Relationship analysis method and device based on news events

By clustering and relationship extraction of multiple news texts and combining correlation analysis, the problem of in-depth background information cannot be provided in the existing technology, the deep mining of news events and information fusion are achieved, and more comprehensive news understanding is provided. .

CN113158672BActive Publication Date: 2024-11-08TSINGHUA UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110326769.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-03-26
Publication Date
2024-11-08
Estimated Expiration
2041-03-26

AI Technical Summary

Technical Problem

The existing technology is difficult to provide in-depth and comprehensive background information for news events, cannot effectively aggregate different categories of news information, and cannot integrate cross-category information, resulting in readers being unable to obtain deeper or more comprehensive news understanding.

Method used

By obtaining multiple news texts, clustering, entity linking, entity extraction, entity relationship extraction, event relationship extraction and entity-event relationship extraction, combined with correlation analysis, cross-category fusion of information and background knowledge supplementation.

Benefits of technology

It has achieved in-depth exploration of news events, providing more comprehensive and in-depth information, so that readers can understand news events more deeply.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113158672B_ABST
    Figure CN113158672B_ABST
Patent Text Reader

Abstract

The present invention provides a relationship analysis method and device based on news events, wherein the method comprises: obtaining multiple news texts; clustering the multiple news texts to obtain multiple news events after clustering; performing entity linking, entity extraction, entity relationship extraction, event relationship extraction, and entity and event relationship extraction on the multiple news events, and performing association analysis based on the entity linking results, entity extraction results, entity relationship extraction results, event relationship extraction results, and entity and event relationship extraction results to obtain analysis results. The present invention realizes news event level analysis by clustering the original news texts to obtain multiple events, and then extracts entities, entity relationships, event relationships, and event-entity relationships from the event texts. Furthermore, through entity linking, the entity background knowledge is supplemented to realize in-depth mining of the event, so that readers can obtain more comprehensive and in-depth information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a relationship analysis method and device based on news events. Background Art

[0002] With the increasing penetration of the Internet and the degree of information disclosure, the Internet has become an important platform for reporting and disseminating news events. The number of news pages on the Internet has increased dramatically, and the increase in news volume and redundant information has brought new challenges to reading and information analysis. Due to the open nature of the Internet, the news information posted on the website may be complicated and disordered. News information describing the same news event may be scattered on different websites, which is not conducive to users' in-depth understanding. When users want to understand the ins and outs of this news event, they need to enter it again and again to search, which is very time-consuming.

[0003] In order to analyze news events, existing technologies mainly simply classify the news text or extract corresponding keywords based on the news content, which cannot fully reflect the information contained in a piece of news. Or they only cluster news events and display the same events in a centralized manner, so that readers can obtain more relevant information. However, this method cannot aggregate different types of news, cannot integrate information across categories, cannot reflect the background knowledge not included in the news, and cannot enable readers to obtain more in-depth or comprehensive information, thereby deepening their understanding of the news. Summary of the invention

[0004] The present invention provides a relationship analysis method and device based on news events, which are used to solve the defect in the prior art that readers cannot obtain in-depth and comprehensive background information, realize cross-category fusion of information, supplement background knowledge information in news information, and enable readers to understand the news more deeply and comprehensively.

[0005] In a first aspect, the present invention provides a relationship analysis method based on news events, comprising:

[0006] Get multiple news texts;

[0007] Clustering the multiple news texts to obtain multiple clustered news events;

[0008] Perform entity linking, entity extraction, entity relationship extraction, event relationship extraction, and entity and event relationship extraction on the multiple news events to obtain entity linking results, entity extraction results, entity relationship extraction results, event relationship extraction results, and entity and event relationship extraction results;

[0009] An analysis result is obtained by performing association analysis on the entity linking result, the entity extraction result, the entity relationship extraction result, the event relationship extraction result, and the entity and event relationship extraction result.

[0010] Furthermore, according to a relationship analysis method based on news events provided by the present invention, clustering the multiple news texts to obtain multiple clustered news events includes:

[0011] K-means clustering, density-based clustering, mean shift clustering or hierarchical clustering are performed on the multiple news texts to obtain multiple news events.

[0012] Furthermore, according to a relationship analysis method based on news events provided by the present invention, entity relationship extraction is performed on the multiple news events, including one or more of the following:

[0013] Performing corpus-level entity relationship extraction on the multiple news events;

[0014] Extracting sentence-level entity relationships for the multiple news events;

[0015] The entity linking results obtained from the entity linking of the multiple news events are subjected to entity relationship expansion.

[0016] Furthermore, according to a relationship analysis method based on news events provided by the present invention, open domain relationship extraction is adopted when extracting entity relationships of the multiple news events; wherein the open domain relationship extraction refers to describing the relationship between entities based on entity pairs and connectives in the context.

[0017] Furthermore, according to a relationship analysis method based on news events provided by the present invention, event relationship extraction of the multiple news events includes one or more of the following:

[0018] Extracting causal event relationships from the multiple news events;

[0019] Extracting sub-event relations of the multiple news events;

[0020] Extracting temporal event relationships of the multiple news events.

[0021] Furthermore, according to a relationship analysis method based on news events provided by the present invention, entities and event relationships are extracted from the multiple news events, including:

[0022] Determine the relationship between entities and events;

[0023] According to the association relationship between the entities and the events, entity and event relationships are extracted for the multiple news events.

[0024] Furthermore, entity linking is performed on the multiple news events to obtain entity linking results, including:

[0025] Establish a joint representation model of words and entities;

[0026] Based on the joint representation model of the words and entities, a probabilistic entity model is adopted to establish a link relationship between the character strings in multiple news events and the entities in the knowledge base to obtain an entity link result.

[0027] In a second aspect, the present invention provides a relationship analysis device based on news events, comprising:

[0028] The first acquisition module is used to acquire multiple news texts;

[0029] A second acquisition module is used to cluster the multiple news texts to obtain multiple clustered news events;

[0030] A relationship processing module, used to perform entity linking, entity extraction, entity relationship extraction, event relationship extraction, and entity and event relationship extraction on the multiple news events, and obtain entity linking results, entity extraction results, entity relationship extraction results, event relationship extraction results, and entity and event relationship extraction results;

[0031] The relationship analysis module is used to perform association analysis on the entity linking result, the entity extraction result, the entity relationship extraction result, the event relationship extraction result, and the entity and event relationship extraction result to obtain an analysis result.

[0032] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any step of the above-mentioned relationship analysis method based on news events when executing the program.

[0033] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-mentioned relationship analysis methods based on news events.

[0034] The present invention provides a relationship analysis method based on news events. It implements news event level analysis by clustering original news texts to obtain multiple news events. It performs entity linking, entity extraction, entity relationship extraction, event relationship extraction and entity and event relationship extraction on the multiple news events to obtain entity linking results, entity extraction results, entity relationship extraction results, event relationship extraction results and entity and event relationship extraction results; it performs association analysis on the entity linking results, the entity extraction results, the entity relationship extraction results, the event relationship extraction results and the entity and event relationship extraction results to obtain analysis results. It can be seen that the present invention performs entity linking, entity extraction, entity relationship extraction, event relationship extraction and entity and event relationship extraction on news events to obtain entity linking results, entity extraction results, entity relationship extraction results, event relationship extraction results and entity and event relationship extraction results, and then performs entity extraction results, entity relationship extraction results, entity relationship extraction results, event relationship extraction results and entity and event relationship extraction results based on background knowledge in the obtained entity linking results. The event relationship extraction results and the entity-event relationship extraction results are improved, and finally a more complete relationship analysis result of the news event is obtained. It can be seen that the present invention realizes the deep mining of events, so that readers can obtain more comprehensive and in-depth information. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0036] Figure 1 This is one of the flow charts of the relationship analysis method based on news events provided by the present invention;

[0037] Figure 2 This is the second flow chart of the relationship analysis method based on news events provided by the present invention;

[0038] Figure 3 is a relational connection diagram referred to in the first embodiment of the present invention;

[0039] Figure 4 It is a structural schematic diagram of a relationship analysis device based on news events provided by the present invention;

[0040] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] Combine the following Figure 1-Figure 2 Description The present invention provides a relationship analysis method based on news events, comprising:

[0043] Step 100: Obtain multiple news texts.

[0044] Step 200: Cluster multiple news texts to obtain multiple clustered news events.

[0045] Step 300: Perform entity linking, entity extraction, entity relationship extraction, event relationship extraction, and entity and event relationship extraction on the multiple news events to obtain entity linking results, entity extraction results, entity relationship extraction results, event relationship extraction results, and entity and event relationship extraction results.

[0046] Step 400: Perform association analysis based on the entity linking results, the entity extraction results, the entity relationship extraction results, the event relationship extraction results, and the entity and event relationship extraction results to obtain analysis results.

[0047] Specifically, in step 100, the news page information of each news website is obtained by data mining technology. The news website here can be a website that specializes in providing news information, and can include large national news portals (such as Xinhuanet, People's Daily Online, etc.), business portals (Sina News, NetEase News, etc.), local news portals (Yangtze River Network, etc.), and industry portals; and can also include some social platforms (Sina Weibo, etc.). The news information of these news sites is all-encompassing and has a large number of user visits. By mining the news pages of these news sites, more comprehensive news data can be obtained.

[0048] Then, according to step 200, the acquired news information is clustered. The clustering refers to the process of dividing a collection of physical or abstract objects into multiple classes consisting of similar objects. It can refer to clustering based on the correlation between news page titles or the correlation between search terms. Clustering news texts refers to clustering news texts based on keywords or other information in the text content. News texts clustered into one class are called news events. For example, by searching for news related to "James" from August 1, 2020 to October 31, 2020, a total of 216 news items were obtained, which were clustered into 57 events. For example, one of the events was the news on September 28, 2020, "The Lakers returned to the finals after 10 years, and James took over to continue the purple and gold legend", another event was the news on September 28, 2020, "The Lakers and Heat write a new chapter in the finals", and one event was the news on October 12, 2020, "Lakers championship! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the Finals FMVP".

[0049] In step 300, after obtaining the clustered events of the news text, the obtained clustered multiple news events are extracted. The extracted objects are entities in the news events. Entities represent specific things in the real world, or abstract concepts. Such as people, organizations, places, or "machine learning", "artificial intelligence", etc. Different from the "named entities" in most studies, the entities referred to in this article include named entities (mainly referring to people, organizations, places), common entities (such as movies, books, songs, cultural customs, food, materials, etc.) and abstract concepts (concepts without physical forms generated from human abstract thinking). In the knowledge base, an entity may correspond to multiple concepts, such as Michael Jordan belongs to both the category "basketball player" and the category "Presidential Medal of Freedom Winner" in Wikipedia. Entity recognition refers to identifying entities with specific meanings in the text, mainly including names of people, places, names of organizations, proper nouns, etc. For example, taking the three events of the cluster obtained above as an example, entity extraction is performed, and the entities that can be obtained include: "Los Angeles Lakers", "Davis", "James", "Miami Heat", "Championship", "Liverpool", etc.

[0050] In addition, the event text is extracted, and the extracted content includes different types such as the relationship between entities, the relationship between entities and events, and the relationship between events. The event text is also processed by entity linking.

[0051] The above-mentioned entity relationship refers to the relationship between entities. Taking the two entities "James" and "Los Angeles Lakers" extracted above as an example, the entity relationship can be extracted to find that the relationship between "James" and "Los Angeles Lakers" is "play for", which means "James plays for the Los Angeles Lakers".

[0052] Regarding the relationship between entities and events, that is, the relationship between entities and participants as subjects of events, entities as content objects of events, or entities as places where events occur. Similarly, the entities and events in the above text are used as examples to illustrate that the relationship between the extracted entity "James" and the event "Lakers Championship! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the Finals FMVP" can be judged as "James" is the experiencer of this event, that is, "James" is the subject of the event "Lakers Championship! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the Finals FMVP".

[0053] For the relationship between events, that is, the relationship between different events, it can be a temporal succession relationship, or a supplementary relationship in content, a logical causal relationship, or there is no special relationship between the two events, etc. Taking the event "September 28, 2020 News 'Lakers Heat write a new chapter in the Finals'" and the current event "October 12, 2020 News 'Lakers championship! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the Finals FMVP'" in the above text as an example, the relationship between the two events can be extracted as the second event "October 12, 2020 News 'Lakers championship! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the Finals FMVP'" is the subsequent event of the first event "September 28, 2020 News 'Lakers Heat write a new chapter in the Finals'", that is, the two are in a temporal succession relationship.

[0054] Entity linking for multiple news events. Entity linking refers to a given document and a knowledge base. Entity linking aims to identify all entity mentions in the text and find the entity corresponding to each entity mention in the knowledge base. If the knowledge base does not include the entity referred to by the entity mention, it needs to be mapped to an empty entity. The knowledge base here, also known as the knowledge graph, is used to describe the relationship between entities. Knowledge is organized into a directed graph, which represents all nodes of the knowledge graph. Each node represents an entity. The entity has rich information descriptions, such as entity categories, information boxes, text descriptions, etc., which represent the relationship set between two nodes. The knowledge graph can also be represented as a set of triples. Entity mentions represent text fragments that mention entities in the text. It may be a word or several consecutive words. For example, the basketball player "Michael Jordan" may have different entity mentions in different documents, such as "Jordan" or "Flying Man".

[0055] Map some strings in a text to the corresponding entities in the knowledge base. Take the entity "James" in the above text as an example. In obtaining the background knowledge of the entity, "LeBron James (LeBron James), full name LeBron Raymone James, was born on December 30, 1984 in Akron, Ohio, USA. He is an American professional basketball player, a small forward, nicknamed "Little Emperor", and plays for the NBA Los Angeles Lakers. James was selected by the Cleveland Cavaliers with the first pick in the first round of the 2003 NBA draft. In 2009 and 2010, he won the NBA regular season Most Valuable Player (MVP). In 2010, James moved to the Miami Heat. In 2011, relying on his popularity in the international sports arena, James reached a cooperation agreement with Fenway Sports Group. He became the exclusive global senior image representative of Liverpool Football Club, and the remuneration was a number of Liverpool shares. In 2012, James won the third regular season MVP, first championship and Finals MVP of his NBA career, and won the gold medal in the London International Top Tournament on behalf of the US men's basketball team, tying the record set by Michael Jordan in 1992. In 2013, James won his fourth regular season MVP, second NBA championship and second Finals MVP, achieving two consecutive championships. In 2014, James returned to the Cavaliers. In 2016, James led the Cavaliers to a reversal victory over the defending champion Warriors to win the first championship in team history and his third Finals MVP. On July 10, 2018, James officially signed a four-year, $153 million contract with the Lakers. In the 2019-20 season, James was elected the NBA assists king, won his fourth championship and his fourth Finals MVP. James has an extremely high basketball IQ and sharp breakthroughs. With excellent vision and passing skills, he is considered one of the most versatile players in NBA history. In the 2019 Forbes 100 list, James ranked 17th. On December 27, 2020, James was elected the 2020 Associated Press Best Male Athlete. "

[0056] Step 400 is to perform association analysis on the entity linking results, entity extraction results, entity relationship extraction results, event relationship extraction results, and entity and event relationship extraction results obtained on the basis of step 300 to obtain analysis results. Although the corresponding relationship is obtained in step 300, the relationship is not organically combined. For example, the method in step 300 cannot make the reader understand the meaning when seeing the following information. For example, the event "Liverpool official tweets congratulations to the Lakers for winning the championship and calls James: champion shareholder!" and the event "Lakers champion! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the finals FMVP". When extracting event relationships, the extracted event relationship is that the previous event "Liverpool official tweets congratulations to the Lakers for winning the championship and calls James: champion shareholder!" is the subsequent event of the latter event "Lakers champion! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the finals FMVP". However, readers cannot know what kind of relationship there is, and why "Liverpool" congratulated "James" for winning the championship. In order to solve this problem, the present invention supplements the relevant information by adopting entity linking, so that the relationship between events can be understood smoothly.

[0057] The background knowledge obtained after entity linking the entity "James" in step 300 includes "In 2011, relying on his popularity in the international sports arena, James reached a cooperation agreement with Fenway Sports Group and became the exclusive global senior image representative of Liverpool Football Club, and the reward was a certain amount of Liverpool's equity." It can be seen that "James" and "Liverpool" are "shareholders". This also explains why "Liverpool" congratulated "James" on winning the championship. That is, the event is explained from the supplement of the background knowledge of the entity "James", thereby realizing the mining of the event at a deeper level.

[0058] Furthermore, in this embodiment, entity relationships are extracted from clustered events, and event-entity relationships and relationships between events are integrated. At the same time, corresponding entities and supplementary entity background knowledge are added to perform overall association analysis, thereby obtaining analysis results.

[0059] The invention provides a relationship analysis method based on news events, which realizes news event level analysis by clustering original news texts, obtains multiple events, and then extracts entities, entity relationships, event relationships, and event-entity relationships from event texts. At the same time, by entity linking the text, the entity background knowledge is supplemented to obtain entities with richer and more comprehensive information, and realizes in-depth mining of events, so that readers can obtain more comprehensive and in-depth information.

[0060] Furthermore, in one embodiment of the present invention, a relationship analysis method based on news events is provided according to the present invention, wherein a plurality of news texts are clustered to obtain a plurality of clustered news events, including:

[0061] K-means clustering, density-based clustering, mean shift clustering or hierarchical clustering are performed on multiple news texts to obtain multiple clustered news events.

[0062] Furthermore, in one embodiment of the present invention, a relationship analysis method based on news events is provided according to the present invention, wherein entity relationship extraction is performed on multiple news events, including one or more of the following:

[0063] Perform corpus-level entity relationship extraction for multiple news events. Perform sentence-level entity relationship extraction for multiple news events. Perform entity relationship expansion on entity linking results obtained from entity linking for multiple news events.

[0064] Specifically, corpus-level entity relationship extraction for news events refers to extracting the relationship between two corresponding entities from the context of the event text. For example, from the event "October 12, 2020 News 'Lakers Championship! The Lakers won the 17th championship in team history after a decade, and James won the Finals FMVP'" it can be concluded that the relationship between the entity "James" and the entity "Lakers" is "'James' plays for the 'Lakers'".

[0065] The sentence-level entity relationship extraction of news event text refers to the relationship between two entities in the grammatical sentences in the event text, that is, judging the relationship between two entities from a sentence. For example, the sentence "James plays for the Los Angeles Lakers" can get two entities "James" and "Los Angeles Lakers", and the relationship between these two entities is "James", "plays for", "Los Angeles Lakers".

[0066] The entity linking results obtained from entity linking for multiple news events are extended with entity relationships. That is, the relationship between entities cannot be obtained from the event text or sentence, but there is information about the relationship between two entities in the background knowledge of the entity. For example, in the background knowledge of the entity "James", "In 2011, relying on his popularity in the international sports arena, James reached a cooperation agreement with Fenway Sports Group. He became the exclusive global senior image representative of Liverpool Football Club, and the remuneration was a number of Liverpool shares." It can be known that the relationship between the entity "James" and the entity "Liverpool" is that "James" is a shareholder of "Liverpool".

[0067] By adopting these three methods of judging the relationship between entities, the relationship between different entities can be fully and comprehensively presented, allowing readers to obtain more comprehensive information.

[0068] Furthermore, in one embodiment of the present invention, a relationship analysis method based on news events is provided according to the present invention, wherein, when performing entity relationship extraction on multiple news events, open domain relationship extraction is adopted; wherein, open domain relationship extraction refers to describing the relationship between entities based on entity pairs and connectives in the context.

[0069] Specifically, open domain relation extraction does not require pre-defined relations, but uses some words in the context of entity pairs to describe the relations between entities. For example, in "Yao Ming was born in Shanghai", the extraction result of the open domain relation extraction system is (Yao Ming, born in, Shanghai), where "Yao Ming" and "Shanghai" are entity pairs with a relationship, and "born in" represents the relationship.

[0070] Furthermore, according to the present invention, a relationship analysis method based on news events is provided, wherein event relationships are extracted from multiple news events, including one or more of the following:

[0071] Extract causal event relationships from multiple news events. Extract sub-event relationships from multiple news events. Extract temporal event relationships from multiple news events.

[0072] Specifically, causal event relationships are extracted for multiple news events. For example, between the event "Liverpool official Twitter congratulates Lakers on winning the championship and calls out to James: Champion shareholder!" and the event "Lakers champion! Lakers win 17th championship in team history after ten years, James wins Finals FMVP", combined with the information in the entity link of the entity "James", it can be seen that the latter event is the cause of the former event, that is, "James" is a shareholder of "Liverpool". Therefore, after "James" won the "Finals FMVP", "Liverpool" congratulated "James". In other words, "James" won the award first, and "Liverpool" congratulated him for his award.

[0073] The so-called sub-event relationship extraction of multiple events means that a certain event contains multiple small events, such as "a certain university held an autumn sports meeting on the weekend and many school records were broken" and another event is "the men's 800-meter school record was broken". These two events themselves have no self-connection, but by performing entity linking on the entity "Autumn Sports Meeting", it can be known that "Men's 800 meters" is a sub-event of "Autumn Sports Meeting", so it can be judged that the latter event is a sub-event of the former event.

[0074] Similarly, the so-called extraction of temporal event relations between multiple events means that the order between the two events cannot be seen in the two events, but there is some information in the entity link, which can be used to supplement the explanation of the time order between the two events. For example, between the event "Lakers Heat write a new chapter in the Finals" and the event "Lakers championship! The Lakers won the 17th championship in team history after a lapse of ten years, and LeBron James won the Finals FMVP", the latter event is the post-order event of the former event, that is, the latter is after the former in time order.

[0075] Furthermore, in one embodiment of the present invention, a relationship analysis method based on news events is provided according to the present invention, wherein entity and event relationships are extracted for multiple news events, including:

[0076] Determine the relationship between entities and events.

[0077] According to the association relationship between entities and events, entity and event relationships are extracted for the multiple news events.

[0078] Specifically, in this embodiment, since the relationship between the entity and the event needs to be extracted, it is necessary to determine whether there is an association relationship between the entity and the event. The association relationship here is a relationship in a broad sense, which means that there is a strong or weak relationship between the entity and the event, which can also be a superficial connection or an implicit connection. Among them, a strong relationship is a relationship that is manifested in situations such as when the entity is an experiencer of the event, while a weak relationship means that the entity does not directly participate in or experience the event, but is not unrelated to the event. It can be assumed that in a basketball game, the experiencers or participants are players, referees, and coaches, etc., but as the owner or shareholder of the team, although they have no direct relationship with the game, they have a weak connection with the game.

[0079] For the surface connection or implicit connection between entities and events, the surface connection refers to the relationship between the two that can be directly derived through the event, such as the event "Lakers return to the finals after 10 years, James takes over and continues to write the purple and gold legend" can directly deduce "James" and "play for" "Lakers". However, there are also implicit relationships between entities and events, which need to be obtained from background information, such as the relationship between the entity "James" and the event "Liverpool official Twitter congratulates the Lakers on their championship and calls James: champion shareholder!" In the entity link of the entity "James", the following information exists: "In 2011, relying on his popularity in the international sports arena, James reached a cooperation agreement with Fenway Sports Group, and he became the exclusive global senior image representative of Liverpool Football Club, and the reward was a number of Liverpool shares." explains the relationship between the entity "James" and the event "Liverpool official Twitter congratulates the Lakers on their championship and calls James: champion shareholder!"

[0080] Therefore, to determine the relationship between an entity and an event, it is necessary to first determine the association relationship between the entity and the event, and then determine the specific relationship between the entity and the event based on the association relationship.

[0081] In the present invention, the news event data mentioned can be data of any range, such as news within a certain period of time, news in a certain field such as "sports", and news related to "James". There are many clustering methods, such as K-means clustering, density-based clustering method, mean shift clustering, hierarchical clustering and other methods. Then, the present invention performs entity, entity relationship extraction, and event relationship extraction on the events obtained by clustering. The relationship between entities extracted by the present invention includes the following three categories: the first category is corpus (chapter) level relationship extraction; the second category is sentence level relationship, that is, to determine what kind of semantic relationship is between two entities from a sentence; the third category is the relationship given from the linked background knowledge. In the present invention, entity relationship extraction is mainly open domain relationship extraction. Open relationship extraction does not require pre-defined relationships, but uses some words in the context of entity pairs to describe the relationship between entities. For example, in "Yao Ming was born in Shanghai", the extraction result of the open domain relationship extraction system is (Yao Ming, born in, Shanghai), where "Yao Ming" and "Shanghai" are entity pairs with relationships, and "born in" represents the relationship.

[0082] In this embodiment, the entity and event relationship extraction is mainly the relationship between entities and events. The event relationship extraction in this invention mainly focuses on the causal, sub-event, time sequence and other relationships between events. Then, the entities are linked to the corresponding background knowledge. Finally, the entity, entity background knowledge, entity relationship and event relationship are analyzed as a whole to obtain all the corresponding analysis results.

[0083] Here is a specific example: search for news related to "James" from August 1, 2020 to October 31, 2020, there are 216 news articles in total, clustered into 57 events, such as the news on September 28, 2020 "The Lakers return to the finals after 10 years, James takes over and continues the purple and gold legend", the news on September 28, 2020 "The Lakers and Heat write a new chapter in the finals", and the news on October 12, 2020 "Lakers championship! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the Finals FMVP". The entities can be extracted to obtain entities such as "Los Angeles Lakers", "Davis", "James", "Miami Heat", "Championship", "Liverpool" and so on; for the entity link part, the background knowledge of the entity is introduced, for example, the linked "James" can obtain the background knowledge "LeBron James (LeBron James), full name LeBron Raymone James, was born on December 30, 1984 in Akron, Ohio, USA. He is an American professional basketball player, a small forward, nicknamed "Little Emperor", and plays for the NBA Los Angeles Lakers. James was selected by the Cleveland Cavaliers with the first pick in the first round of the 2003 NBA draft. In 2009 and 2010, he won the NBA regular season Most Valuable Player (MVP). In 2010, James moved to the Miami Heat. In 2012, James won the third regular season MVP, the first championship and the finals MVP in his NBA career, and won the gold medal in the London International Top Event on behalf of the US men's basketball team, tying Michael Jordan in 1992. In 2013, James won his fourth regular season MVP, second NBA championship and second Finals MVP, achieving two consecutive championships. In 2014, James returned to the Cavaliers. In 2016, James led the Cavaliers to a comeback victory over the defending champion Warriors to win the first championship in team history and his third Finals MVP. On July 10, 2018, James officially signed a 4-year, $153 million contract with the Lakers. In the 2019-20 season, James was elected NBA assists king, won his fourth championship and his fourth Finals MVP. James has a very high basketball IQ, sharp breakthroughs, excellent vision and passing skills, and is considered one of the most versatile players in NBA history. In the 2019 Forbes 100 Celebrity List, James ranked 17th. On December 27, 2020, James was elected the 2020 Associated Press Best Male Athlete."The entity relationship can be extracted to find that the relationship between "James" and "Los Angeles Lakers" is "play for", which means "James plays for the Los Angeles Lakers"; the relationship between the entity "James" and the event "Lakers Championship! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the Finals FMVP" can be extracted as "James" is the experiencer of this event; for the news on September 28, 2020 "Lakers Heat write a new chapter in the Finals" and the news on October 12, 2020 "Lakers Championship! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the Finals FMVP", the event relationship extraction can obtain the event "Lakers Championship! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the Finals FMVP" is a subsequent event of the event "Lakers Heat write a new chapter in the Finals". The event "Liverpool official tweeted congratulations to the Lakers on winning the championship and called out to James: Championship shareholder! "is the event "Lakers Championship! The Lakers won the 17th championship in team history after a lapse of ten years, and James won the Finals FMVP. At first glance, I don’t know what the relationship is between them. “Why did Liverpool congratulate James on winning the championship?” In the results of the entity link of “James”, I can see that “In 2011, relying on his popularity in the international sports arena, James reached a cooperation agreement with Fenway Sports Group. He became the global exclusive senior image representative of Liverpool Football Club, and the remuneration was a number of Liverpool’s shares. ", from which it can be obtained that "James" and "Liverpool" are "shareholders", which explains why "Liverpool" "congratulates" "James" on winning the "championship". Here, the event is explained from the supplement of the background knowledge of the entity "James", thereby realizing the mining of the event at a deeper level. It can be seen that this embodiment realizes news event level analysis through clustering of original news texts. This embodiment realizes entity recognition, entity relationship extraction, event relationship extraction, and event-entity relationship extraction through event text extraction. This embodiment supplements the entity background knowledge through entity linking of entities to realize in-depth mining of events. The present invention realizes event tracking, entity portrait modeling, and mining of implicit entity relationships in events through the analysis of news event language at a higher level.

[0084] Reference Figure 3 , Figure 3 At the bottom is the timeline. Each square represents a clustered event, and the dots above represent each entity. The lines represent the relationship between entities, the relationship between entities and events, etc. The final result can be analyzed from these diagrams.

[0085] Further, in one embodiment of the present invention, a relationship analysis method based on news events is provided according to the present invention, and entity linking is performed on multiple news events to obtain entity linking results, including:

[0086] Establish a joint representation model of words and entities;

[0087] Based on the joint representation model of the words and entities, a probabilistic entity model is adopted to establish a link relationship between the entities in the news event and the entities in the knowledge base, and obtain an entity link result.

[0088] In this embodiment, the joint representation model of words and entities includes a combination of a skip-gram model, a knowledge base model, and an anchor text context model; accordingly, a joint representation model of words and entities is established, including:

[0089] Build a skip-gram model to learn the representation of words in a given text corpus by predicting the neighboring words of the word;

[0090] Build a knowledge base model to learn entity representation by predicting entities adjacent to the target entity;

[0091] Establish an anchor text context model, replace the anchor text with the corresponding entity, predict the words around the entity, and map the words and entities into the same semantic space;

[0092] When training the joint representation model of words and entities, the overall goal of the joint representation model of words and entities is to maximize the linear combination of three model objective functions.

[0093] In this embodiment, based on the joint representation model of words and entities, a probabilistic entity model is used to establish a link relationship between entities in news events and entities in the knowledge base, including:

[0094] Given an entity m i , according to entity m i Find relevant entities e from the knowledge base;

[0095] Based on entity e, generate entity m i context;

[0096] Generate entity m i Other entities in

[0097] Where, given an input document D and an entity m i , find out m in the knowledge base i The corresponding entity problem is determined to be of the following form:

[0098]

[0099] Among them, the generation process corresponding to each entity is independent of each other. Represents the context of the entity, and N represents the set of other unambiguous entities in the input document;

[0100] Given an input document D and an entity mi, the corresponding entity in the final knowledge base is the entity that maximizes the posterior probability P(ei|mi,D), so it is formalized as follows:

[0101]

[0102] P(e) is the prior distribution of the entity, which is defined as the prior distribution over the entire dataset. In order to control the impact of different priors in different fields, the impact factor α is introduced:

[0103]

[0104] Among them, A e,* is the set of anchor texts pointing to entity e, A *,* is the set of all anchor texts in the knowledge base KB; α=0 means that the entity prior is 1, and the value has no effect on the posterior probability P(e|m); α=1 means that the prior probability is not subject to any control;

[0105] is the context text distribution corresponding to e. An entity is more likely to appear in a context that matches its meaning. The context vector is represented by the average of the word vectors in the context of the entity:

[0106]

[0107] in, is the set of words in the context;

[0108] Create an Aho-Corasick index in advance for the vocabulary corresponding to the word vector, and directly use the Aho-Corasick algorithm to match the words in the entity context without the need for word segmentation.

[0109] is the cosine similarity between the candidate entity e and the context vector. Since the value range of cosine similarity is [-1, 1], it is mapped to the range of [0, 1];

[0110] P(N|e) is the probability distribution of context entities given entity e;

[0111] The method for calculating entity consistency includes: finding the initial unambiguous entity in the currently processed document, adding the initialized unambiguous entity set N, and determining the entity with a prior probability P^(e|m)>θ as an unambiguous entity, where P^(e|m)=|A e,m | / |A *,m |;

[0112] Process the discovered entities {m1,m2,,m |M|}, each time a disambiguated entity is obtained, it is added to set N, and the vector of set N is represented as the average of all entity vectors in it:

[0113]

[0114] Among them, E N Represents an unambiguous entity.

[0115] Specifically, entity linking, given a document and a knowledge base, entity linking aims to identify all entity mentions in the text and find the entity corresponding to each entity mention in the knowledge base. If the knowledge base does not include the entity referred to by the entity mention, it needs to be mapped to an empty entity. The entity linking task is generally divided into three steps: entity discovery, candidate entity generation, and candidate entity disambiguation. Entity discovery aims to identify all entity mentions in the document, and candidate entity generation finds the knowledge base entity that each entity mention may refer to, which is called a candidate entity set. Candidate entity disambiguation determines the knowledge base entity referred to by the entity mention.

[0116] The present invention uses a joint representation model of words and entities. The model is mainly based on the skip-gram model. Skip-gram was first proposed to learn the embedded representation of words, and its central idea is to use the target word to predict the words in its context. The joint representation model of words and entities used in the present invention has three parts based on the skip-gram model: 1) a conventional skip-gram model, which learns the representation of words in a given text corpus by predicting the adjacent words of the word; 2) a knowledge base model, which learns the representation of the entity by predicting the adjacent entities of the target entity; 3) an anchor text context model, which replaces the anchor text with the corresponding entity, predicts the words in the context around the entity, and maps the words and entities to the same semantic space. When training the joint representation model, the overall goal of the model is to maximize the linear combination of the three model objective functions.

[0117] At the same time, the present invention can regard the entity linking process as a generative process. Given an entity mention mi, first, find a related entity e from the knowledge base KB based on the entity, then, based on the entity e, generate the context of the entity mention mi, and finally, generate other entities in the entity mention mi. Therefore, given an input document D and an entity mention mi, the problem of finding the corresponding entity of mi in the knowledge base can be inferred as in Formula 1:

[0118]

[0119] The present invention assumes that the generation process corresponding to each entity mention is independent of each other, where represents the context of the entity mention, and N represents the set of other unambiguous entities in the input document.

[0120] Given an input document D and an entity mention mi, the corresponding entity in the final knowledge base is the entity that maximizes the posterior probability P(ei|mi,D), so it can be formalized as 2:

[0121]

[0122] P(e) is the prior distribution of the entity. The present invention defines the prior distribution of the entity as the prior distribution on the entire data set, that is, the present invention is the prior distribution in the field of electric power customer service. In a large-scale corpus, the more times an entity is mentioned, the more well-known this entity may be. However, in different fields, the prior probability of an entity may be different. For example, in the field of electric power customer service, the probability that the term "system internal overvoltage" refers to "power system content overvoltage" in the field of electric power customer service is greater than in the open field. Therefore, in order to control the impact of different priors in different fields, the present invention introduces an impact factor α, as shown in Formula 3.

[0123]

[0124] Among them, Ae,* is the set of anchor texts pointing to entity e, and A*,* is the set of all anchor texts in KB. α=0 means that the entity prior is 1, that is, its value has no effect on the posterior probability P(e|m), and α=1 means that the prior probability is not controlled in any way.

[0125] is the context text distribution corresponding to e. An entity is more likely to appear in a context environment that matches its meaning. For example, when there are words such as "monthly fee" and "total fee" in the context, the entity mention "electricity fee" is more likely to refer to the entity "monthly electricity fee" rather than "electricity fee per kilowatt-hour". Therefore, the present invention uses the average of the word vectors in the context of the entity mention to represent the context vector, that is, Formula 4:

[0126]

[0127] in, It is a collection of words in the context. The present invention establishes an Aho-Corasick index in advance for the vocabulary corresponding to the word vector, so the Aho-Corasick algorithm can be directly used to match the words in the entity mention context without the need for operations such as word segmentation. The present invention regards it as the cosine similarity between the candidate entity e and the context vector. Since the value range of cosine similarity is [-1, 1], the present invention maps it to the range of [0, 1].

[0128] P(N|e) is the probability distribution of contextual entities for a given entity e. In a related context, such as a news article, entities usually belong to the same topic, and these entities are usually close in the semantic space. Therefore, this distribution can also be regarded as the distribution of topic consistency of entities. The present invention designs a two-step method for calculating entity consistency. First, the present invention finds the initial unambiguous entity in the currently processed document and adds it to the initialized unambiguous entity set N. The present invention defines the entity with a prior probability P^(e|m)>θ as an unambiguous entity, where P^(e|m)=|Ae,m| / |A*,m|. In the system, the present invention selects θ=0.95. Then, the present invention processes the mentions of the discovered entities {m1,m2,, m|M|} in order from left to right (L2R) or from simple to complex (S2C). After each processing to obtain a disambiguated entity, it is added to the set N. The vector of the set N is expressed as the average of all entity vectors therein, that is, Formula 5:

[0129]

[0130] Wherein, EN represents an unambiguous entity. P(N|e) is calculated by the cosine similarity between the vector of set N and the vector of the candidate entity. The present invention finds that the difference in the results brought by the processing order of S2C and L2R is not obvious. Considering that the order of S2C needs to be sorted again, the present invention adopts the order of L2R ​​in the specific implementation.

[0131] In this embodiment, it should be noted that entity linking, given a document and a knowledge base, aims to identify all entity mentions in the text and find the entity corresponding to each entity mention in the knowledge base. If the knowledge base does not include the entity referred to by the entity mention, it needs to be mapped to an empty entity.

[0132] In this embodiment, the entity linking part, for the entity discovery result {m1, m2,...,m |M|} and its corresponding candidate entity set C1,C2,...,C |M| , the entity link is for each m i In its corresponding candidate entity set C i Find a corresponding entity e in i *This part mainly includes two tasks: 1) learning the joint representation of words and entities; 2) using a probabilistic disambiguation model for entity linking based on the joint representation of words and entities.

[0133] The joint representation of words and entities is explained and illustrated below. This embodiment uses a joint representation model of words and entities. The model is mainly based on the skip-gram model. Skip-gram was first proposed to learn the embedded representation of words, and its central idea is to use the target word to predict the words in its context. The joint representation model of words and entities used in the present invention has three parts based on the skip-gram model: 1) a conventional skip-gram model, which learns the representation of words in a given text corpus by predicting the adjacent words of the word; 2) a knowledge base model, which learns the representation of entities by predicting the adjacent entities of the target entity; 3) an anchor text context model, which replaces the anchor text with the corresponding entity, predicts the words in the context around the entity, and maps the words and entities into the same semantic space.

[0134] (1) Word Representation Learning

[0135] Given a word sequence w1,w2,...,w containing T words T , the goal of the skip-gram model is to maximize the objective function in the following formula:

[0136]

[0137] Where c is the size of the context window, w t represents the target word, w t+j Represents the context word. Conditional probability P(w t+j |w t ) is calculated according to softmax, as shown in the following formula:

[0138]

[0139] Among them, W is the set of all words, V w and U w Represents the vector of word w in matrix V and matrix U.

[0140] (2) Knowledge base model

[0141] In the electric customer service knowledge base, each entity has links to other entities, which the present invention calls "external links". The present invention uses the external link relationship between entities to learn the relevance between entities. Another method to measure the relevance between entities is the Wikipedia Link-based Measure (WLM), which is applied as a feature of entity links. WLM is calculated according to the following formula:

[0142]

[0143] Among them, E is the entity in the knowledge base KB, C e is a set of links pointing to entity e. The assumption of the WLM method is that entities with more intersections of link sets are more related. Inspired by the WLM method, based on the skip-gram model, the connection relationship between entities can be regarded as the contextual relationship between words. Therefore, the knowledge base model is formalized as follows:

[0144]

[0145] Similarly, the conditional probability P(e o |e i ) can also be calculated using softmax.

[0146] (3) Anchor Text Context Model

[0147] If we simply combine the skip-gram model of a word with the knowledge base model, the word and the entity are not in the same vector space. Therefore, we can use the anchor text to replace the anchor text with the entity it represents. Based on the skip-gram model, we can use the entity to predict the words in its context. The objective function of the model is as follows:

[0148]

[0149] Among them, A is the anchor text set, and Q is the set of context words around the anchor text. During training, the overall goal of the model is to maximize the linear combination of the three model objective functions, as shown in the following formula:

[0150] L=L w +L e +L a

[0151] The following is a detailed introduction to the probabilistic entity model. The entity linking process can be regarded as a generative process. Given an entity mention m i First, find a related entity e from the knowledge base KB according to the entity, and then generate entity mention m according to entity e i Finally, we generate entity mention m i Therefore, given an input document D and entity mention m i , find out m in the knowledge base i The corresponding entity problem can be inferred to be in the form of:

[0152]

[0153] This example assumes that the generation process of each entity mention is independent of each other. represents the context of the entity mention, and N represents the set of other unambiguous entities in the input document.

[0154] In this example, given an input document D and entity mention m i , the corresponding entity in the final knowledge base is the one that maximizes the posterior probability P(e i |m i ,D) entity, therefore, can be formalized as follows:

[0155]

[0156] P(e) is the prior distribution of the entity. This embodiment defines the prior distribution of the entity as the prior distribution over the entire data set, for example, it can be the prior distribution in the field of electric power customer service. In a large-scale corpus, the more times an entity is mentioned, the more well-known this entity may be. However, in different fields, the prior probability of an entity may be different. For example, in the field of electric power customer service, the probability that the term "system internal overvoltage" refers to "power system content overvoltage" in the field of electric power customer service is greater than in the open field. Therefore, in order to control the impact of different priors in different fields, this embodiment introduces an impact factor α, as shown in the following formula:

[0157]

[0158] Among them, A e,* is the set of anchor texts pointing to entity e, A *,* is the set of all anchor texts in the KB. α=0 means that the entity prior is 1, that is, its value has no effect on the posterior probability P(e|m), and α=1 means that the prior probability is not controlled in any way.

[0159] is the context text distribution corresponding to e. An entity is more likely to appear in a context that matches its meaning. For example, when there are words such as "monthly fee" and "total fee" in the context, the entity mention "electricity fee" is more likely to refer to the entity "monthly electricity fee" rather than "electricity fee per kilowatt-hour". Therefore, this embodiment uses the average of the word vectors in the context of the entity mention to represent the context vector, that is, the following formula:

[0160]

[0161] in, It is a collection of words in the context. The present invention establishes an Aho-Corasick index in advance for the vocabulary corresponding to the word vector, so the Aho-Corasick algorithm can be directly used to match the words in the entity mention context without the need for operations such as word segmentation. The present invention regards it as the cosine similarity between the candidate entity e and the context vector. Since the value range of cosine similarity is [-1, 1], this embodiment maps it to the range of [0, 1].

[0162] P(N|e) is the probability distribution of context entities for a given entity e. In a related context, such as a news article, entities usually belong to the same topic, and these entities are usually close in the semantic space. Therefore, this distribution can also be regarded as the distribution of topic consistency of entities. This embodiment designs a two-step method for calculating entity consistency. First, the present invention finds the initial unambiguous entity in the currently processed document and adds it to the initialized unambiguous entity set N. This embodiment defines entities with prior probability P^(e|m)>θ as unambiguous entities, where P^(e|m)=|A e,m | / |A *,m |, in the system, this embodiment selects θ = 0.95; Then, this embodiment processes the entity mentions {m1, m2,, m |M|}, each time a disambiguated entity is obtained, it is added to set N. The vector of set N is represented as the average of all entity vectors in it, that is, the following formula:

[0163]

[0164] Among them, E N Represents an unambiguous entity. P(N|e) is calculated by the cosine similarity of the vector of the set N and the vector of the candidate entity. In this embodiment, it is found that the difference in the results brought by the processing order of S2C and L2R is not obvious. Considering that the order of S2C needs to be sorted again, the present invention adopts the order of L2R ​​in the specific implementation.

[0165] In addition, the following part is explained and illustrated about the result pruning. After the process of entity discovery and entity linking, for each document D, the present invention can obtain a result sequence R = {r1, r2, ..., r|M|}, ri = (m i ,e i *,score i ), where score iis the value of P(e|m) in the entity linking result, which is regarded as the confidence of entity linking in the present invention. Since the number of dictionaries built based on encyclopedias is huge, it is inevitable that some noise entities will be left if all entities are retained in entity discovery. Therefore, the present invention designs a simple filtering algorithm to filter the final results, and filters R according to r i .score i Sort from high to low, retaining the first k% of the results Will link_prob(r i .m i )<∈. The two parameters k and ∈ are determined by experience. i .score i Sort from high to low, retaining the first k% of the results "This is not considered a priority.

[0166] In this embodiment, examples of entity recognition and entity linking data can be referred to Table 1 and Table 2 below.

[0167] Table 1 Entity recognition

[0168]

[0169] Table 2 Entity linking

[0170]

[0171] The following are examples of Chinese and English. The knowledge base uses Wikipedia and Baidu Encyclopedia. The number of anchor texts is huge. According to statistics, there are nearly 10 million anchor texts in English Wikipedia pages and more than 3 million anchor texts in Baidu Encyclopedia. Rich anchor texts provide sufficient data support for entity discovery and entity linking. By extracting anchor texts from Baidu Encyclopedia and English Wikipedia, an anchor text dictionary is constructed. Each key in the dictionary can be regarded as a mention of an entity, and the corresponding value can be regarded as the entity that should be mentioned in the knowledge base.

[0172] Table 3: Display of some anchor text dictionaries

[0173]

[0174] In addition, since the English Wikipedia also includes disambiguation pages, that is, for an entity name, all possible entities it may refer to are given, and this part of the data is also added to the English anchor text dictionary. Finally, the total amount of the obtained anchor text dictionary is as follows: The English Wikipedia's anchor text dictionary has 4,843,616 entity mention-entity matching pairs, and Baidu Encyclopedia has 2,895,610 entity mention-entity matching pairs.

[0175] In all the anchor texts of the present invention, the following filtering rules are applied for filtering: where length is taken as 1, prob is taken as 0.01, and count is taken as 2. (The value range can be enlarged)

[0176] 1) Remove the anchor texts with length(m) ≤ 1;

[0177] 2) Remove the anchor texts with link_prob(m) ≤ 0.01;

[0178] 3) Remove the anchor texts with count(m) ≤ 2.

[0179] The following is an example for illustration. For example, for the example where m1 is "Nanjing Yangtze River Bridge" and m2 is "Yangtze River Bridge", at this time length(m1) and length(m2) are 7 and 4 respectively, then there is length(m1) > length(m2), so m1 is retained. For the example where m1 is "time-of-use electricity price for electric heating" and m2 is "step electricity price for electric heating", at this time length(m1) and length(m2) are both 7, that is, length(m1) = length(m2), but link_prob(m1) < link_prob(m2), in this case m2 should be retained.

[0180] For the result pruning part, the value of ∈ is taken as 0.0008. (The range can be enlarged) "Sort R according to r i .score i in descending order, and retain the top k% of the results as The rules are not filtered.

[0181] According to the above technical solution, in this embodiment, an unsupervised generative probability model based on the joint representation of words and entities is designed to solve the ambiguity of entities.

[0182] Combined with Figure 4 , the following describes a relationship analysis device based on news events provided by the present invention. The relationship analysis device based on news events described below can be mutually corresponding and referred to the relationship analysis method based on news events described above.

[0183] In one embodiment of the present invention, the present invention provides a relationship analysis device based on news events, comprising:

[0184] The first acquisition module 41 is used to acquire multiple news texts.

[0185] The second acquisition module 42 is used to cluster the multiple news texts to obtain multiple clustered news events.

[0186] The relationship processing module 43 is used to perform entity linking, entity extraction, entity relationship extraction, event relationship extraction, and entity and event relationship extraction on the multiple news events to obtain entity linking results, entity extraction results, entity relationship extraction results, event relationship extraction results, and entity and event relationship extraction results.

[0187] Relationship analysis module 44: used to perform association analysis on the entity linking results, the entity extraction results, the entity relationship extraction results, the event relationship extraction results, and the entity and event relationship extraction results to obtain analysis results.

[0188] Since the device provided in the embodiment of the present invention can be used to execute the method described in the above embodiment, its working principle and beneficial effects are similar, so they are not described in detail here. For specific contents, please refer to the introduction of the above embodiment.

[0189] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute a relationship analysis method based on news events, the method comprising: obtaining multiple news texts; clustering the multiple news texts to obtain multiple news events after clustering; performing entity linking, entity extraction, entity relationship extraction, event relationship extraction and entity and event relationship extraction on the multiple news events to obtain entity linking results, entity extraction results, entity relationship extraction results, event relationship extraction results and entity and event relationship extraction results; performing association analysis based on the entity linking results, the entity extraction results, the entity relationship extraction results, the event relationship extraction results and the entity and event relationship extraction results to obtain analysis results.

[0190] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0191] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the above-mentioned methods for performing a relationship analysis method based on news events.

[0192] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor. Through the description of the above implementation methods, technicians in this field can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware. Based on such an understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiment.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A relationship analysis method based on news events, characterized in that: include: Get multiple news texts; Clustering the multiple news texts to obtain multiple clustered news events; Performing entity linking, entity extraction, entity relationship extraction, event relationship extraction, and entity-event relationship extraction on the multiple news events to obtain entity linking results, entity extraction results, entity relationship extraction results, event relationship extraction results, and entity-event relationship extraction results; Performing association analysis on the entity linking result, the entity extraction result, the entity relationship extraction result, the event relationship extraction result, and the entity and event relationship extraction result to obtain an analysis result; The entity linking is performed on the multiple news events to obtain entity linking results, including: Establish a joint representation model of words and entities; Based on the joint representation model of the word and entity, a probabilistic entity model is used to establish a link relationship between entities in multiple news events and entities in the knowledge base to obtain an entity link result; Among them, the joint representation model of words and entities is established, including: Build a skip-gram model to learn the representation of words in a given text corpus by predicting the neighboring words of the word; Build a knowledge base model to learn entity representation by predicting entities adjacent to the target entity; Establish an anchor text context model, replace the anchor text with the corresponding entity, predict the words around the entity, and map the words and entities into the same semantic space; When training the joint representation model of the word and the entity, the overall goal of the joint representation model of the word and the entity is to maximize the linear combination of three model objective functions; Among them, based on the joint representation model of the word and entity, a probabilistic entity model is adopted to establish the link relationship between the entities in multiple news events and the entities in the knowledge base, and obtain the entity link result, which specifically includes: Extract unambiguous entities whose prior probabilities meet preset conditions in the currently processed document to obtain an initialized unambiguous entity set; For the current entity m in the unambiguous entity set i Match the candidate entities in the knowledge base to obtain the related entity e that maximizes the posterior probability; the posterior probability is calculated based on the cosine similarity between the candidate entity and the context vector, the probability distribution of the context entity of the candidate entity, the prior distribution of the candidate entity and the preset influence factor; the context vector is the current entity m i The average of the word vectors in the context; According to the entity e, generate the current entity m i context; Generate current entity m i other entities in .

2. The method for analyzing the relationship between news events according to claim 1, characterized in that: Clustering the multiple news texts to obtain multiple clustered news events includes: K-means clustering, density-based clustering, mean shift clustering or hierarchical clustering are performed on the multiple news texts to obtain multiple news events.

3. The relationship analysis method of news events according to claim 1 is characterized in that: Entity relationships are extracted from the multiple news events, including one or more of the following: Performing corpus-level entity relationship extraction on the multiple news events; Extracting sentence-level entity relationships for the multiple news events; The entity linking results obtained from the entity linking of the multiple news events are subjected to entity relationship expansion.

4. The method for analyzing the relationship between news events according to any one of claims 1 to 3, characterized in that: When performing entity relationship extraction on the multiple news events, open domain relationship extraction is adopted; wherein, the open domain relationship extraction refers to describing the relationship between entities based on entity pairs and connectives in the context.

5. The method for analyzing the relationship between news events according to claim 1, characterized in that: Extracting event relations from the multiple news events includes one or more of the following: Extracting causal event relationships from the multiple news events; Extracting sub-event relations of the multiple news events; Extracting temporal event relationships of the multiple news events.

6. The method for analyzing the relationship between news events according to claim 1, characterized in that: Extracting entities and event relationships from the multiple news events includes: Determine the relationship between entities and events; According to the association relationship between the entities and the events, entity and event relationships are extracted for the multiple news events.

7. A relationship analysis device based on news events, characterized in that: include: The first acquisition module is used to acquire multiple news texts; A second acquisition module is used to cluster the multiple news texts to obtain multiple clustered news events; A relationship processing module, used to perform entity linking, entity extraction, entity relationship extraction, event relationship extraction, and entity-event relationship extraction on the multiple news events, and obtain entity linking results, entity extraction results, entity relationship extraction results, event relationship extraction results, and entity-event relationship extraction results; A relationship analysis module, used to perform association analysis on the entity linking result, the entity extraction result, the entity relationship extraction result, the event relationship extraction result, and the entity and event relationship extraction result to obtain an analysis result; The entity linking is performed on the multiple news events to obtain entity linking results, including: Establish a joint representation model of words and entities; Based on the joint representation model of the word and entity, a probabilistic entity model is used to establish a link relationship between entities in multiple news events and entities in the knowledge base to obtain an entity link result; Among them, the joint representation model of words and entities is established, including: Build a skip-gram model to learn the representation of words in a given text corpus by predicting the neighboring words of the word; Build a knowledge base model to learn entity representation by predicting entities adjacent to the target entity; Establish an anchor text context model, replace the anchor text with the corresponding entity, predict the words around the entity, and map the words and entities into the same semantic space; When training the joint representation model of the word and the entity, the overall goal of the joint representation model of the word and the entity is to maximize the linear combination of three model objective functions; Among them, based on the joint representation model of the word and entity, a probabilistic entity model is adopted to establish the link relationship between the entities in multiple news events and the entities in the knowledge base, and obtain the entity link result, which specifically includes: Extract unambiguous entities whose prior probabilities meet preset conditions in the currently processed document to obtain an initialized unambiguous entity set; For the current entity m in the unambiguous entity set i Match the candidate entities in the knowledge base to obtain the related entity e that maximizes the posterior probability; the posterior probability is calculated based on the cosine similarity between the candidate entity and the context vector, the probability distribution of the context entity of the candidate entity, the prior distribution of the candidate entity and the preset influence factor; the context vector is the current entity m i The average of the word vectors in the context; According to the entity e, generate the current entity m i context; Generate current entity m i other entities in .

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.