the candidate entity, and then make this group of named entities Retrieve a seed in a search engine to obtain the search result webpage related to the candidate entity, extract the title and summary information of the first n search results from the search result webpage, and obtain the extended text of the candidate entity, and the candidate entity is in The context text in the knowledge base together with the extension text constitutes the extended context information of the candidate entity.

The described method for linking named entities based on a search engine is characterized in that: in the step 5, the method for calculating the degree of matching between the entity designation and each candidate entity is:

For each candidate entity in the candidate entity list, calculate the cosine similarity a between the extended context information referred to by the entity and the extended context information of the candidate entity, calculate the candidate entity popularity b, and calculate the entity The matching degree c=w _a ×a+w _b ×b between the reference and each candidate entity, wherein: w _a and w _b are preset weights, and w _a +w _b =1; the candidate entity The popularity of is: the ratio of the number of times the entity reference links to the candidate entity page as a hyperlink to the total number of times the entity reference links to all candidate entities as a hyperlink.

Compared with the prior art, the beneficial effects of the present invention are reflected in:

1. The present invention adopts a two-stage method for generating candidate entity lists. First, calculate the string similarity to get the initial candidate entity list. Then, according to the method of classifying entity types, the initial candidate entity list is screened to obtain the candidate entity list. The method of generating candidate entity lists in two stages can not only generate more candidate entities related to entity references, but also remove some noisy candidate entities through entity type classification to reduce the complexity of disambiguation.

2. The key contribution of the present invention is to provide a method for extending entity references and context text information of candidate entities to solve the problem of named entity linking. The length of entity reference context text is limited, and the entity context information contained in the knowledge base is also limited, which leads to low accuracy in calculating the similarity between entity reference and candidate entity context based on word co-occurrence information. Retrieve entity references and candidate entities through search engines, get all the webpages related to entity references and candidate entities, and get extended context information after sorting out the webpages, which will help us extract more entity features, so as to improve the entity link Accuracy provides additional information.

Description of drawings

FIG. 1 is a flowchart of a method for linking named entities based on a search engine in the present invention.

Detailed ways

A method for linking named entities based on a search engine includes the following steps:

Step 1. Segment the input query text and identify a group of named entities in the text that need to be linked to the knowledge base.

Step 2. For each named entity, obtain a list of candidate entities through entity reference.

Step 3, use the search engine to expand the context information of the entity reference.

Step 4, using the search engine to expand the context information of each candidate entity in the candidate entity list.

Step 5, calculate the matching degree between the entity reference and each candidate entity.

Step 6, sort the candidate entities according to their matching degree, and select the candidate entity with the highest matching degree as the target entity for linking.

In step 1, named entities are person names, organization names, place names, and all other entities identified by names, and entities are objects or concepts in real life; query text is text or text sets entered by users, including entity references and their contexts Text, whose text length is a fixed length, the knowledge base is a collection of several entity entries, including people, institutions, place names and all other entities identified by names; entity entries are the specific information stored in the knowledge base for each entity, including The context text, entity attribute and attribute value of the entity, the context information of the entity is the text describing the entity in detail, and the entity reference is the noun reference of the entity to be linked in the query text.

In step 2, the method of obtaining candidate entities through entity references is: take out each entity entry in the knowledge base in turn, traverse its name attribute value and alias attribute value, and calculate the entity reference in the query text and the name attribute value of each entity entry , the string similarity between the alias attribute values. If the similarity between the entity reference of the query text and the name attribute value of the entity entry is greater than the preset threshold, the current entity entry will be output to the candidate entity list; if the similarity between the entity reference of the query text and the alias attribute value of the entity entry is greater than the preset threshold, the entity entry corresponding to the alias attribute is output to the candidate entity list; the initial candidate entity list is obtained through the above method. Filter the initial candidate entity list to obtain the candidate entity list. The method is as follows: extract some entities in the knowledge base as the training set, and use the labeled <entity, type> pair set to train the entity-type classifier. Then use the entity-type classifier to identify the type of the entity in the candidate entity list, keep the entity of the same type as the entity to be linked in the query text, delete the entity of different type, and obtain the final candidate entity list, the type of the entity is the entity Category attributes, including people, places, institutions, and time; <entity, type> pairs are mappings between entities and their types, where the correctly mapped <entity, type> pair is a positive example for training a classifier, otherwise it is a negative example; The entity-type classifier is a multi-class classification model trained by <entity, type>. The multi-class classification model is to divide the entity into multiple categories. An entity belongs to and only belongs to one of the multiple categories. Between different categories are mutually exclusive, and the type of entity can be judged through the multi-class classification model.

In step 3, using the search engine to expand the context information of entity references is as follows: use the entity references of a group of named entities in the query text in step 1 as seeds to search in the search engine, and obtain the search results related to the entity references The result web page, extracting the title and abstract information of the first m search results from the search result web page to obtain the extended text of the entity reference, and the query text and the extended text together constitute the extended context information of the entity reference.

In step 4, using a search engine to expand the context information of each candidate entity in the candidate entity list is as follows: segment the context text of each candidate entity in the candidate entity list in the knowledge base, and perform word segmentation on the text after word segmentation Named entity recognition obtains a group of named entities related to the candidate entity, and then uses this group of named entities as a seed to retrieve in a search engine to obtain a search result webpage related to the candidate entity, and from the search result webpage The title and abstract information of the first n search results are extracted to obtain the extended text of the candidate entity, and the context text of the candidate entity in the knowledge base and the extended text constitute the extended context information of the candidate entity.

In step 5, the method for calculating the matching degree between the entity reference and each candidate entity is: for each candidate entity in the candidate entity list, calculate the extended context information of the entity reference and the extended context information of the candidate entity the cosine similarity a between them, calculate the candidate entity popularity b, and calculate the matching degree c=w _a ×a+w _b ×b between the entity reference and each candidate entity. Wherein: w _a and w _b are preset weights, and w _a +w _b =1. The popularity of a candidate entity is: the ratio of the number of times the entity reference links to the candidate entity page as a hyperlink to the total number of times the entity reference links to all candidate entities as a hyperlink.

Specific examples:

The present embodiment provides a method for linking named entities based on a search engine. The steps of the method for linking named entities based on a search engine in this embodiment are described below in conjunction with FIG. 1:

(1) As shown in S101 of FIG. 1 , word segmentation is performed on the input query text, and a group of named entities in the text that need to be linked to the knowledge base are identified. A user enters a query text, for example, a paragraph in a news article "Li Na's retirement ceremony and Nadal presented flowers, and sister Na burst into tears to enjoy the cheers of the audience." First, the input query text is segmented, and then the named entity recognition tool is used to Identify a set of named entities in text. Among them, named entities are people's names, organization names, place names, and all other entities identified by names, and broader entities include numbers, dates, currencies, and addresses.

Carrying out word segmentation to the input query text is to utilize the word segmentation tool to divide the query text into words according to the Chinese dictionary. In the present embodiment, the used tool for word segmentation is the NLPIR Chinese word segmentation tool provided by the Institute of Computing Technology, Chinese Academy of Sciences (web site: http:// /ictclas.nlpir.org/downloads). The set of words and their types marked by word segmentation is: W={Li Na/nr retires/vi ceremony/n Nadal/nr presents flowers/vi, /wd Najie/nr tears/n Ben/v enjoys/vg audience/ n cheers /v}, where "/nr", "/vi", "/n", "/wd", "/v", "/vg" are the type annotations of words.

After word segmentation obtains a set of words, use the named entity recognition technology to identify the entity set contained in the text after word segmentation: E={Li Na, Nadal, Najie}, the named entity recognition method adopted in this embodiment is based on word segmentation The obtained words and their types are marked, and the words whose types are marked as nr, ns, nt, and nz are extracted to form a named entity set. The "Li Na", "Nadal" and "Najie" here are the entity references of the identified named entities "Li Na", "Nadal" and "Najie". This embodiment introduces the named entity linking process by taking the linking process of the entity referring to "Li Na" as an example.

(2) As shown in S102 of FIG. 1 , for each named entity, a candidate entity list is obtained through entity reference. Take out each entity entry in the knowledge base in turn, traverse its name attribute value and alias attribute value, and calculate the string similarity between the entity reference in the query text and the name attribute value and alias attribute value of each entity entry. If the similarity between the entity reference of the query text and the name attribute value of the entity entry is greater than the preset threshold, the current entity entry will be output to the candidate entity list; if the similarity between the entity reference of the query text and the alias attribute value of the entity entry is greater than the preset threshold, the entity entry corresponding to the alias attribute is output to the candidate entity list. The initial candidate entity list is obtained through the above method. The knowledge base used in this embodiment is the Chinese knowledge base issued by Tsinghua University (URL: http://keg.cs.tsinghua.edu.cn/project/ChineseKB), which contains 800,000 different entities and their attributes, names The attribute value is included in the rdf:about attribute of the <rdf:Description rdf:about=""></rdf:Description> tag, and the alias attribute is included in the <ont:alias></ont:alias> tag. In this embodiment, the preset threshold value is 0.7. For example, the initial candidate entity entries that meet the threshold obtained by searching for the entity reference of Li Na are shown in Table 1:

Table 1 The initial candidate entity list for the entity referred to as "Li Na"

In Table 1, the entity name is the name attribute value of the entity entry in the extracted knowledge base, which is included in the rdf:about attribute of the <rdf:Description rdf:about=""></rdf:Description> tag, and the entity context is The context attribute value of the entity entry in the extracted knowledge base, which is included in the <ont:ABSTRACT></ont:ABSTRACT> tag, and other attributes are other attribute values of the entity entry in the knowledge base, included in except <rdf: Descriptionrdf:about=""></rdf:Description>, <ont:ABSTRACT></ont:ABSTRACT> tags in other tags.

Then, filter the initial candidate entity list to obtain the candidate entity list. The method is: extract some entities in the knowledge base as the training set, and use the marked "<entity, type>" set to train the "entity-type" classifier. The method for extracting some entities in the knowledge base as the training set in this embodiment is: randomly extract 200 entity entries from each type of entity entry set as the training set, and the classifier model adopted is SVM (Support Vector Machine) . Then use the "entity-type" classifier to identify the type of entities in the candidate entity list, keep the entities of the same type as the entities to be linked in the query text, delete entities of different types, and obtain the final candidate entity list. The entity type is the category attribute of the entity, including person, place, institution, and time. The list of candidate entities after screening is shown in Table 2:

Table 2 A list of candidate entities referred to as "Li Na"

(3) As shown in S103 of FIG. 1 , use a search engine to expand the context information of entity references. Taking the entity reference "Li Na" as an example, the query text of the entity reference is "Li Na's retirement ceremony Nadal presented flowers, sister Na burst into tears to enjoy the cheers of the audience.", the entity reference of a group of named entities identified from the query text is " Li Na, Nadal, and Sister Na" are searched as the seeds of the search engine, and the search result webpage related to the entity alleged "Li Na" is obtained, and the title and abstract information of the first m search results are extracted from the search result webpage, and the results are as shown in the table The entity referred to in 3 is the extended text of "Li Na", and the query text and the extended text together constitute the extended context information of the entity referred to as "Li Na". In this embodiment, the search engine is the Bing search engine, and the title and summary information of the search results are extracted using a CSS selector, and the value of m is 8.

Table 3 The extended text of the entity referring to "Li Na"

(4) As shown in S104 of FIG. 1 , use a search engine to expand the context information of each candidate entity in the candidate entity list. First, word segmentation is performed on the context text of each candidate entity in the knowledge base in the candidate entity list. In this embodiment, the context information of the candidate entity is included in the <ont:ABSTRACT></ont:ABSTRACT> tag. Taking the candidate entity "Li Na (tennis player)" as an example, the result of its context segmentation is W ^* ={Li Na/nr(/wkz 1982/t February/t 26th/t—/wp)/wky,/ wd China/nr women/n tennis/n players/n. /wj 2011/t June/t 4th/t, /wd in/p France/nr/ude1 paris/nr west/f mont/vi ter/d high/a land/n/ude1 Roland/nrf ·Gal/b Luo/b Si/b stadium/n inside/f, /wd Li Na/nr won/v French Open/n women's singles/n champion/n. /wj become the /v history /dl first /m /q win /v big /a grand slam /n tennis match /n event / n champion / ude1 Asia /nr people /n. /wj}, perform named entity recognition on the word-segmented text to obtain a set of named entities related to candidate entities. The named entity recognition method adopted in this embodiment is to extract words whose types are marked as nr, ns, nt, and nz to form a named entity set according to the words obtained by word segmentation and their types. The result of named entity recognition is E ^* ={Li Na, China, France, Paris, Asia}. Then, use this group of named entities "Li Na, China, France, Paris, Asia" as a seed to retrieve in the search engine, and get a group of search result pages related to the candidate entity "Li Na (tennis player)". The title and abstract information of the first n search results are extracted from the result webpage, and the extended text of the candidate entity "Li Na (tennis player)" is obtained as shown in Table 4, and the context of the candidate entity "Li Na (tennis player)" in the knowledge base The text together with the extended text constitutes the extended context information of the candidate entity "Li Na (tennis player)". In this embodiment, the search engine is the Bing search engine, and the title and summary information of the search results are extracted using a CSS selector, and the value of n is 10.

Table 4 The extended text of the candidate entity "Li Na (tennis player)"

(5) As shown in S105 of FIG. 1 , calculate the matching degree between the entity reference and each candidate entity. First, for each candidate entity in the candidate entity list, calculate the cosine similarity a between the extended context information of the entity reference and the extended context information of the candidate entity, calculate the popularity b of the candidate entity, and calculate the relationship between the entity reference and each candidate The degree of matching between entities c=w _a ×a+w _b ×b. Wherein: w _a and w _b are preset weights, and w _a +w _b =1. The popularity of a candidate entity is: the ratio of the number of times the entity reference links to the candidate entity page as a hyperlink to the total number of times the entity reference links to all candidate entities as a hyperlink. In this embodiment, w _a =0.6 and w _b =0.4 are preset. Taking the calculation of the matching degree between the entity reference "Li Na" and the candidate entity "Li Na (tennis player)" in step (3) as an example, the extended context information of the entity reference "Li Na" and the extension of the candidate entity "Li Na (tennis player)" The cosine similarity a=0.53 between the context information of the candidate entity "Li Na (tennis player)", the popularity b=0.88 of the candidate entity "Li Na" and the candidate entity "Li Na (tennis player)" matching degree c =0.6*0.53+0.4*0.88=0.67. Table 5 lists the matching degree between the entity reference "Li Na" and each candidate entity.

Table 5 The matching degree between the entity referring to "Li Na" and each candidate entity

candidate entity cosine similarity Popularity of Candidate Entities suitability Li Na (tennis player) 0.53 0.88 0.67 Li Na (singer) 0.43 0.056 0.28 Li Na (Professor of Peking University) 0.39 0 0.23 Lee Na Young 0.19 0.31 0.24

(6) As shown in S106 of FIG. 1 , the candidate entities are sorted according to their matching degree, and the candidate entity with the highest matching degree is selected as the target entity for linking. In this embodiment, Table 5 is sorted from large to small according to the matching degree between the entity reference and the candidate entity to obtain Table 6. The candidate entity in item 1 of Table 6 is the candidate entity with the highest matching degree, thus it can be obtained The entity refers to the target entity of the "Li Na" link as "Li Na (tennis player)".

Table 6 Sorting results of the candidate entities referred to as "Li Na" according to the matching degree from large to small

candidate entity cosine similarity Popularity of Candidate Entities suitability Li Na (tennis player) 0.53 0.88 0.67 Li Na (singer) 0.43 0.056 0.28 Lee Na Young 0.19 0.31 0.24 Li Na (Professor of Peking University) 0.39 0 0.23

Claims (6)

1. a kind of name entity link method based on search engine, it is characterised in that：Include the following steps：

Step 1, the query text of input is segmented, identifies that one group of name for needing to be linked in text in knowledge base is real Body；

Step 2, it to each name entity, censures to obtain candidate list of entities by entity；

Step 3, the contextual information censured using search engine, extension entity；

Step 4, using search engine, the contextual information of each candidate entity in candidate list of entities is extended；

Step 5, computational entity censures the matching degree between each candidate entity；

Step 6, it is sorted to candidate entity according to the matching degree of candidate entity, selects the maximum candidate entity of matching degree as target Entity is linked.

2. a kind of name entity link method based on search engine according to claim 1, it is characterised in that：The step In rapid 1：

Name entity is name, mechanism name, place name and other all entities with entitled mark, and entity is in actual life Object or concept；

Query text is text input by user or text set, including entity is censured and its context text, and text size is Regular length；

Knowledge base is the set of several entity entries, including personage, mechanism, place name and other all realities with entitled mark Body；

Entity entries are the specifying information that each entity stores in knowledge base, include context text, the entity attribute of entity And attribute value, the contextual information of entity is the text that entity is described in detail, and it is to wait for chain in query text that entity, which is censured, Connect the noun reference of entity.

3. a kind of name entity link method based on search engine according to claim 1, it is characterised in that：The step In rapid 2, it is by the method that entity censures to obtain candidate entity：

Every entity entries in knowledge base are taken out successively, traverse its name attribute value, alias property value, are calculated in query text Entity censure the similarity of character string between the name attribute values of each entity entries, alias property value；If query text Entity censure and be more than pre-set threshold value with the name attribute value similarities of entity entries, then it is current entity entry is defeated Go out into candidate list of entities, is set in advance if the entity of query text censures to be more than with the alias property value similarity of entity entries The threshold value set then is output to the entity entries corresponding to the alias property in candidate list of entities；

Initial candidate list of entities is obtained by the above method, initial candidate list of entities is screened to obtain candidate entity row Table, method are：

The part entity in knowledge base is extracted as training set, utilizes mark<Entity, type>To set training entity-type Then grader utilizes entity-type sorter to identify the type of entity in candidate list of entities, with chain is waited in query text The entity reservation that entity belongs to same type is connect, different types of entity is deleted, obtains final candidate list of entities；

The type of the entity is the category attribute of entity, including personage, place, mechanism, time；It is described<Entity, type>It is right For the mapping of entity and its type, wherein mapping is correct<Entity, type>To the positive example for training grader, otherwise it is Negative sense example；Entity-the type sorter is to pass through<Entity, type>To the multicategory classification model of set training, pass through institute State the type that model may determine that entity.

4. a kind of name entity link method based on search engine according to claim 1, it is characterised in that：The step In rapid 3, using search engine, the method for the contextual information that extension entity is censured is：

The entity of one group of name entity in query text in step 1 is censured and is retrieved in a search engine as seed, obtain with The entity censures relevant search result web page, the title and abstract of m search result before being extracted from search result web page Information, obtains the expanded text that the entity is censured, and query text constitutes the extension that the entity is censured together with expanded text Contextual information.

5. a kind of name entity link method based on search engine according to claim 1, it is characterised in that：The step In rapid 4, using search engine, the method for extending the contextual information of each candidate entity in candidate list of entities is：

Context text of the candidate entity of each of candidate list of entities in knowledge base is segmented, to the text after participle Originally it is named Entity recognition and obtains one group and the relevant name entity of the candidate entity, then this group name entity is done Retrieved in a search engine for a seed, obtain with the candidate relevant search result web page of entity, from search result net The title and summary info of n search result before being extracted in page, obtain the expanded text of the candidate entity, and candidate entity is being known Know the contextual information that the context text in library constitutes the extension of the candidate entity together with expanded text.

6. a kind of name entity link method based on search engine according to claim 1, it is characterised in that：The step In rapid 5, the method that computational entity censures the matching degree between each candidate entity is：

To the candidate entity of each of candidate list of entities, the contextual information for the extension that the entity is censured and the time are calculated The cosine similarity a between the contextual information of the extension of entity is selected, the candidate entity popularity b is calculated, computational entity refers to Claim matching degree c=w between each candidate entity_a×a+w_b× b, wherein：w_a、w_bFor preset weights, and w_a+w_b=1；It is described The popularity of candidate entity is：Entity denotion is linked to the number of the candidate physical page as hyperlink and entity denotion is done The ratio of the total degree of all candidate entities is linked to for hyperlink.