CN105468791B - An Integrity Expression Method Based on Interactive Q&A Community-Baidu Knows Geographical Entities - Google Patents

Abstract

The present invention relates to a method for expressing the integrity of a geographic location entity based on an interactive question-and-answer community-Baidu, comprising the following steps: step 1): extracting a defective geographic location entity defectLoc through data processing; step 2): generating a question for the extracted defectLoc : "Which area does a certain defectLoc belong to", search through Baidu Know; step 3): extract features according to the search results, calculate the score of each area where the defectLoc belongs to, and construct the feature vector of the area to which the defectLoc belongs; step 4): use the rules Complete the defectLoc. The present invention is based on micro-blog city complaint texts, aiming at the non-standard and unstructured expression of geographic location entities, which makes it difficult for staff to perform statistical analysis work, the present invention proposes a geographic location entity based on Baidu The integrity expression method has a high accuracy rate for the integrity of the defect geographic location entity, which can well meet the needs of practical applications.

Description

An interactive question-and-answer community-Baidu knows the integrity of geographic location entities expression method

technical field

The invention belongs to the technical field of completeness expression of geographic location entities in microblog city complaint texts, and in particular relates to a method for complete expression of geographic location entities based on an interactive question-and-answer community-Baidu Know.

Background technique

In recent years, with the rise of microblogs asking about politics, more and more government departments have set up official microblogs to interact with ordinary people. For microblog city complaint information, due to the huge number of complaint microblogs received every day, geographical location entities sometimes lack regional information. A complete geographical location entity should include two parts of place name area and place name, for example, "R&F Festival Walk, Dougezhuang Township, Chaoyang District". However, the geographical location entity in the Weibo city complaint text has the following phenomena: 1. The place name area is missing, such as "Zhongguancun"; 2. The place name area is vague, such as "Chang'an Avenue". Due to the existence of missing or blurred geographical names, it has brought great difficulties to the staff's statistical analysis work, so that it is difficult for the staff to count the accidents in each area and cannot prevent the occurrence of accidents in time. In the present invention, the geographic location entities with the above two situations are collectively referred to as defective geographic location entities, which are denoted as defectLoc. Moreover, with the passage of time, place names and area information have also changed, making it more difficult to analyze the subordinate areas of place names. For example, "Chongwenmen Xinjingjiayuan" originally belonged to Chongwen District, but now it belongs to Dongcheng District. How to find place names in time Changes in information about the area to which it belongs become particularly important. Integrity representation of geographic location entities, adding missing area information, such as normalizing "Zhongguancun" to "Zhongguancun, Haidian District" or deterministic fuzzy areas such as normalizing "Chang'an Street" to "Chang'an Street, Dongcheng District" or "Xicheng District Chang’an Avenue” can facilitate city managers to conduct statistics and analysis, further discover problems existing in the region, and prevent them, realize the early warning function, and provide decision support for future work.

At present, domestic research is focused on the identification of place names and geographic location entities, and there are few studies on the integrity of geographic location entities. Aiming at the problem of missing regional information, related research mostly solves this problem by constructing geographic ontology and geographic knowledge base. However, the construction of geographic ontology and geographic knowledge base requires the participation of domain experts, and the consistency and integrity of the constructed geographic ontology and geographic knowledge base are maintained. Maintaining such a large geographic ontology and geographic knowledge base requires a lot of manpower. And it is impossible to update the data in time, especially when the affiliation relationship changes, it usually needs to modify more nodes, and it is difficult to achieve real-time performance.

Contents of the invention

Aiming at the problems existing in the above-mentioned prior art, the object of the present invention is to provide a geographical location entity integrity representation method based on the interactive question-and-answer community-Baidu Zhizhi that can avoid the above-mentioned technical defects.

In order to realize the foregoing invention object, the technical scheme that the present invention adopts is as follows;

A method for representing the integrity of geographical location entities based on the interactive question-and-answer community-Baidu Know, comprising the following steps:

Step 1): extract the defect geographic location entity through data processing; wherein, the defective geographic location entity is a geographic location entity with missing or fuzzy regions, which is recorded as defectLoc;

Step 2): Generate a question for the defectLoc extracted in step 1): "Which district does a certain defectLoc belong to", and search through Baidu Zhizhi;

Step 3): extracting features according to the result retrieved in step 2), calculating the scores that defectLoc belongs to each region, and constructing the region feature vector of defectLoc;

Step 4): Use the rules to complete the defectLoc to realize the integrity representation of the geographic location entity.

Further, the step 1) is specifically:

Step A: Analyze the identified geographic location entity to determine whether there is area information, exit if it exists; if not, go to step B;

Step B: Locate the original Weibo, segment the words of the original Weibo through NLPIR, and extract all the contents of @ to form an @ array, determine whether there is unique area information in the array, complete the defectLoc if it exists, and filter it ; does not exist go to step C;

Step C: extract the defectLoc to be processed to form a defectLoc set.

Further, the features extracted in the step 3) are specifically:

Feature 1: content features;

Feature 2: Baidu knows the feature;

Feature 3: Search Feedback Feature.

Further, the feature one is specifically:

(1) Whether there is regional information in the feedback question and answer pair;

The score ScoreA of the area is shown in formula (1):

ScoreA(QA _j ，area _i )

＝(1-λ)×(area _i /10)+λ×(area _i ％10) (1)

Among them, i is the i-th area, j is the j-th question-answer pair that Baidu knows the feedback, and λ is the weight of the area information in the answer; area _i is calculated as shown in formula (2) (3):

(2) Question similarity set;

The problem similarity set is recorded as Simq={simq ₁ , simq ₂ ,..., simq ₁₀ }, wherein, simq _1-10 is the similarity between the proposed question tq and the question-and-answer pair QA set of feedback from Baidu, where The calculation formula is shown in formula (4):

Among them, A and B are two n-dimensional vectors, A is [A1, A2, ..., An], B is [B1, B2, ..., Bn], A _i and B _i indicate that the same character appears in A and B respectively The frequency of , n is all non-repeating single characters in A and B.

Further, the second feature is specifically:

(1) Whether it is a recommended answer;

in, Indicates the weight of the recommended answer;

(2) Number of likes;

ScoreI(QA _i ,Agree)=θ×count(QA _i ,Agree) (6)

Where θ is the weight of each like, and count(QA _i , Agree) is the number of likes in the i-th QA.

(3) Answer time;

Limit the answering time, the unit is year, the calculation formula is shown in formula (7)(8):

time _i = Now-Ans Time _i (7)

Where i is the i-th QA, Now is the current time, and AnsTime is the time to answer the question.

Further, the third feature is specifically:

The first three query results of the feedback results are regarded as having the same weight, and the weight of the latter results gradually decreases as the ranking increases. The specific distribution is shown in formula (9), where i is the i-th QA pair.

Further, the step 3) is specifically:

The missing geographic location entity defectLoc belongs to the score Score(area _i |defectLoc) of area i, and the calculation formula is shown in formula (10):

Among them, RowScore(QA _j , area _i ) is the score of the area i to which the jth QA belongs, and the calculation formula is shown in formula (11):

RowScore(QA _j ，area _i )

=ScoreA(QA _j , area _i )×simq _j ×(1+Rec(j))

×(1+ScoreI(QA _j ，Agree))×(1+ScoreT(time _j ))

×(1+Pos(j)) (11)

According to the score value Score of all areas of defectLoc, the score feature vector of defectLoc is finally constructed

{

Score(area ₁ |defectLoc), Score(area ₂ |defectLoc),..., Score(area ₁₆ |defectLoc)

}.

Further, the rules in step 4) are specifically:

Rule 1: There are two situations for a clear geographic location entity. First, if the search result contains only one area information, this area information is the area information of defectLoc; second, if there is Max(P(area _i |defectLoc ))≥γ, this area _i is the area information of defectLoc;

Among them, the clear geographic location entity is the defectLoc that appears in the search results and only appears in one area, or Max(P(area _i |defectLoc))≥γ, which is recorded as clearLoc; the probability calculation formula is shown in formula (12):

Rule 2: For ambiguous geographic location entities, use countLoc to disambiguate defectLoc; where countLoc counts the number of each area, and multiple identical area information appears in one QA, calculate once, and finally get Max(countLoc|area _i ), the area information of defectLoc is area _i : if there are 2 or more areas in Max(countLoc|area _i ), take the area information of the first Max(countLoc|area _i );

Among them, the ambiguous geographic location entity is the defectLoc in which multiple areas appear in the retrieval results and Max(P(area _i |Location))<γ, which is recorded as ambiguityLoc;

Rule 3: For entities with zero geographic location, region completion cannot be performed;

The zero geographic location entity is the defectLoc that does not appear in the search results, and is denoted as zeroLoc.

The method for expressing the completeness of geographical location entities based on the interactive question-and-answer community-Baidu Knowing provided by the present invention is based on micro-blog city complaint texts, aiming at the non-standard and unstructured characteristics of geographical location entities, which makes it very difficult for staff It is difficult to carry out statistical analysis work. The present invention proposes a completeness expression method based on the geographic location entity known by Baidu, which has a high accuracy rate for the integrity of the defective geographic location entity and can well meet the needs of practical applications.

Description of drawings

Fig. 1 is a flowchart of the present invention;

Figure 2 shows the proportion of defectLoc categories.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

As shown in Figure 1, a method for representing the integrity of geographic location entities based on the interactive question-and-answer community-Baidu Know includes the following steps:

Step 1): extract the defect geographic location entity through data processing; wherein, the defective geographic location entity is a geographic location entity with missing or fuzzy regions, which is recorded as defectLoc;

The present invention first utilizes the geographic location entity automatic recognition method proposed by Li×w to identify the geographic location entity, and after the geographic location entity is identified, defectLoc is further extracted. When a user posts a complaint on Weibo, in addition to "@北京12345", sometimes he will also @ the relevant area. According to the characteristics of relevant areas of microblog @, the present invention extracts the content of all complained microblog @, when there is unique area information in the content of @, such as "@ Chaoyang District Government Hotline", this area is taken as the belonging of this defectLoc The integrity of the area is represented, and finally a part of defectLoc is filtered. The algorithm for extracting the pending defectLoc is as follows:

Step A: Analyze the identified geographic location entity to determine whether there is area information, exit if it exists; if not, go to step B;

Step B: Locate the original Weibo, segment the words of the original Weibo through NLPIR, and extract all the contents of @ to form an @ array, determine whether there is unique area information in the array, complete the defectLoc if it exists, and filter it ; does not exist go to step C;

Step C: extract the defectLoc to be processed to form a defectLoc set.

Step 2): Generate a question for the defectLoc extracted in step 1): "Which district does a certain defectLoc belong to", and search through Baidu Zhizhi;

Baidu Knows is one of the most popular Chinese interactive question-and-answer communities. From 2005 to 2015, Baidu Knows has solved more than 377 million questions in a total of ten years. In just two years after its establishment, Baidu Zhizhi generated 17,596,864 problems and solved 17,012,767 problems, with a problem solving rate of 96.7%. At the same time, Baidu Zhizhi is a knowledge community with strong participation rate and interaction. More than 10 million users visit Zhizhi every day, generating an average of 71,308 questions and 223,907 answers every day, and each question attracts an average of 3.14 users to participate in the interaction. Since Baidu knows that it has a large user base and Q&A data, it is very suitable to solve the problem of region completion of defectLoc.

The present invention mainly utilizes the open interactive question-and-answer community-Baidu Zhizhi to generate a question for the defectLoc extracted in step 1), the question being "which district does the defectLoc belong to?", for example, "which district does Zhongguancun belong to", through "zhidao.baidu. com" to search for related issues, realize the search for the area to which defectLoc belongs, and finally present the feedback results as structured data. For example, for "Zhongguancun", first submit "Which district does Zhongguancun belong to" as a search string to Baidu Know, Feedback a set of QA pairs with 10 similar questions, and make a structured representation of the feedback results, as shown in Table 1 is the first 6 sets of QA pairs intercepted.

Table 1: "Zhongguancun" structured data representation

Step 3): extracting features according to the result retrieved in step 2), calculating the scores that defectLoc belongs to each region, and constructing the region feature vector of defectLoc;

Step 4): Use the rules to complete the defectLoc to realize the integrity representation of the geographic location entity.

Wherein, the feature extracted in the step 3) is specifically:

Feature 1: content features;

This feature describes that Baidu knows the content of the question and answer feedback. First, it is necessary to confirm whether there is regional information in the content of the question and answer. At the same time, if the feedback question has a high similarity with the proposed question, it is considered that the regional information in the question and answer is more important.

Feature 2: Baidu knows the feature;

Baidu Zhizhi features refer to some attributes of Baidu Zhizhi itself, reflecting the credibility of Baidu Zhizhi's QA pairs.

Feature 3: Search Feedback Feature.

According to the order of the search feedback results, the weight is calculated by using the pseudo-feedback technology of the search engine. Baidu knows that the higher the ranking in the feedback results, the more relevant it is to the area information of defectLoc.

(1) Whether there is regional information in the feedback question and answer pair;

Construct a bag={QA ₁ , QA ₂ ,...,QA ₁₀ } according to the feedback question-answer pair, and the target area set is Area={area ₁ , area ₂ ,...,area ₁₆ }, where QA _i is the number i question-answer pairs, each QA corresponds to an Area set. As a judging question, the present invention uses tens and ones 1 to respectively represent whether there is an area _i in the question and the answer, as shown in formula (1)(2):

For different areas, each QA constructs a set containing all 16 areas. Since the areas appearing in the QA question are different from the areas appearing in the answer, the answer is the answer to the missing area problem. Therefore, the area in the answer Information is more important. The score ScoreA of the area is shown in formula (3):

ScoreA(QA _j ，area _i )

＝(1-λ)×(area _i /10)+λ×(area _i ％10) (3)

Where i is the i-th area, j is the j-th question-answer pair that Baidu knows to feedback, and λ is the weight of the area information in the answer;

(2) Question similarity set;

This feature is used to measure the similarity between the proposed question tq and all questions in the QA set, recorded as Simq, then Simq={simq ₁ , simq ₂ ,..., simq ₁₀ }, where simq _1-10 is tq and The similarity of each question in the QA set. Since the result of the cosine similarity is only between [0, 1], and the two questions that need to calculate the similarity have fewer words, usually around 10 words, the present invention uses words as vectors and cosine similarity as Question similarity calculation method. Suppose A and B are two n-dimensional vectors, A is [A1, A2, ..., An], B is [B1, B2, ..., Bn], where A _i and B _i represent the same character in The frequency of occurrences in A and B, n is all non-repeating single characters in A and B, then the cosine similarity between A and B can be expressed as:

Wherein, the second feature is specifically:

(1) Whether it is a recommended answer;

The recommended answer is a better-quality answer recommended by senior Zhizhi netizens on the Baidu Zhizhi platform. Therefore, recommended answers usually have higher credibility and are more important than other answers, and φ represents the weight of recommended answers.

(2) Number of likes;

In Baidu Zhizhi, other users' "agreements" can affirm the accuracy of the answer through the behavior of thumbs up. Generally, the more likes the answer has, the higher its quality is. The calculation of the number of likes in the present invention is expressed as:

ScoreI(QA _i ,Agree)=θ×count(QA _i ,Agree) (6)

Where θ is the weight of each like, and count(QA _i , Agree) is the number of likes in the i-th QA.

(3) Answer time;

The answer time comes from the time when the user answers the question in the QA pair. Because the regional attribution problem of the geographic location changes with time, usually the closer to the answer at the current time, the higher its accuracy. Therefore, the answer time of the present invention Make a limit, the unit is year, the calculation formula is shown in formula (7) (8):

time _i = Now-Ans Time _i (7)

Where i is the i-th QA, Now is the current time, and AnsTime is the time to answer the question.

Among them, the third feature is specifically:

The first three query results of the feedback results are regarded as having the same weight, and the weight of the latter results gradually decreases as the ranking increases. The specific distribution is shown in formula (9), where i is the i-th QA pair.

According to the existence of regional information, question similarity, recommendation, number of likes, answer time and feedback ranking results, a scoring model for the region to which defectLoc belongs to each QA is constructed, in which whether there is regional information and question similarity as its base score , and according to the importance of different features, the calculated score is modified every time a feature is added. If the feature value is 0, the total score remains unchanged. On the contrary, the larger the feature value, the more the total score increases. The step 3) is specifically:

The missing geographic location entity defectLoc belongs to the score Score(area _i |defectLoc) of area i, and the calculation formula is shown in formula (10):

Among them, RowScore(QA _j , area _i ) is the score of the area i to which the jth QA belongs, and the calculation formula is shown in formula (11):

RowScore(QA _j ，area _i )

=ScoreA(QA _j , area _i )×simq _j ×(1+Rec(j))

×(1+ScoreI(QA _j ，Agree))×(1+ScoreT(time _j ))

×(1+Pos(j)) (11)

According to the score value Score of all areas of defectLoc, the score feature vector of defectLoc is finally constructed

{

Score(area ₁ |defectLoc), Score(area ₂ |defectLoc),..., Score(area ₁₆ |defectLoc)

}.

Table 2: Score feature vectors for all regions of defectLoc

Through the observation and analysis of the data, it is found that the Score(area _i |defectLoc) value of the defective geographic location entity and the number of areas appearing in the retrieval results play a decisive role in the integrity of the defective geographic location entity. The invention utilizes the rules to represent the region completeness of different types of defective geographic location entities. Wherein, the rules in the step 4) are specifically:

Rule 1: There are two situations for a clear geographic location entity. First, if the search result contains only one area information, this area information is the area information of defectLoc; second, if there is Max(P(area _i |defectLoc ))≥γ, this area _i is the area information of defectLoc; as in 6 in Table 2, although there are multiple area scores, the area to which it belongs can be determined;

Among them, the clear geographic location entity is the defectLoc that appears in the search results and only appears in one area, or Max(P(area _i |defectLoc))≥γ, which is recorded as clearLoc; the probability calculation formula is shown in formula (12):

Rule 2: For ambiguous geographic location entities, use countLoc to disambiguate defectLoc; where countLoc counts the number of each area, and multiple identical area information appears in one QA, calculate once, and finally get Max(countLoc|area _i ), the area information of defectLoc is area _i ; if there are 2 or more areas in Max(countLoc|area _i ), take the area information of the first Max(countLoc|area _i ); as in Table 3 2. The countLoc of Haidian is the maximum value of 7, and the result of the final integrity normalization is "Wulu Residence, Haidian District";

Among them, the ambiguous geographic location entity is the defectLoc in which multiple areas appear in the retrieval results and Max(P(area _i |Location))<γ, which is recorded as ambiguityLoc;

Table 3: countLoc for all areas of defectLoc

Rule 3: For zero geographic location entities, region completion operations cannot be performed; since such geographic location entities do not necessarily belong to the Beijing region, such as 3 in Table 2.

The zero geographic location entity is the defectLoc that does not appear in the search results, and is denoted as zeroLoc.

For each defective geographic location entity, classify each defective geographic location entity through its all area scores, and then complete the defective geographic location entity through the above rules, and finally normalize it into a complete geographic location entity, as shown in Table 4.

Table 4: Entity Integrity Representation for Defective Geolocation in Table 2

The corpus of the present invention comes from Sina Weibo, with "@北京12345" as the keyword, searches through the search page "s.weibo.com" of Sina Weibo, and writes a directional crawler program to automatically collect related Weibo. Since the geographical locations of complaints about microblogs are concentrated in the Beijing area, the geographical location entities include 14 districts and 2 counties, namely Dongcheng District, Xicheng District, Chaoyang District, Fengtai District, Shijingshan District, Haidian District, and Mentougou District , Fangshan District, Daxing District, Changping District, Shunyi District, Tongzhou District, Huairou District, Pinggu District, Miyun County, Yanqing County.

The present invention uses 1480 Sina city complaint microblogs as experimental corpus, extracts 1482 geographic location entities in total according to the method of Li×w, and proofreads them by professionals. Among them, 840 place names contain clear regional information, which can provide help for subsequent statistics, and there are 642 defective geographic location entities, accounting for 43.32% of the entire corpus. After the previous data processing, according to the Weibo characteristics of @related area information, among the 642 defective geographic location entities, 218 defective geographic location entities can be represented in integrity, and the remaining 424 defective geographic location entities cannot be completed. express. However, 90 of these 424 defective geographic location entities have repeated occurrences, such as "Guomao", "Shuangjing" and other common geographic location entities. After removing these duplicates, there are a total of 334 defective geographic location entities that need to be completed. express.

From the above data, it can be seen that the integrity research of the geographic location entity is necessary, and the present invention mainly conducts the integrity research on 334 defective geographic location entities. After repeated experiments, the regional information in the answer usually contributes more to the region than the regional information in the question. The recommended answer is more authoritative for the explanation of the question, while the number of likes in the Baidu know feature has a smaller contribution to the region. For the defective geographic location entity, if there is a region whose score exceeds or is equal to half of the sum of all regional scores, it can be determined as a clear geographic location entity. Therefore, the present invention takes λ=0.7, θ=0.1, γ=0.5.

The present invention uses accuracy (Accuracy) to evaluate the experimental results, that is, the ratio of the number of correct and complete defective geographic location entities to all defective geographic location entities, and its calculation method is:

Among them, right represents the number of defective geographic location entities that are correctly completed, and total represents the number of all defective geographic location entities to be completed.

Need to carry out complete representation to 334 defective geographic location entities through the data processing stage, the experimental method of the present invention is divided into following 3 steps and carries out:

1) Retrieve questions and give structured feedback results. Through data processing, the present invention needs to perform problem retrieval on 334 defectLocs, and perform structural processing on the problem retrieval results according to the structure in Table 1, and finally form 334 feedback data tables.

2) Feature extraction, calculate the scores of all regions, and construct the score feature vector of defectLoc. The present invention adopts the above-mentioned eigenvalue calculation method and the scoring model of the corresponding area, calculates the scores of each area of each defectLoc through the feedback data table, and constructs the score feature vector.

3) According to the score feature vector of defectLoc, all defectLocs are classified, and the completeness is represented by rules. The present invention classifies and expresses 334 defectLocs, including 290 definite geographic location entities, 35 ambiguous geographic location entities, and 9 zero geographic location entities. As shown in Figure 2, clearLoc accounts for 87% of all defectLocs, indicating that most of the defectLocs in urban complaint microblogs are clearLocs, and although zeroLocs that cannot be completed account for only 3%, other methods still need to be found. Integrity representation.

It can be seen from the experimental results in Table 5 that the accuracy rate of the complete clearLoc of the method of the present invention reaches 96.21%, and the accuracy rate of ambiguityLoc reaches 85.71%. The completeness of clearLoc is through rule 1. Since Baidu knows that the retrieved area is the only area or Max(P(area _i |defectLoc))≥γ, there will be basically no ambiguous area information, so the accuracy rate is the highest. The accuracy of ambiguityLoc's completeness is slightly lower than that of clearLoc, mainly because there are multiple ambiguous regions and the scores are relatively close, so errors sometimes occur in the process of disambiguating multiple regions. The method of the invention can realize integrity representation for most defectLocs, and the coverage rate reaches 97.31%. For a small number of zeroLocs that do not return retrieval results, the method of the present invention is powerless. To sum up, the method of the present invention is suitable for the integrity representation of defectLoc.

Table 5: Distribution table and correct rate of various types in defect geographic location entities

The method for expressing the completeness of geographical location entities based on the interactive question-and-answer community-Baidu Knowing provided by the present invention is based on micro-blog city complaint texts, aiming at the non-standard and unstructured characteristics of geographical location entities, which makes it very difficult for staff It is difficult to carry out statistical analysis work. The present invention proposes a completeness expression method based on the geographic location entity known by Baidu, which has a high accuracy rate for the integrity of the defective geographic location entity and can well meet the needs of practical applications.

The above-mentioned embodiments only express the implementation manner of the present invention, and the description thereof is relatively specific and detailed, but should not be construed as limiting the patent scope of the present invention. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention should be based on the appended claims.

Claims (1)

1. A method for expressing the integrity of a geographic location entity based on an interactive question-and-answer community-Baidu, characterized in that it comprises the following steps: Step 1): extract the defect geographic location entity through data processing; wherein, the defective geographic location entity is a geographic location entity with missing or fuzzy regions, which is recorded as defectLoc; Step 2): Generate a question for the defectLoc extracted in step 1): "Which district does a certain defectLoc belong to", and search through Baidu Zhizhi; Step 3): extracting features according to the result retrieved in step 2), calculating the scores that defectLoc belongs to each region, and constructing the region feature vector of defectLoc; The step 3) is specifically: The missing geographic location entity defectLoc belongs to the score Score(area _i |defectLoc) of area i, and the calculation formula is shown in formula (10): Among them, RowScore(QA _j , area _i ) is the score of the area i to which the jth QA belongs, and the calculation formula is shown in formula (11): RowScore(QA _j ，area _i ) =ScoreA(QA _j , area _i )×simq _j ×(1+Rec(j)) ×(1+ScoreI(QA _j ，Agree))×(1+ScoreT(time _j )) ×(1+Pos(j)) (11) According to the score value Score of all areas of defectLoc, the score feature vector of defectLoc is finally constructed { Score(area ₁ |defectLoc), Score(area ₂ |defectLoc),..., Score(area ₁₆ |defectLoc) }; Among them, i is a positive integer, and 1≤i≤16; Step 4): Use the rules to complete the defectLoc to realize the integrity representation of the geographic location entity; Described step 1) is specifically: Step A: Analyze the identified geographic location entity to determine whether there is area information, exit if it exists; if not, go to step B; Step B: Locate the original Weibo, segment the words of the original Weibo through NLPIR, and extract all the contents of @ to form an @ array, determine whether there is unique area information in the array, complete the defectLoc if it exists, and filter it ; does not exist go to step C; Step C: Extract the defectLoc to be processed to form a defectLoc set; The feature extracted in the step 3) is specifically: Feature 1: content features; Feature 2: Baidu knows the feature; Feature 3: Search Feedback Feature; Described characteristic one is specifically: (1) Whether there is regional information in the feedback question and answer pair; The score ScoreA of the area is shown in formula (1): ScoreA(QA _j ，area _i ) ＝(1-λ)×(area _i /10)+λ×(area _i ％10) (1) Where i is the i-th area, j is the j-th question-answer pair that Baidu knows the feedback, λ is the weight of the area information appearing in the answer, λ=0.7; area _i is calculated as shown in formula (2) (3): Among them, QA is the question-and-answer pair that Baidu knows the feedback; (2) Question similarity set; The problem similarity set is recorded as Simq={simq ₁ , simq ₂ ,..., simq ₁₀ }, wherein, simq _1-10 is the similarity between the proposed question tq and the question-and-answer pair QA set of feedback from Baidu, where The calculation formula is shown in formula (4): Among them, A and B are two n-dimensional vectors, A is [A1, A2, ..., An], B is [B1, B2, ..., Bn], A _i and B _i indicate that the same character appears in A and B respectively The frequency of , n is all non-repeating single characters in A and B; The second feature is specifically: (1) Whether it is a recommended answer; in, Indicates the weight of the recommended answer, (2) Number of likes; ScoreI(QA _i ,Agree)=θ×count(QA _i ,Agree) (6) Where θ is the weight of each like, θ=0.1, count(QA _i , Agree) is the number of likes in the i-th QA; (3) Answer time; Limit the answering time, the unit is year, the calculation formula is shown in formula (7)(8): time _i = Now-Ans Time _i (7) Where i is the i-th QA, Now is the current time, and AnsTime is the time to answer the question; The third feature is specifically: The first three query results of the feedback results are regarded as having the same weight, and the weight of the latter results gradually decreases as the ranking increases. The specific distribution is shown in formula (9), where i is the i-th QA pair; The rules in the step 4) are specifically: Rule 1: There are two situations for a clear geographic location entity. First, if the search result contains only one area information, this area information is the area information of defectLoc; second, if there is Max(P(area _i |defectLoc ))≥γ, this area _i is the area information of defectLoc; where γ=0.5; Among them, the clear geographic location entity is the defectLoc that appears in the search results and only appears in one area, or Max(P(area _i |defectLoc))≥γ, which is recorded as clearLoc; the probability calculation formula is shown in formula (12): Rule 2: For ambiguous geographic location entities, use countLoc to disambiguate defectLoc; where countLoc counts the number of each area, and multiple identical area information appears in one QA, calculate once, and finally get Max(countLoc|area _i ), the area information of defectLoc is area _i ; if there are 2 or more areas in Max(countLoc|area _i ), take the area information of the first Max(countLoc|area _i ); Among them, the ambiguity geographic location entity is the defectLoc in which multiple areas appear in the retrieval results and Max(P(area _i |Location))<γ, which is recorded as ambiguityLoc; Rule 3: For entities with zero geographic location, region completion operations cannot be performed; The zero geographic location entity is the defectLoc that does not appear in the search results, which is denoted as zeroLoc.