News thread-based search display method and device, server, and storage medium

Abstract

The application discloses a news thread-based search display method and device, a server and a storage medium, relates to the technical field of information processing, and comprises the following steps: determining hot news based on the attention degrees of each news information in a set time by a user, and collecting the hot news into a news candidate library; performing topic induction on the hot news and generating event nodes, performing news search and vector conversion based on the event nodes, news topics and news content, and establishing a vector index in a vector index library; the topic induction is used for classifying the hot news, and the event nodes are used for describing the content information of the hot news under each topic; in response to receiving a search request, performing vector conversion on the request content, performing reverse index on the converted vectors stored in the vector index library through the event nodes, and generating a news thread list according to the index result. The scheme can greatly reduce the time cost of user search, help the user quickly understand news events, and improve the news click rate of the user in the in-site search scene.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of information processing technology, and in particular to a search and display method, device, server, and storage medium based on news context. Background Art

[0002] With the rapid development of the internet, people consume a variety of news information daily, including videos, images, and text. However, as more and more news is published online, the amount of redundant information becomes overwhelming. This is because users' understanding of news generally falls into two stages: from complete unfamiliarity to a certain level of understanding. This process requires a timeline and tracking of progress to gradually grasp the news content.

[0003] When users are completely unfamiliar with news, relevant app search results are based on keyword matching, lacking detailed descriptions of the events themselves. This leads to an excessive amount of low-quality content, resulting in high sifting costs and difficulty understanding the underlying story. When users have a more established understanding of news, app search results often display mainstream perspectives, which can be one-sided and make it difficult to track subsequent related news. For example, for trending online topics, users may need significant time to research and analyze the event before gaining a general understanding. Otherwise, they only receive the final outcome, resulting in fragmented information and a lack of rapid understanding of the developments and context of the news story. Summary of the Invention

[0004] The present application provides a search and display method, device, server and storage medium based on news context, which reduces the time cost for users to understand news events and helps users quickly understand news events.

[0005] In one aspect, the present application provides a search and display method based on news context, the method comprising:

[0006] S1, based on the user's attention to each piece of news information within a set time period, determine hot news and include it in the news candidate library;

[0007] S2: Topically summarize the hot news and generate event nodes. Based on the event nodes, news topics, and news content, perform news search and vector conversion, and establish a vector inverted index in the vector index library. Topic summary is used to classify the hot news, and the event nodes are used to describe the content information of the hot news under each topic.

[0008] S3, in response to receiving the search request, performing vector conversion on the request content, performing vector index matching with the inverted index established in the vector index library through the converted vector, and generating a news context list according to the event node and the index result.

[0009] On the other hand, the present application provides a search and display device based on news context, the device comprising:

[0010] Hot news determination module, used to determine hot news based on the user's attention to various news information within a set time, and include it in the news candidate library;

[0011] A node conversion module is used to summarize the hot news topics and generate event nodes, perform news search and vector conversion based on the event nodes, news topics, and news content, and establish a vector inverted index in the vector index library; topic summarization is used to classify the hot news, and the event nodes are used to describe the content information of the hot news under each topic;

[0012] The news context generation module is used to convert the request content into a vector in response to receiving a search request, match the converted vector with the inverted index established in the vector index library, and generate a news context list based on the event node and the index result.

[0013] On the other hand, the present application provides a server, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the search and display method of news context described in any of the above aspects.

[0014] On the other hand, the present application provides a computer-readable storage medium, which stores at least one instruction, at least one program, code set or instruction set. The at least one instruction, the at least one program, the code set or instruction set is loaded and executed by a processor to implement the search and display method of news context described in any of the above aspects.

[0015] The beneficial effects of the technical solution provided by the embodiment of the present application include at least: determining hot news by crawling and analyzing the published news according to the set time, and storing it in the news candidate library to facilitate the subsequent capture of related news; for the hot news included in the candidate news library, the topic is summarized according to the news content, and an event node describing the hot news content information under the topic is generated. In order to improve the accuracy of retrieval display, the hot news and the corresponding event nodes and topics are vectorized by using a vector indexing method, and an inverted index is established in the vector index library to facilitate subsequent user searches. In the user search process, instead of using a keyword-only search method, the text content is vectorized, and the news is hit and returned based on the spatial mapping relationship with the vector data in the vector index library. This calculation method based on vector space mapping can improve the accuracy of retrieval, so that the returned news context is more accurate. Compared with the traditional list-based display method, it can greatly reduce the time cost of user searches, help users quickly understand news events, and improve the user's news click-through rate in the site search scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is an example diagram of news search based on keyword matching in related technologies;

[0017] Figure 2 This is a schematic diagram of a news context display method used in related art;

[0018] Figure 3 This is a flowchart of a news context-based search and display method provided in an embodiment of the present application;

[0019] Figure 4 This is a framework diagram for searching and displaying news context based on an inverted index provided by an embodiment of the present application;

[0020] Figure 5 Comparison chart of hits based on keyword search and vector search;

[0021] Figure 6 This is a flowchart of a news context-based search and display method provided by another embodiment of the present application;

[0022] Figure 7 It is a schematic diagram for summarizing hot news and related news topics and generating event nodes;

[0023] Figure 8 This is a diagram of the vector retrieval architecture;

[0024] Figure 9 This is an exemplary interface diagram for returning to the news context;

[0025] Figure 10This is a structural block diagram of a news context-based search and display device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0027] In this document, "plurality" refers to two or more. "And / or" describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone. The character " / " generally indicates an "or" relationship between the associated objects.

[0028] Figure 1 This is an example diagram of news search based on keyword matching in related technologies. Taking the commonly used news search as an example, after searching for the keyword "car price reduction", the page displayed is full of various news containing the keyword. For the service provider's search engine, the news returned after the search is basically based on the most recent release, that is, the current hot news time. For social hot topics or public opinion topics, when the duration is long, many major node events and reversal information will occur in the middle. These intermediate information generally require users to search and figure out the whole story on their own. As for another type of technology that displays the context of news. Figure 2 This is a schematic diagram of a news thread display method used in related technologies. A search engine search for the keyword "XXX live streaming with goods" returns all news threads related to the event. However, it should be noted that the news collected by this news thread is manually calibrated by the backend, not automatically created by the site's search engine. Without timely calibrating of news by backend staff, the news thread cannot be updated. Furthermore, the news threads displayed by the search engine are all condensed news headlines, making it difficult to directly understand the content of the news. One still needs to click and read each piece of news to grasp the information.

[0029] Figure 3 Flowchart of a news context-based search and display method provided by an embodiment of the present application, comprising the following steps:

[0030] S1, based on the user's attention to each piece of news information within a set time period, hot news is determined and included in the news candidate library.

[0031] The timeframe can be set by the search engine's server or backend to determine the scope of news crawling. It can be set to news within a year or other timeframes. Hot news or topics can be identified by metrics such as user clicks or reads within a certain period. News context primarily targets topical news with a long time span, rich information content, and wide social impact. Therefore, this context requires judgment when indexing news. This inclusion facilitates subsequent crawling of related news and identifies topic types, such as "Wang Moumou films a new movie" or "Car price reductions."

[0032] S2 summarizes hot news topics and generates event nodes. It then performs news search and vector conversion based on event nodes, news topics, and news content, and establishes a vector inverted index in the vector index library.

[0033] Topic summarization is used to classify hot news, determine the topic type, and generate event nodes based on the topic type. Event nodes are used to describe the content information of hot news under each topic, that is, news summaries generated based on the news content (because most news do not have summaries). For example, an event node for news about car price reductions is described as: ABC car prices will be reduced in March. The "car price reduction" is a specific news topic type, and the news content is the web page content, including but not limited to text, pictures, and videos. In this embodiment, the event nodes, news topics, and news content are packaged as a whole and converted into vectors, and then an inverted index is established in the vector index library. The vector retrieval technology based on the inverted index can express each entity (text, pictures, videos, etc.) in a mathematical vector manner, and map the entity to some points in space. Search and matching are performed based on the inverted index in subsequent indexing.

[0034] S3, in response to receiving the search request, performs vector conversion on the request content, performs vector index matching with the inverted index established in the vector index library, and generates a news context list based on the event nodes and index results.

[0035] Figure 4 This is a framework diagram for searching and displaying news context based on an inverted index provided in an embodiment of the present application.

[0036] Step 1 involves crawling and retrieving news content after the webmaster (news publisher) posts it to the webpage. Step 2 involves storing the crawled news image, i.e., in the webpage database. Step 3 summarizes topics and generates event nodes. Steps 3 through 4 are transformed into vectors using the transformation model. Step 4 then creates an inverted index and stores it in the vector index library.

[0037] The search request is the search content sent by the user, which can be search text content or image content. Figure 4Steps a and b are transformed into vectors using a transformation model, and then matched against the index data in the vector index library. Step c sorts the matched data to generate a news context, and step d returns the context to the user for display. This solution sorts the returned news context by event nodes, for example, displaying it in ascending or descending order based on news release time. Displayed content includes, but is not limited to, time and news summaries.

[0038] The traditional method is to split the text into keywords, then match them against the news database and return a list. This can lead to inaccurate results due to inaccurate keywords. Furthermore, if the relevant news for the keyword hits hasn't been manually tagged and contextualized by the backend, the results won't be returned.

[0039] In one possible implementation, Figure 5 The following chart compares keyword and vector searches. When a user enters "BBA Cars," traditional word segmentation searches split it into "BBA" and "Cars," returning only the index list for these two keywords. However, using vector control mapping, "BBA Cars" is transformed into a vector and indexed using word vectors. This results in "B1, B2, and A1 Cars," which meets user expectations.

[0040] In summary, the embodiment of the present application determines hot news by crawling and analyzing the published news according to the set time, and stores it in the news candidate library to facilitate the subsequent capture of related news; for the hot news included in the candidate news library, the topic is summarized according to the news content, and an event node describing the hot news content information under the topic is generated. In order to improve the accuracy of retrieval and display, a vector indexing method is used to convert hot news and corresponding event nodes and topics into vectors, and an inverted index is established in the vector index library to facilitate subsequent user searches. In the user search process, the keyword unique search method is not used, but the text content is converted into a vector, and the news is hit and returned based on the spatial mapping relationship with the vector data in the vector index library. This calculation method based on vector space mapping can improve the accuracy of retrieval, so that the returned news context is more accurate. Compared with the traditional list-based display method, it can greatly reduce the time cost of user search, help users quickly understand news events, and improve the user's news click-through rate in the site search scenario.

[0041] Figure 6 This is a flowchart of a news context-based search and display method provided by another embodiment of the present application, comprising the following steps:

[0042] Step 601: crawl all web news in the web news library based on the news release time, and obtain at least one of the news click volume, comment volume, and forwarding volume.

[0043] Step 602: Calculate news attention based on the number of news clicks, comments, and reposts; when the news attention exceeds a set threshold, it is determined as hot news and included in the news candidate library.

[0044] All the news included in the candidate database are hot news, that is, news with high attention. The news candidate database is a subset of the web page database.

[0045] Step 603: Determine the newly added hot news as the initial event news, generate the initial event node based on the news content and news time, and determine the news topic.

[0046] Hot news in the news candidate database are categorized by topic, such as "car price reduction" and "ABB star." For newly added hot news, keyword information must first be obtained. This keyword information can be determined based on the news summary and text content, as keywords are crucial for identifying topics. News content includes at least one of the title, summary, and text content. News time includes at least one of the news release information and the time of the event.

[0047] In one possible implementation, the keyword information of the initial event is determined based on the news release time and news summary content (based on news text content if there is no summary), and the news topic corresponding to the initial event node is determined based on the keyword information, and the newly added hot news is moved into the news topic.

[0048] It should be noted that the first appearance of a topic type is not necessarily the beginning of a thread, because the initial news may not necessarily become popular. Therefore, it is possible to capture news that has not reached the same popularity as the hot news in the future. If the candidate news library already has the same news topic as the initial event news, it means that such a topic has been established. The newly added hot news meets the requirement of attracting widespread social attention at the beginning of its release, so it can be directly added to the topic. The other existing hot news are related news (related news must already have an event node) and are used to generate news threads in the future.

[0049] Step 604: crawl related news on the web page based on the initial event node and extract the news body of the related news.

[0050] Related news is news that is still included in the candidate news library, such as news that was published a long time ago or has not yet reached the inclusion criteria. Related news is obtained by crawling based on the keyword information of the initial event news. The extracted news body is used to determine the news topic and time information. The news body includes at least a first-degree body and a second-degree body. The first-degree body is the keyword information used to describe the topic news, and the second-degree body is used to describe the news time information of the topic news. Figure 7 This is a schematic diagram for summarizing hot news and related news topics and generating event nodes. Because the first-degree subject is used to describe keyword information, related news can also be searched based on the event node of the initial event. All relevant news is extracted. Taking car price reductions as an example, the first-degree subject can be expressed as "car" and "price reduction", while the second-degree subject can be described as "this year" or "March" and other time information. Because the time basis displayed by different users' searches varies slightly, there is a certain difference between the news editing time and the event occurrence time. Therefore, this time information can be further described as the news release time or the event occurrence time. For example, if the event occurrence time exceeds three years after the news editing time, the crawler's news is clearly not relevant to the topic and will be filtered out.

[0051] Step 605: determining related news based on the news body, and generating relevant event nodes according to the news content and news body of the related news.

[0052] After the first-degree and second-degree subject matching described above, the related news corresponding to the initial event can be screened out and integrated to generate related event nodes, that is, the event nodes of the related news (news summaries with time information).

[0053] Of course, in some other embodiments, more subjects may be included to describe the event in more detail, which is also conducive to subsequent indexing and accurate display. This application does not limit the number and representational significance of specific news subjects.

[0054] The method of generating news summaries based on first-degree and second-degree subjects can adopt the key-point generative summary model Leader+Writer. The Leader-Writer model mainly generates summaries by mining the key points in the conversation (such as background, keywords, and conclusions). The auxiliary task of key point sequence prediction is introduced, and the key point sequence information of the conversation is used to guide the model to generate summaries that are logical, complete, and have correct key information. The Leader-Writer model encodes each utterance with a layer of Transformer encoder, classifies the key points of each utterance with a Leader decoder, and generates a summary with a Writer decoder. The output of the Leader decoder decoding is used as the input of the initial state of the Writer decoder to utilize the key point information of different conversation segments.

[0055] Step 606 : The initial event news and related news are converted into vectors based on the news content and event nodes, and stored in the vector database. The vector index library creates an inverted index based on the added insertion vector to generate vector index data.

[0056] For each news item matching the topic, its content and event nodes are converted to vectors. The resulting vectors are then stored in the vector database. Furthermore, the vector index database is updated based on the added insertion vectors, essentially re-performing the inverted index to generate vector index data. When a subsequent user search request is received, the spatial similarity of each vector is calculated based on the vector index data in this database.

[0057] Figure 8 This is a vector retrieval architecture diagram. The news context is usually not established based on user request triggers, but is automatically established by the server background according to the set mechanism. Therefore, the process of converting the web page database to the deep model (step 1) and storing the vector database (step 2) is done in advance. The deep model converts the query text in real time based on the request. Step 3 corresponds to the process of calculating the similarity based on the spatial vector, and then generating the ranking and returning (steps 4 and 5) to the user end. The calculation steps and principles of the inverted index do not fall within the scope of protection of the end point of this application and will not be repeated here.

[0058] Step 607: Obtain the text content in the search request and perform vector conversion to obtain a query vector.

[0059] For web search engines, the search request is text content, which is converted into a query vector through a deep model.

[0060] The deep model is deployed on the server in advance, and its training scheme for automatically outputting node sequence is as follows:

[0061] (1) In the initial stage, tens of thousands of manual annotations of news context, event nodes and sequences are required to train the cold start model.

[0062] (2) In the mid-term, after the cold start model is launched, some hot events can be pushed in a targeted manner, and data iteration and training can be performed based on user click data. At the same time, thousands of manually labeled data are collected every month for incremental training.

[0063] (3) In the later stage, when the confidence of online data reaches a certain level, full training can be performed entirely through online click data without manual labeling.

[0064] Step 608: Match the query vector with the vector index data established in the vector index library, calculate the vector similarity and generate a vector index list.

[0065] The vector index list generated at this time is completely arranged based on similarity. Some news in the candidate vectors may not belong to the target topic, which is a retrieval error. In this case, they need to be screened again.

[0066] Step 609 : Screen and sort the event nodes corresponding to the candidate vectors to obtain the target vector.

[0067] This process mainly screens and sorts event nodes, and specifically includes the following steps:

[0068] A. Obtain the event nodes of each candidate vector in the vector index list, match them with the first-degree subject, determine the topic type of the candidate vector with the highest vector similarity as the target topic, and perform secondary screening on the remaining candidate vectors to obtain the target vector;

[0069] Here, we use the first-order subject as the matching criterion for similarity calculation. This means that vectors containing the same keyword are necessarily more similar than those without the keyword. The purpose of determining the target topic is to filter out other types of candidate news, resulting in only the target news (corresponding vector) under the same target topic.

[0070] B. Sort the vectors based on the second-degree subject of the target vector to obtain a list of news context vectors;

[0071] The existence of the second-degree subject is used for sorting, the aforementioned positive or reverse order, based on the time when the news editor published it or the time when the event occurred, etc., to obtain a list of news context vectors.

[0072] C. Crawl target news from the web news library based on the news context vector list, and generate a news context list of the target hot news based on the list order, and return it to the sending end for display.

[0073] Step 610: Send the target vector to the vector database, and capture the target news based on the selected target vector and return it.

[0074] The returned content includes news time and event nodes, etc. For news that already have a summary, the summary information will be displayed. Figure 9 The figure shows an exemplary interface for displaying the news context. Taking the search for "car price reduction" as an example, the event summary displays hot news topics and event node information in chronological order. This kind of context news display is only generated after becoming a hot topic, that is, it is a complementary solution to keyword retrieval in the form of vector indexing, because web news still needs to be displayed when the hot topic is not met. So in Figure 9At the end of the news context, regular web news returned by keyword search is displayed. These two complementary display methods can significantly reduce user search time, help users quickly understand news events, and increase user click-through rates for news in site searches.

[0075] Figure 10 : is a structural block diagram of a news context-based search and display device provided in an embodiment of the present application, the device comprising:

[0076] Hot news determination module 1001, for determining hot news based on the user's attention to each piece of news information within a set time period, and adding the hot news to the news candidate library;

[0077] Node conversion module 1002 is used to summarize the hot news topics and generate event nodes, perform news search and vector conversion based on the event nodes, news topics, and news content, and establish a vector inverted index in the vector index library; topic summarization is used to classify the hot news, and the event nodes are used to describe the content information of the hot news under each topic;

[0078] The news context generation module 1002 is used to convert the request content into a vector in response to receiving a search request, perform vector index matching with the converted vector and the inverted index established in the vector index library, and generate a news context list based on the event node and the index result.

[0079] On the other hand, the present application also provides a server, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the search and display method of news context described in any of the above aspects.

[0080] On the other hand, the present application also provides a computer-readable storage medium, which stores at least one instruction, at least one program, code set or instruction set. The at least one instruction, the at least one program, the code set or instruction set is loaded and executed by a processor to implement the search and display method of news context described in any of the above aspects.

[0081] The above describes the preferred embodiments of the present invention; it should be understood that the present invention is not limited to the above-mentioned specific embodiments, and the devices and structures not described in detail should be understood to be implemented in a common manner in the art; any technician familiar with the art can make many possible changes and modifications without departing from the technical solution of the present invention, or modify them into equivalent embodiments with equivalent changes, which does not affect the essential content of the present invention; therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention that do not depart from the content of the technical solution of the present invention are still within the scope of protection of the technical solution of the present invention.

Claims

1. A search and display method based on news context, characterized in that: The method comprises: S1, based on the user's attention to each news information within a set time, determine the hot news and include it in the news candidate library; S2, summarize the hot news topics and generate event nodes, perform news search and vector conversion based on the event nodes, news topics and news content, and establish a vector inverted index in the vector index library; topic summary is used to classify the hot news, and the event nodes are used to describe the content information of the hot news under each topic; specifically, the newly added hot news is determined as the initial event news, and the initial event node is generated and the news topic is determined based on the news content and news time; crawl the web page related news based on the initial event node, and extract the news body of the related news; the news body includes at least a first-degree body and a second-degree body, the first-degree body is the keyword information used to describe the topic news, and the second-degree body is used to describe the news time information of the topic news; determine the related news based on the news body, and generate related event nodes according to the news content of the related news and the news body; convert the initial event news and the related news into vectors based on the news content and the event nodes, and store them in the vector database; the vector index library establishes an inverted index based on the inserted vector; S3, in response to receiving a search request, performs vector conversion on the request content, performs vector index matching with the converted vector and the inverted index established in the vector index library, and generates a news context list based on the event nodes and index results; specifically, determines the target topic and filters to obtain the target vector based on the first-degree subject, and performs vector sorting based on the second-degree subject to obtain a news context vector list.

2. The search and display method based on news context according to claim 1, characterized in that: Step S1 includes: crawling all web news in the web news library based on news release time, and obtaining at least one of the number of clicks, comments, and forwarding of the news; The news attention is calculated based on the number of news clicks, comments and reposts; when the news attention exceeds a set threshold, it is determined as hot news and included in the news candidate library.

3. The search and display method based on news context according to claim 2, characterized in that: The news content includes at least one of title information, summary information and text content information, and the news time includes at least one of news release information and event occurrence time; Step S21 includes: The keyword information of the initial event is determined based on the news content and the news time, and the news topic corresponding to the initial event node is determined based on the keyword information, and the newly added hot news is moved into the news topic.

4. The search and display method based on news context according to claim 1, characterized in that: Step S3 includes: S31, obtaining the text content in the search request and performing vector conversion to obtain a query vector; S32, matching the query vector with the inverted index data established in the vector index library, calculating vector similarity and generating a vector index list; the vector index list includes candidate vectors that meet the vector similarity; S33, filtering and sorting the event nodes corresponding to the candidate vectors to obtain a target vector; S34, sending the target vector to the vector database, capturing target news based on the selected target vector and returning it.

5. The search and display method based on news context according to claim 4, characterized in that: Step S33 includes: Obtaining the event node of each candidate vector in the vector index list, matching the first-degree subject, determining the topic type of the candidate vector with the highest vector similarity as the target topic, and performing secondary screening on the remaining candidate vectors to obtain the target vector; Sorting vectors of the second-degree subjects based on the target vector to obtain a news context vector list; Based on the news context vector list, target news is crawled from the webpage news library, and based on the list order, the news context list of target hot news is generated and returned to the sending end for display.

6. The search and display method based on news context according to claim 5, characterized in that: When the secondary subject is the time of news release, the news context is arranged in ascending or descending order based on the time of news release; When the second-degree subject is the time of occurrence of an event, the news context is arranged in ascending or descending order based on the time of occurrence of the event.

7. The search and display method based on news context according to claim 1, characterized in that: When the same news topic as the initial event news already exists in the candidate news library, all the hot news under the same news topic are directly determined as the related news.

8. A search and display device based on news context, characterized in that: The device comprises: Hot news determination module, used to determine hot news based on the user's attention to various news information within a set time, and include it in the news candidate library; A node conversion module is used to summarize the hot news topics and generate event nodes, perform news search and vector conversion based on the event nodes, news topics and news content, and establish a vector inverted index in the vector index library; topic summarization is used to classify the hot news, and the event nodes are used to describe the content information of the hot news under each topic; specifically, the newly added hot news is determined as the initial event news, and the initial event node is generated and the news topic is determined based on the news content and news time; crawl related news on the web page based on the initial event node, and extract the news body of the related news; the news body includes at least a first-degree body and a second-degree body, the first-degree body is the keyword information used to describe the topic news, and the second-degree body is used to describe the news time information of the topic news; based on the news body, the related news is determined, and related event nodes are generated according to the news content of the related news and the news body; the initial event news and the related news are vector-converted based on the news content and the event nodes, and stored in the vector database; the vector index library establishes an inverted index based on the inserted vector; The news context generation module is used to, in response to receiving a search request, perform vector conversion on the request content, perform vector index matching with the inverted index established in the vector index library through the converted vector, and generate a news context list based on the event node and the index result; specifically, determine the target topic and filter to obtain the target vector based on the first-degree subject, and perform vector sorting based on the second-degree subject to obtain a news context vector list.

Images