Event context information generation method and related equipment

Abstract

The invention discloses an event context information generation method and related equipment. The method comprises the steps of obtaining an event knowledge graph corresponding to a target event; searching event background knowledge information of each sub-event corresponding to the target event through the event knowledge graph; based on the event background knowledge information of each sub-event and the event basic description information of the target event, determining at least one key sub-event of the target event from each sub-event; performing event structure analysis processing on the key sub-events according to the event background knowledge information of the key sub-events and the event basic description information of the key sub-events to obtain event element information of the key sub-events; and based on the event element information of each key sub-event, generating event context information of the target event. According to the method and the device, automatic generation of the event context information is realized, and more reliable background knowledge can be obtained through the event knowledge graph. Therefore, the acquisition efficiency and accuracy of the event venation information are improved.

Description

Technical Field

[0001] The present application relates to the field of computers, and in particular to a method for generating event context information and related equipment, wherein the related equipment includes an event context information generation device, an electronic device, a storage medium and a program product. Background Art

[0002] With the rapid development of the Internet, information has exploded, and the number of reports and information sources of various events has increased day by day. As events progress, relevant information will continue to update and change. In this complex and ever-changing information environment, in order to fully and accurately understand the progress of events, it is usually necessary to systematically sort out the events to obtain the development context of the events, presenting the entire process from the cause of the event to each important stage.

[0003] However, in an environment with excessive information, users often need to actively read a large amount of relevant content about an event in order to clearly determine the overall picture of the event and its evolution process. This is not conducive to quickly obtaining accurate event context information, and the event context information obtained is not necessarily accurate. Summary of the invention

[0004] The embodiments of the present application provide a method for generating event context information and related equipment, which not only improves the efficiency of obtaining event context information, but also avoids deviations and omissions that may occur in the manual sorting process, which is conducive to ensuring that the event context information is highly reliable in terms of accuracy and objectivity, and can truly and comprehensively present the evolution process of the target event.

[0005] The present application provides a method for generating event context information, including: Obtaining an event knowledge graph corresponding to a target event, wherein the event knowledge graph represents an association relationship between the target event and at least one sub-event, and an association relationship between the sub-event and at least one content; Through the event knowledge graph, retrieve the event background knowledge information of each sub-event corresponding to the target event; Based on the event background knowledge information of each sub-event and the event basic description information of the target event, determining at least one key sub-event of the target event from each sub-event; According to the event background knowledge information of the key sub-event and the event basic description information of the key sub-event, the event structure analysis and processing of the key sub-event is performed to obtain the event element information of the key sub-event; Based on the event element information of each key sub-event, the event context information of the target event is generated.

[0006] The present application also provides a device for generating event context information, including: An event knowledge graph acquisition unit for acquiring an event knowledge graph corresponding to a target event, where the event knowledge graph represents the association relationship between the target event and at least one sub-event, and the association relationship between the sub-event and at least one piece of content; An information retrieval unit for retrieving event background knowledge information of each sub-event corresponding to the target event through the event knowledge graph; An event determination unit for determining at least one key sub-event of the target event from each sub-event based on the event background knowledge information of each sub-event and the event basic description information of the target event; An event analysis unit for performing event structure analysis processing on the key sub-event according to the event background knowledge information and the event basic description information of the key sub-event to obtain event element information of the key sub-event; A context generation unit for generating event context information of the target event based on the event element information of each key sub-event.

[0007] In some embodiments, the event knowledge graph includes multiple nodes and edges between the nodes. The nodes include a parent event node corresponding to the target event, a sub-event node corresponding to the sub-event, and a content node corresponding to the content. The edges represent the association relationship between the target event and the sub-event, or the association relationship between the sub-event and the content; The information retrieval unit is used for: Searching for sub-event nodes corresponding to each sub-event corresponding to the target event in the event knowledge graph; Respectively determining local sub-graphs corresponding to each sub-event node in the event knowledge graph. The local sub-graph of the sub-event node includes the sub-event node, the content nodes connected to the sub-event node, and the edges connecting the sub-event node and the content nodes; Determining the event background knowledge information of each sub-event node according to the association relationship between the sub-event and the content in the local sub-graph of each sub-event node.

[0008] In some embodiments, the apparatus further includes an element evaluation unit and an element correction unit: The element evaluation unit is used for performing evaluation processing on the event element information of the key sub-event in at least one dimension through the event knowledge graph to obtain evaluation information; The element correction unit is used for, when the evaluation information indicates that the evaluation of the event element information of the key sub-event fails, correcting the event element information of the key sub-event based on the evaluation information, the event background knowledge information of the key sub-event, and the event basic description information of the key sub-event to generate new event element information.

[0009] In some embodiments, the element evaluation unit is used for: Through the event knowledge graph, perform an evaluation process on the element accuracy of the event element information of the key sub-events to obtain accuracy evaluation information; Through the event knowledge graph, perform a description quality evaluation process on the event element information of the key sub-events to obtain quality evaluation information; Through the event knowledge graph, evaluate the importance of the key sub-events to the target event.

[0010] In some embodiments, the context generation unit is used for Based on the event knowledge graph and the event basic description information of the target event, evaluate the logical relationship between the key sub-events of the target event to obtain event evaluation result information; According to the event evaluation result information, select target key sub-events from the key sub-events; Generate event context information of the target event based on the event element information of the target key sub-events.

[0011] In some embodiments, the context generation unit is used for when the event evaluation result information indicates that there are key sub-events to be removed among the key sub-events, based on the event evaluation result information, the basic event description information of the target event, and the event knowledge graph, optimize each key sub-event to select target key sub-events.

[0012] In some embodiments, the event determination unit is used for: Through the key sub-event recognition model, based on the event background knowledge information of each sub-event, the event basic description information of each sub-event, and the event basic description information of the target event, respectively predict the probability information of each sub-event belonging to the key sub-event of the target event; According to the probability information, determine at least one key sub-event of the target event.

[0013] In some embodiments, the device further includes a content acquisition unit, an information aggregation unit, an event aggregation unit, and a graph construction unit: The content acquisition unit is used to acquire a plurality of initial media contents, identify event information of the plurality of initial media contents, and select media contents containing event information from the plurality of initial media contents according to the recognition results; The information aggregation unit is used to perform an aggregation process on the event information of the media contents to obtain at least one sub-event, and establish an association relationship between each sub-event and each media content; The event aggregation unit is used to perform an aggregation process on at least one sub-event to obtain at least one event, and establish an association relationship between each event and the sub-events; A graph construction unit for constructing an event graph according to the association relationships between each sub - event and each media content, and the association relationships between each event and each sub - event. The event graph includes event knowledge graphs corresponding to each event.

[0014] In some embodiments, an information aggregation unit is configured to perform information comparison on the event information of each media content to deduplicate the repeatedly occurring event information, obtain sub - events corresponding to the repeatedly occurring event information, and mount the media content corresponding to the repeatedly occurring event information to the corresponding sub - events.

[0015] In some embodiments, a graph construction unit is configured to establish an initial event graph based on the association relationships between sub - events and content, and the association relationships between events and sub - events. The initial event graph includes initial event knowledge graphs corresponding to each event. Each initial event knowledge graph corresponding to an event includes a parent event node of the corresponding event, sub - event nodes corresponding to each sub - event of the event, content nodes of the media content corresponding to each sub - event, and edges between the nodes. The edges represent the association relationships between events and sub - events, or the association relationships between sub - events and media content. Perform representation learning on each node and edge in the initial event graph to obtain the event graph.

[0016] In some embodiments, the apparatus further includes a graph update unit, configured to: Obtain incremental sub - events; Based on the event - based description information of the incremental sub - events and the event - based description information of each event in the event graph, calculate the event correlation between the incremental sub - events and each event in the event graph; When the event correlation is greater than a preset correlation, calculate the event aggregation score between the incremental sub - event and the related event based on the media content corresponding to the incremental sub - event and the media content corresponding to each sub - event of the related event; According to the event aggregation score, establish the association relationship between the incremental sub - event and the related event, and update the event graph.

[0017] In some embodiments, the apparatus further includes a graph update unit, configured to: Determine multiple incremental sub - events and the media content corresponding to each incremental sub - event; Calculate the event correlation of the multiple incremental sub - events based on the event - based description information of each incremental sub - event; When the event correlation is greater than a preset correlation, calculate the sub - event aggregation score between the incremental sub - event and the related incremental sub - event according to the media content corresponding to the incremental sub - event and the media content corresponding to its related incremental sub - event; When the sub-event aggregation score is greater than a preset aggregation score, aggregate the incremental sub-events and their related incremental sub-events to obtain new events, and update the event graph based on the association relationship between the new events and their related incremental sub-events, and the association relationship between the incremental sub-events related to the new events and the corresponding media content.

[0018] In some embodiments, the apparatus further includes a heat calculation unit and a graph management unit, configured to: Calculate the event heat information of each event; Delete the event knowledge graph corresponding to each event in the event graph according to the event heat information.

[0019] In some embodiments, the heat calculation unit is configured to determine the first heat sub-heat according to the query frequency of the event; determine the second heat sub-heat according to the query time information; determine the third heat sub-heat according to the event occurrence time information; and determine the event heat information according to the first heat sub-heat, the second heat sub-heat, and the third heat sub-heat.

[0020] In some embodiments, the event analysis unit is configured to perform identification processing on at least one preset event element of the background knowledge information of the key sub-event and the event basic description information of the key sub-event through an event element identification model, to obtain the element content of the key sub-event under each preset event element; and perform content sorting processing on the element content of the key sub-event under each preset event element based on the element priority of each preset event element, to obtain the event element information of the key sub-event.

[0021] An embodiment of the present application further provides an electronic device, including a processor and a memory, where the memory stores multiple instructions; the processor loads the instructions from the memory to execute the steps in any one of the event context information generation methods provided by the embodiments of the present application.

[0022] An embodiment of the present application further provides a computer-readable storage medium, where the computer-readable storage medium stores multiple instructions, and the instructions are suitable for being loaded by a processor to execute the steps in any one of the event context information generation methods provided by the embodiments of the present application.

[0023] An embodiment of the present application further provides a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps in any one of the event context information generation methods provided by the embodiments of the present application are implemented.

[0024] An embodiment of the present application can obtain an event knowledge graph corresponding to a target event. The event knowledge graph represents the association relationship between the target event and at least one sub-event, as well as the association relationship between the sub-event and at least one piece of content. Through the event knowledge graph, retrieve the event background knowledge information of each sub-event corresponding to the target event. Based on the event background knowledge information of each sub-event and the event basic description information of the target event, determine at least one key sub-event of the target event from each sub-event. According to the event background knowledge information and the event basic description information of the key sub-event, perform event structure analysis processing on the key sub-event to obtain the event element information of the key sub-event. Based on the event element information of each key sub-event, generate the event context information of the target event.

[0025] In the present application, since the event knowledge graph corresponding to the target event represents the association relationship between the target event and at least one sub-event, as well as the association relationship between the sub-event and at least one piece of content, thus, through this event knowledge graph, the event background knowledge information of each sub-event corresponding to the target event can be quickly retrieved, and based on the event background knowledge information of each sub-event and the event basic description information of the target event, the key sub-events that play a promoting role in the development of the target event can be screened out. By further analyzing the event background knowledge information and the event basic description information of the key sub-events, event structure analysis processing can be performed on the key sub-events, so as to extract the event element information of the key sub-events. Finally, based on the event element information of each key sub-event, the event context information of the target event can be generated. Since the process of generating this event context information is automated and completely requires no manual intervention, and more reliable background knowledge can be obtained through the event knowledge graph, it not only greatly improves the acquisition efficiency of the event context information, but also avoids the biases and omissions that may occur in the manual collation process. In this way, the generated event context information can be fully guaranteed in terms of accuracy and objectivity, ensuring that it can comprehensively and truly present the evolution process of the target event. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those skilled in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0027] Figure 1a is a schematic diagram of the scenario of the method for generating event context information provided by an embodiment of the present application; Figure 1b is a schematic flowchart of the method for generating event context information provided by an embodiment of the present application; Figure 2aIt is a schematic diagram of the application of the method for generating event context information provided by an embodiment of the present application in various scenarios of multi-agent collaboration; Figure 2b It is a flowchart for constructing an event graph provided by an embodiment of the present application; Figure 2c It is a flowchart for generating event context information provided by an embodiment of the present application; Figure 2d It is a schematic diagram of a scenario for evaluating event elements provided by an embodiment of the present application; Figure 2e It is a schematic diagram of a scenario for evaluating key sub-events provided by an embodiment of the present application; Figure 3 It is a schematic diagram of the structure of an apparatus for generating event context information provided by an embodiment of the present application; Figure 4 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0028] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present application.

[0029] An embodiment of the present application provides a method for generating event context information and related devices. The related devices include an apparatus for generating event context information, an electronic device, a storage medium, and a program product.

[0030] Among them, the apparatus for generating event context information can be specifically integrated in an electronic device, and the electronic device can be a device such as a terminal or a server. Among them, the terminal can be a device such as a mobile phone, a tablet computer, a smart Bluetooth device, a laptop computer, or a personal computer (PC); the server can be a single server or a server cluster composed of multiple servers.

[0031] In some embodiments, the apparatus for generating event context information can also be integrated in multiple electronic devices. For example, the apparatus for generating event context information can be integrated in multiple servers, and the method for generating event context information of the present application is implemented by multiple servers.

[0032] In the embodiments of the present application, the term "module" or "unit" refers to a computer program with a predetermined function or a part of a computer program, which works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, one processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of the overall module or unit that includes the function of the module or unit.

[0033] It can be understood that in the specific implementation of the present application, when it comes to data related to media content, etc., when the embodiments of the present application are applied to specific products or technologies, user permission or consent needs to be obtained, and the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of the relevant countries and regions.

[0034] For example, referring to Figure 1a , the electronic device can obtain an event knowledge graph corresponding to the target event, where the event knowledge graph represents the association relationship between the target event and at least one sub-event, as well as the association relationship between the sub-event and at least one piece of content; through the event knowledge graph, retrieve the event background knowledge information of each sub-event corresponding to the target event; based on the event background knowledge information of each sub-event and the event basic description information of the target event, determine at least one key sub-event of the target event from each sub-event; according to the event background knowledge information of the key sub-event and the event basic description information of the key sub-event, perform event structure analysis processing on the key sub-event to obtain the event element information of the key sub-event; based on the event element information of each key sub-event, generate the event context information of the target event.

[0035] In this application, due to the fact that the event knowledge graph corresponding to the target event represents the association relationship between the target event and at least one sub-event, as well as the association relationship between the sub-event and at least one piece of content, thus, through this event knowledge graph, the event background knowledge information of each sub-event corresponding to the target event can be quickly retrieved, and based on the event background knowledge information of each sub-event and the event basic description information of the target event, the key sub-events that play a promoting role in the development of the target event can be screened out. By further analyzing the event background knowledge information and event basic description information of the key sub-events, event structure analysis processing can be performed on the key sub-events, so as to extract the event element information of the key sub-events. Finally, based on the event element information of each key sub-event, the event context information of the target event can be generated. Since the generation process of this event context information is automated and completely requires no manual intervention, it not only greatly improves the acquisition efficiency of the event context information, but also avoids the biases and omissions that may occur during the manual collation process. In this way, the generated event context information can be fully guaranteed in terms of accuracy and objectivity, ensuring that it can comprehensively and truly present the evolution process of the target event.

[0036] The following will be described in detail respectively. It should be noted that the serial numbers of the following embodiments do not limit the preferred order of the embodiments.

[0037] In this embodiment, a method for generating event context information is provided. As Figure 1b shown, the specific process of this method for generating event context information can be as follows: 101. Obtain the event knowledge graph corresponding to the target event. The event knowledge graph represents the association relationship between the target event and at least one sub-event, as well as the association relationship between the sub-event and at least one piece of content.

[0038] Among them, the target event refers to a specific event whose evolution process needs to be analyzed in detail. Such events can cover multiple fields, such as historical events, enterprise projects, technological breakthroughs, popular events, etc. In other words, target events are those events that need to trace their development context, key nodes, and influencing factors. Whether it is a major turning point in the historical process, a major project advancement within an enterprise, an innovation breakthrough in the technology field, or a hot issue at the social level, they can all be regarded as target events. By sorting out these events, it is possible to better understand their background, process, and results.

[0039] The event knowledge graph clearly shows the association relationship between the target event and at least one sub-event, as well as the association relationship between the sub-event and at least one specific piece of content in a graphical form. It presents the complex relationship network of the target event and its related sub-events in an intuitive structure, providing strong support for the analysis and understanding of the target event.

[0040] A sub - event refers to a specific event that is interrelated with the target event during the evolution of the target event. For example, a sub - event may be a key node in a historical event, a phased task in an enterprise project, an experimental link in a technological breakthrough, or a specific dynamic in a trending event, etc.

[0041] The association relationship between the target event and the sub - event indicates that the sub - event is a component of the target event. This association relationship can specifically be a causal relationship, an influence relationship, a dependency relationship, a supplementary relationship, a parallel relationship, etc.

[0042] The content is a detailed description, supplementary explanation, or specific presentation of the sub - event associated with it, and is used to provide information such as the background, details, and characteristics of the sub - event.

[0043] The association relationship between the sub - event and the content indicates that the content is the source of relevant information for the sub - event. This association relationship can specifically be an information support relationship, an explanatory relationship, a proof relationship, a background supplementary relationship, a dependency relationship, an extension relationship, etc.

[0044] In some embodiments, the event knowledge graph corresponding to the target event can be obtained locally or remotely.

[0045] In some embodiments, in order to identify, aggregate, and associate event information in different media contents, a multi - level and structured event graph is constructed. Before obtaining the event knowledge graph corresponding to the target event, it further includes: Obtain multiple initial media contents, identify event information in the multiple initial media contents, and select the media contents containing event information from the multiple initial media contents according to the identification results; Perform aggregation processing on the event information of the media contents to obtain at least one sub - event, and establish an association relationship between each sub - event and each media content; Perform aggregation processing on at least one sub - event to obtain at least one event, and establish an association relationship between each event and the sub - event; Construct an event graph according to the association relationship between each sub - event and each media content, and the association relationship between each event and the sub - event. The event graph includes the event knowledge graph corresponding to each event.

[0046] Among them, the initial media content refers to the original media materials collected from various channels without any editing or processing. The initial media content can specifically be various forms of materials such as articles, videos, pictures, and audios.

[0047] The channel is the dissemination path of the initial media content, covering many forms such as traditional media channels and digital media channels.

[0048] The recognition result can reflect whether the initial media content contains event information. For example, the recognition result can indicate the position of the media content containing event information in the initial media content, can directly indicate that the initial media content contains event information, can only indicate that the initial media content contains event information, and so on.

[0049] Event information is information describing an event in the original media content. For example, event information may include the event title, the time and place of the event, the event participants, the event process, the event impact, and other content.

[0050] The media content containing event information is material content containing event information extracted from the initial media content after identification and screening.

[0051] The event graph shows events, sub-events and their relationship with media content, and connects this information in an orderly manner in the form of a graph.

[0052] It is understandable that by identifying event information from multiple initial media contents, those media contents containing actual event information can be effectively screened out, ensuring that only relevant and valid media contents are extracted. Aggregating event information from different media contents ensures that reports from different perspectives can be classified as the same sub-event or event, which can avoid missing any key information and improve the comprehensiveness of event understanding. By further aggregating sub-events, a complete event is finally obtained, which enables events to be organized and managed at different levels. This structure helps to show the overall picture of the event more clearly and facilitates subsequent analysis. In the process of constructing the event map, establishing the association between each event, sub-event and media content can trace the source and development process of each event, and at the same time, it can also establish logical connections for information from different sources to ensure the accuracy and consistency of the event map. The event map is not just a collection of events, it also contains rich knowledge (such as the views of various media, reporting time, progress of events, etc.). By constructing such a map, in-depth mining and more detailed analysis of events can be achieved.

[0053] In some embodiments, in order to build a more accurate and structured event knowledge graph and ensure that event information from different sources is not repeated, and to improve the quality and efficiency of data processing, the event information of the media content is aggregated to obtain at least one sub-event, including: Comparing the event information of each media content to eliminate repeated event information and obtain sub-events corresponding to the repeated event information; The media content corresponding to the recurring event information is mounted to the corresponding sub-event.

[0054] Among them, the repeatedly occurring event information is obtained by comparing the event information in each media content and screening out the same or highly similar event information. For example, the repeatedly occurring event information can be compared through at least one of the event title, the time and location of the event occurrence, event participants, event process, event impact, etc.

[0055] For example, randomly select two event information from the event information of each media content, such as event information i and event information j, and use calculation methods such as similarity and keyword matching to compare event information i and event information j. When the information comparison indicates that event information i and event information j are repeatedly occurring event information, perform aggregation processing on event information i and event information j to obtain the sub-event e corresponding to event information i and event information j, and mount the media content corresponding to event information i and the media content corresponding to event information j to the corresponding sub-event e.

[0056] In some embodiments, in order to facilitate the retrieval of event-related content through the event graph, an event graph is constructed according to the association relationship between each sub-event and each media content, and the association relationship between each event and the sub-event, including: Based on the association relationship between the sub-event and the content, and the association relationship between the event and the sub-event, an initial event graph is established. The initial event graph includes the initial event knowledge graph corresponding to each event. The initial event knowledge graph corresponding to each event includes the parent event node of the corresponding event, the sub-event nodes corresponding to each sub-event of the event, the content nodes of the media content corresponding to each sub-event, and the edges between the nodes. The edges represent the association relationship between the event and the sub-event, or the association relationship between the sub-event and the media content; Perform representation learning on each node and edge in the initial event graph to obtain the event graph.

[0057] Among them, the association relationship between the sub-event and the content is used to clarify the correspondence between the sub-event and its related content, so that the specific content corresponding to the sub-event can be quickly located and retrieved through this relationship.

[0058] The association relationship between the event and the sub-event is used to clarify the correspondence between the event and its related sub-events, so that the sub-events related to the event can be quickly found through this relationship.

[0059] The initial event graph uses the structured form of nodes and edges to clearly show the hierarchical structure of each event and the corresponding connection between the sub-event and the media content.

[0060] The initial event knowledge graph is constructed around a single event, which covers the sub-events of the event and the association relationship between the sub-events and the media content.

[0061] The parent event node is at the top of the event hierarchy in the initial event knowledge graph, representing the core event that triggers or dominates a series of related events. It is the root or starting point of the sub-event nodes in the initial event knowledge graph, connecting the subordinate sub-event nodes and content nodes to it through a hierarchical relationship. The parent event node provides a framework for the graph, under which the related sub-events and media content can be further expanded, refined, and associated.

[0062] The sub-event node is a node in the initial event knowledge graph used to represent a specific sub-event associated with an event. It is at the next level below the parent event node and is connected to the parent event node by an edge. This connecting edge indicates that there is a certain logical or semantic association relationship between the event and the sub-event, such as a causal relationship, an inclusion relationship, or a time sequence relationship, etc. This hierarchical structure enables the complex information of the event to be clearly organized and presented, facilitating the understanding and analysis of the whole picture of the event and the interactions between its various components.

[0063] The content node is a node in the initial event knowledge graph used to represent the specific media content related to a sub-event. Each content node represents a specific media material, and these materials provide detailed descriptions and specific information sources for the sub-event.

[0064] An edge is used to indicate that there is an association relationship between the two connected nodes.

[0065] Representation learning is used to transform the nodes and edges in the initial event graph into low-dimensional representations to support efficient retrieval. For example, through representation learning, keywords can be generated for each node for explicit queries, or embedding vectors of the nodes can be generated for implicit queries. In addition, representation learning can also generate event description information for the parent event node and the sub-event node.

[0066] In some embodiments, the event index may include event keywords, event embedding vectors, and event description information. Among them, the event keywords refer to the keywords related to the parent event node or the sub-event node, and the event embedding vectors are the embedding vectors corresponding to these parent event nodes or sub-event nodes.

[0067] In some embodiments, in order to dynamically expand and update the knowledge related to the events in the event graph, the method further includes: Obtain incremental sub-events; Based on the event basic description information of the incremental sub-events and the event basic description information of each event in the event graph, calculate the event correlation between the incremental sub-events and each event in the event graph; When the event correlation is greater than the preset correlation, calculate the event aggregation score between the incremental sub-event and the related event based on the media content corresponding to the incremental sub-event and the media content corresponding to each sub-event of the related event; Establish the association relationship between the incremental sub-event and the related event according to the event aggregation score, and update the event graph.

[0068] Among them, the incremental sub-event refers to a newly discovered sub-event, which may be an event that has just occurred recently, or a sub-event that was not previously recorded in the event graph but is related to the event, and so on.

[0069] The event basic description information of the incremental sub-event is a refined description of the content related to the incremental sub-event, aiming to summarize the core elements and key features of the incremental sub-event. These description information can clearly and accurately convey the main content of the incremental sub-event, and ensure coverage of the key dimensions of the incremental sub-event, such as time, place, participants and their mutual relationships, etc., so as to provide a basis for subsequent analysis and understanding.

[0070] The event basic description information of an event is a refined description of the content related to the event in the event graph.

[0071] Event correlation refers to the degree of association between the newly added sub-event and the event in the event graph.

[0072] The preset correlation is a threshold set in advance for evaluating the degree of association between the newly added sub-event and the event in the event graph.

[0073] A related event refers to an event in the event graph whose event correlation with the incremental sub-event exceeds the preset correlation.

[0074] The event aggregation score is used to evaluate the possibility of merging the incremental sub-event and the related event, and measure whether the incremental sub-event can be included in the related event.

[0075] The association relationship between the incremental sub-event and the related event helps to clarify their correspondence, so that the incremental sub-event associated with the related event can be quickly identified by means of this relationship.

[0076] In some embodiments, in order to construct the association knowledge system of new events in the event graph, the method further includes: Determine a plurality of incremental sub-events and the media content corresponding to each incremental sub-event; Calculate the event correlation of a plurality of incremental sub-events based on the event basic description information of each incremental sub-event; When the event correlation is greater than a preset correlation, calculate the sub-event aggregation score between the incremental sub-event and the media content corresponding to its related incremental sub-events based on the media content corresponding to the incremental sub-event and the media content corresponding to its related incremental sub-events. When the sub-event aggregation score is greater than a preset aggregation score, perform an aggregation process on the incremental sub-event and its related incremental sub-events to obtain a new event, and update the event graph based on the association relationship between the new event and its related incremental sub-events and the association relationship between the incremental sub-events related to the new event and the corresponding media content.

[0077] Among them, a related incremental sub-event refers to when the association relationship between a certain incremental sub-event and other incremental sub-events exceeds the preset correlation, the other incremental sub-events are regarded as the related incremental sub-events of this incremental sub-event.

[0078] The sub-event aggregation score is used to measure the possibility of merging between the incremental sub-event and the related incremental sub-events, and evaluate whether they can be integrated into a more extensive event.

[0079] The preset aggregation score is a pre-set scoring criterion used to determine whether the sub-event aggregation score reaches the merging standard. When the sub-event aggregation score is greater than the preset aggregation score, it indicates that the incremental sub-event and the related incremental sub-events have sufficient relevance and consistency and can be merged into a more extensive event; otherwise, they cannot be merged.

[0080] A new event refers to the event newly obtained after aggregating the incremental sub-event and its related incremental sub-events.

[0081] The association relationship between the new event and its related incremental sub-events helps to clarify their corresponding relationship, so as to quickly identify the incremental sub-events related to the new event. Here, the incremental sub-events related to the new event refer to the incremental sub-events used to form the new event during the aggregation process and the related incremental sub-events of this incremental sub-event.

[0082] The association relationship between the incremental sub-events related to the new event and the corresponding media content is used to clarify the correspondence between the incremental sub-events related to the new event and their related media content, so as to quickly locate and retrieve the specific content corresponding to the incremental sub-events related to the new event through this relationship.

[0083] 102. Through the event knowledge graph, retrieve the event background knowledge information of each sub-event corresponding to the target event.

[0084] Among them, the event background knowledge information of the sub-event refers to various information retrieved through the event knowledge graph that can provide a detailed background description and explanation for this sub-event. These information help to clearly present the origin, development process and result of the sub-event.

[0085] In some embodiments, in order to determine the background knowledge information of each sub-event through the event knowledge graph, the event knowledge graph includes multiple nodes and edges between the nodes. The nodes include a parent event node corresponding to the target event, a sub-event node corresponding to the sub-event, and a content node corresponding to the content. The edges represent the association relationship between the target event and the sub-event, or the association relationship between the sub-event and the content; Through the event knowledge graph, retrieve the event background knowledge information of each sub-event corresponding to the target event, including: Search for the sub-event nodes corresponding to each sub-event of the target event in the event knowledge graph; In the event knowledge graph, respectively determine the local sub-graphs corresponding to each sub-event node. The local sub-graph of the sub-event node includes the sub-event node, the content nodes connected to the sub-event node, and the edges connecting the sub-event node and the content nodes; According to the association relationship between the sub-event and the content in the local sub-graph of each sub-event node, determine the event background knowledge information of each sub-event node.

[0086] Among them, a node is the basic unit in the event knowledge graph and is used to represent various entities. These entities can be the target event, the sub-events of the target event, or the content related to the sub-events, etc.

[0087] The edges between the nodes are used to describe the association relationship between two connected nodes.

[0088] The parent event node is the central node in the event knowledge graph and is used to represent the target event.

[0089] The sub-event node is the node connected to the parent event node by an edge in the event knowledge graph and is used to represent the sub-event related to the target event.

[0090] The content node is the node connected to the sub-event node by an edge in the event knowledge graph and is used to represent the content of the sub-event related to the target event. The sub-event node can be connected to multiple content nodes by edges, that is, it can indicate that the sub-event represented by the sub-event node is associated with multiple contents.

[0091] The local sub-graph is a sub-graph constructed with the sub-event node as the center in the event knowledge graph. It consists of the sub-event node, the content nodes connected to the sub-event node, and the edges connecting the sub-event node and these content nodes.

[0092] In some embodiments, through the association relationship between the sub-event and the content in the local sub-graph, the content associated with the sub-event can be retrieved, and the event background knowledge information of the sub-event node can be summarized through these contents.

[0093] 103. Determine at least one key sub - event of the target event from each sub - event based on the event background knowledge information of each sub - event and the event basic description information of the target event.

[0094] Among them, the event basic description information of the target event is the basic information obtained by refining and summarizing the relevant content of the target event, covering the main elements and core features of the target event, aiming to concisely present the overall situation of the target event. For example, the event basic description information can specifically cover the occurrence time, location, main participants, reasons for triggering the event, development process of the event, and final result of the target event, etc.

[0095] A key sub - event refers to a sub - event that promotes the development of the target event during the development process of the target event. For example, a key sub - event can be a sub - event that plays a decisive role in the development of the target event, or a sub - event that promotes the escalation or transformation of the target event, or a sub - event generated by the access of external forces to the target event, and so on.

[0096] In some embodiments, in order to identify the sub - events that promote the development of the target event, based on the event background knowledge information of each sub - event and the event basic description information of the target event, determining at least one key sub - event of the target event from each sub - event includes: Through a key sub - event recognition model, based on the event background knowledge information of each sub - event, the event basic description information of each sub - event, and the event basic description information of the target event, respectively predict the probability information of each sub - event belonging to the key sub - event of the target event; According to the probability information, determine at least one key sub - event of the target event.

[0097] Among them, the key sub - event recognition model is a model used to screen out the sub - events that promote the development of the target event from each sub - event of the target event. For example, the key sub - event recognition model can be a large - language model, a multi - modal model, a content - processing model, a key - sub - event recognition task template based on a large - language model, etc. The content - processing model can specifically be a graph neural network model, a deep - learning model, a decision - tree model, etc. Among them, the key - sub - event recognition task template based on a large - language model aims to guide the large - language model to identify those key sub - events that form the story skeleton from the input content. This template details the content of the task prompt to ensure that the model can specifically focus on those segments that have a decisive impact on the overall event, so as to accurately capture the essence of each key sub - event.

[0098] The probability information represents the possibility of a sub - event becoming a key sub - event of the target event, and its value range is between 0 and 1. The higher the value, the greater the impact of the sub - event on the target event, and the more likely it is to be recognized as a key sub - event of the target event.

[0099] 104. Perform event structure analysis on the key sub-event based on the event background knowledge information of the key sub-event and the event basic description information of the key sub-event to obtain event element information of the key sub-event.

[0100] Among them, event element information refers to the basic components that constitute key sub-events, which can specifically cover time elements, location elements, character elements, event description information and other aspects.

[0101] In some embodiments, in order to quickly generate event element information of a key sub-event, an event structure analysis is performed on the key sub-event according to the event background knowledge information of the key sub-event and the event basic description information of the key sub-event to obtain the event element information of the key sub-event, including: Using the event element recognition model, the background knowledge information of the key sub-event and the event basic description information of the key sub-event are processed to recognize at least one preset event element, so as to obtain the element content of the key sub-event under each preset event element; Based on the element priority of each preset event element, the element content of the key sub-event under each preset event element is sorted to obtain event element information of the key sub-event.

[0102] The event element recognition model is a model for recognizing event elements. For example, the event element recognition model can be a large language model, a multimodal model, a content processing model, an event element recognition template based on a large language model, etc. The event element recognition template based on a large language model is intended to guide the large language model to recognize various event elements of key sub-events from the input content.

[0103] The preset event elements are event elements that are preset in the event element recognition model and need to be recognized. For example, the preset event elements may include time, location, person, description, summary, etc.

[0104] The element content refers to the specific content that matches each preset event element, which is identified from the background knowledge information of the key sub-event and its event basic description information. By using the event element recognition model, the element content of each preset event element can be identified at the same time. For example, the element content can be the specific time point of the event, the geographical location, the relevant people, the event description information, etc.

[0105] Element priority refers to the order of importance of each preset event element set for an event, which is used to clarify the relative priority of each event element in the event.

[0106] For example, after extracting the elements such as the time, location, characters, and description of the key sub-events, the content of these elements is sorted according to the element priorities of each preset event element to generate ordered element content. This sorted element content constitutes the event element information of the key sub-events, which is organized in the priority order of time, location, characters, and description. 105. Based on the event element information of each key sub-event, generate the event context information of the target event.

[0107] Among them, the event context information can completely depict the whole process of the target event from occurrence to end, show the cause, development context, and final result of the event, so as to truly and comprehensively reflect the evolution track of the target event.

[0108] In some embodiments, in order to ensure the accuracy of the event element information of the key sub-events, before generating the event context information of the target event based on the event element information of each key sub-event, the method further includes: Through the event knowledge graph, perform an evaluation process on the event element information of the key sub-events in at least one dimension to obtain evaluation information; When the evaluation information indicates that the evaluation of the event element information of the key sub-event fails, based on the evaluation information, the event background knowledge information of the key sub-event, and the event basic description information of the key sub-event, perform a correction process on the event element information of the key sub-event to generate new event element information.

[0109] Among them, the evaluation information is the result obtained by evaluating the event element information of the key sub-event in one dimension.

[0110] The new event element information is the information obtained by adjusting and optimizing the event element information of the key sub-event based on the modification suggestions provided by the evaluation information, combined with the event background knowledge information and the event basic description information of the key sub-event.

[0111] In some embodiments, in order to comprehensively evaluate the event element information of the key sub-events from multiple aspects, through the event knowledge graph, perform an evaluation process on the event element information of the key sub-events in at least one dimension to obtain evaluation information, including: Through the event knowledge graph, perform an evaluation process on the element accuracy of the event element information of the key sub-event to obtain accuracy evaluation information; Through the event knowledge graph, perform a description quality evaluation process on the event element information of the key sub-event to obtain quality evaluation information; Through the event knowledge graph, evaluate the importance of the key sub-event to the target event.

[0112] Among them, the accuracy evaluation information refers to the result obtained by evaluating the event element information of key sub-events through the event knowledge graph, which reflects the correctness and reliability of the event element information.

[0113] The quality evaluation information refers to evaluating the integrity and accuracy of the information expression by analyzing the event element information of key sub-events through the event knowledge graph. Such information reflects the detail level of the event element information and the clarity and coherence of the expression.

[0114] The importance refers to the result obtained by evaluating the role played by key sub-events in the process of promoting the target event through the event knowledge graph, reflecting the relative importance of key sub-events in the development of the target event.

[0115] In some embodiments, in order to optimize the key sub-events required to generate the event context information of the target event, so that the event context information is more concise and clear, based on the event element information of each key sub-event, the event context information of the target event is generated, including: Based on the event knowledge graph and the event basic description information of the target event, evaluate the logical relationship between each key sub-event of the target event to obtain event evaluation result information; According to the event evaluation result information, select target key sub-events from the key sub-events; Generate the event context information of the target event based on the event element information of the target key sub-events.

[0116] Among them, the logical relationship refers to the mutual connection and mutual influence among different key sub-events in the development process of the target event, how they affect, depend on or occur in sequence with each other, so as to jointly promote the evolution of the event.

[0117] The event evaluation result information is the conclusion obtained by analyzing the logical relationship between each key sub-event in the target event, which reflects the sub-events that play a key role in promoting the development of the target event.

[0118] The target key sub-events are the sub-events that are screened out from all key sub-events and have a key promoting effect on the development of the target event through the event evaluation result information.

[0119] In some embodiments, in order to quickly screen out the key sub-events required to generate the event context information of the target event, according to the event evaluation result information, select target key sub-events from each key sub-event, including: When the event evaluation result information indicates that there are key sub-events to be removed among the key sub-events, optimize each key sub-event based on the event evaluation result information, the basic event description information of the target event and the event knowledge graph to select the target key sub-events.

[0120] Among them, the key sub-events to be removed refer to the sub-events that are determined to contribute less to the development of the target event in the event evaluation result information. Such sub-events have limited effect on the overall progress during the evolution of the target event. Therefore, based on the event evaluation result information, the basic event description information of the target event, and the analysis of the event knowledge graph, the key sub-events to be removed will be excluded from the construction of the event context information of the target event to ensure the simplicity and core relevance of the event context information.

[0121] In some embodiments, to achieve dynamic management of the event graph and prevent it from continuously expanding, the method further includes: Calculating the event heat information of each event; Deleting the event knowledge graph corresponding to each event in the event graph according to the event heat information. Among them, the event heat information refers to the degree of attention received by the corresponding event within a preset time period. The preset time period is the time range for measuring event heat and is used to evaluate the influence or attention degree of the event within this time period. For example, the event heat information may include at least one of the query frequency, discussion degree, attention degree, query time of the event, event occurrence time, etc. of the event.

[0122] In some embodiments, the event graph is migrated from the cache (the first storage space) to the low-speed storage (the second storage space), where the events in the first storage space have a faster access speed, while the events in the second storage space have a relatively slower access speed. For example, each event in the event graph contains event heat information, and based on this information, hierarchical management of the event index is performed: the events with a higher "heat" are stored in the first storage space, while the events with a lower "heat" are stored in the regular storage. When the capacity of the cache reaches the upper limit, the system will eliminate the event with the lowest "heat". For newly occurred events, a relatively high initial "heat" value can be set for them to ensure that they can enter the cache preferentially, thereby improving the access efficiency.

[0123] In some embodiments, to calculate the event heat, calculating the event heat information of each event includes: Determining the first heat sub-heat according to the query frequency of the event; Determining the second heat sub-heat according to the query time information; Determining the third heat sub-heat according to the event occurrence time information; Determining the event heat information according to the first heat sub-heat, the second heat sub-heat, and the third heat sub-heat.

[0124] Among them, the query frequency of the event refers to the frequency of querying the event within a preset time period, and this frequency reflects the attention degree and dissemination degree of the event. Generally, the higher the query frequency, the more attention the event receives.

[0125] The first sub - heat is used to measure the heat of an event based on the query frequency. That is, an event with a high query frequency will be assigned a higher heat value, thus reflecting its attention intensity within a preset time period.

[0126] The query time information refers to the time distribution of users' queries about an event within a preset time period. Query activities in different time periods can reveal the heat fluctuation of the event, such as peak periods, trough periods, etc.

[0127] The second sub - heat is the heat value determined according to the query time information. That is, the change in the query frequency of an event within a preset time reflects the heat of the event within the preset time period. It may reach a peak in some time periods, thus affecting the overall heat assessment of the event.

[0128] The event occurrence time information refers to the specific time and duration when the event itself occurs, including the start time, end time of the event, and key nodes during the occurrence. These time information help to understand the heat performance of the event.

[0129] The third sub - heat is used to evaluate the heat of an event based on the event occurrence time information. For example, if an event occurs on a holiday, the anniversary of a major historical event, or a special time node, it may trigger more discussions and attention, thus increasing the heat value of the event. In addition, an event that occurs on the same day is usually assigned a higher heat value to reflect its timeliness and immediate attention.

[0130] In some embodiments, methods such as weighted summation, averaging, normalization processing, or standard deviation analysis can be used to comprehensively calculate the first sub - heat, the second sub - heat, and the third sub - heat, so as to obtain the overall heat of the event.

[0131] In some embodiments, the method further includes: Displaying the event context information of the target event on the query page of the target event.

[0132] Wherein, the query page refers to the page used to search for the target event. For example, the query page can be a web page related to the target event, or a discussion page on social media about the target event, etc.

[0133] To meet users' diverse needs for event context information, platforms such as social media applications, search websites, and instant messaging applications can construct event graphs through the technical solution of this application. When a new sub-event is obtained, the event knowledge graph in the existing event graph that matches the new sub-event can be dynamically updated, or if there is no event in the event graph that matches the new sub-event, a corresponding event knowledge graph can be created based on the new sub-event. To help users quickly grasp the evolution process of events, the event context information of each event can also be generated through the event knowledge graphs of the events in the event graph, so as to facilitate querying and understanding the development trajectories and correlation relationships of different events. When an event becomes a hot topic on a social media application or a search website, the social media application and the search website can embed the event context information in the relevant discussion pages or sections to provide users with a clear context of the event development and help them quickly grasp the overall picture of the event. At the same time, the instant messaging application can push the event context information of the event to users in real time to ensure that users understand the key progress and evolution process of the event in a timely manner.

[0134] In addition, enterprises can use the technical solution of this application to construct an event graph related to project events, and the event graph contains the event knowledge graph corresponding to each project event. As the project events progress, the event graph can be dynamically updated to ensure that it always reflects the latest project status. Through the event knowledge graphs in the event graph, the event context information of each event can be generated, so as to provide users with a clear event development trajectory query function when analyzing projects. This enables enterprises to better understand the evolution process of projects, optimize decision-making and management efficiency.

[0135] The technical solution of this application can also be applied to computer operation events. By constructing an event graph related to computer operation events, the event graph contains the event knowledge graph corresponding to each operation event. As the operation events change dynamically, the event graph can be updated in real time to ensure that it accurately reflects the latest operation status. Based on the event knowledge graphs in the event graph, the event context information of each operation event can be generated, so as to provide users with a clear operation event development trajectory query function. This helps to better understand the evolution process of operation events and improve the efficiency of problem troubleshooting and system management.

[0136] In summary, the technical solution of this application can be widely applied to multiple scenarios, helping users promptly grasp event development and improving management, decision-making, and system operation and maintenance efficiency.

[0137] As can be seen from the above, in the embodiments of the present application, since the event knowledge graph corresponding to the target event represents the association relationship between the target event and at least one sub-event, and the association relationship between the sub-event and at least one piece of content, thus, through this event knowledge graph, the event background knowledge information of each sub-event corresponding to the target event can be quickly retrieved, and based on the event background knowledge information of each sub-event and the event basic description information of the target event, the key sub-events that play a promoting role in the development of the target event can be screened out. By further analyzing the event background knowledge information and event basic description information of the key sub-events, the event structure analysis process can be performed on the key sub-events, so as to extract the event element information of the key sub-events. Finally, based on the event element information of each key sub-event, the event context information of the target event can be generated. Since the process of generating this event context information is automated and completely requires no manual intervention, and more reliable background knowledge can be obtained through the event knowledge graph, it not only greatly improves the acquisition efficiency of the event context information, but also avoids the biases and omissions that may occur during the manual collation process. In this way, the generated event context information can be fully guaranteed in terms of accuracy and objectivity, ensuring that the evolution process of the target event can be presented comprehensively and truthfully.

[0138] The method described in the above embodiments will be further described in detail below.

[0139] In this embodiment, taking the generation of event context information through the collaborative cooperation of multiple agents as an example, the method of the embodiments of the present application will be described in detail.

[0140] As Figure 2a shown, the specific process of a method for generating event context information is as follows: 201. Obtain the event knowledge graph corresponding to the target event, where the event knowledge graph represents the association relationship between the target event and at least one sub-event, and the association relationship between the sub-event and at least one piece of content.

[0141] In some embodiments, before obtaining the event knowledge graph corresponding to the target event, it further includes: Obtain multiple initial media contents, identify the event information of the multiple initial media contents, and select the media contents containing event information from the multiple initial media contents according to the identification results; Aggregate the event information of the media contents to obtain at least one sub-event, and establish the association relationship between each sub-event and each media content; Aggregate at least one sub-event to obtain at least one event, and establish the association relationship between each event and the sub-event; Construct an event graph based on the association relationships between each sub-event and each media content, and the association relationships between each event and each sub-event. The event graph includes an event knowledge graph corresponding to each event.

[0142] In some embodiments, as Figure 2b shown, through an event recognition agent, media content with event information can be recognized from the initial multimedia content of the network. The initial multimedia content may include a title, text, pictures, videos, scanned documents, or voice files, etc. By analyzing the description of the initial multimedia content, it is determined whether it contains event information, and an association relationship between the event information and the media content is established. The steps for the event recognition agent to process the initial multimedia content are as follows: 1) Clean and process the initial multimedia content; 2) When the event recognition agent returns result 1, establish an association relationship between the event information and the media content, where 0 indicates no event information (non-event information), and 1 indicates there is event information.

[0143] As Figure 2b shown, then continue through an event extraction agent to further extract event information such as event names and descriptions based on the media content containing event information recognized by the event recognition agent.

[0144] The event extraction agent extracts event information from the initial media content containing event information: 1) Clean the initial multimedia content; 2) Use in-context learning to analyze the cleaned initial media content to obtain the media content containing event information and generate event information such as event names and descriptions.

[0145] In some embodiments, aggregate the event information of the media content to obtain at least one sub-event, including: Compare the event information of each media content to eliminate duplicate event information and obtain sub-events corresponding to the duplicate event information; Mount the media content corresponding to the duplicate event information to the corresponding sub-event.

[0146] In some embodiments, as Figure 2b shown, duplicate event information can be recognized and processed through an event deduplication agent to avoid multiple identical event records. By comparing the event information of each media content, merge the duplicate event information to obtain sub-events corresponding to the duplicate event information, and mount the media content corresponding to the duplicate event information under the corresponding sub-events to reduce redundant calculations.

[0147] Steps for the event deduplication agent to aggregate the event information of media content: 1). Input event e i and the event information of event e j ; 2). Calculate the vector representation e i of event e i based on the event information of event e i , and calculate the vector representation e j of event e j based on the event information of event e j ; 3). Calculate the correlation between e i and e j : ; 4). When the event correlation is greater than the preset correlation , perform deduplication aggregation (delete e j , and mount the media content corresponding to e j under the sub - event corresponding to e j ), otherwise do not aggregate. Specifically, when the event correlation is greater than the preset threshold , the event deduplication agent outputs 1, and deduplication aggregation is required. When the event correlation is not greater than the preset threshold , the event deduplication agent outputs 0, and no deduplication aggregation is required.

[0148] In some embodiments, the method further includes: Determine multiple incremental sub - events and the media content corresponding to each incremental sub - event; Calculate the event correlations of multiple incremental sub - events based on the event - based description information of each incremental sub - event; When the event correlation is greater than the preset correlation, calculate the sub - event aggregation score between the incremental sub - event and its related incremental sub - event according to the media content corresponding to the incremental sub - event and the media content corresponding to its related incremental sub - event; When the sub - event aggregation score is greater than the preset aggregation score, perform aggregation processing on the incremental sub - event and its related incremental sub - event to obtain a new event, and update the event graph based on the association relationship between the new event and its related incremental sub - events, and the association relationship between the incremental sub - events related to the new event and the corresponding media content.

[0149] In some embodiments, the method further includes: Obtain incremental sub - events; Based on the event-based description information of the incremental sub-events and the event-based description information of each event in the event graph, calculate the event correlation between the incremental sub-events and each event in the event graph; When the event correlation is greater than the preset correlation, calculate the event aggregation score between the incremental sub-event and the related events based on the media content corresponding to the incremental sub-event and the media content corresponding to each sub-event of the related events; According to the event aggregation score, establish the association relationship between the incremental sub-event and the related events, and update the event graph.

[0150] In some embodiments, as Figure 2b shown, events can be aggregated by an event aggregation agent. Specifically, multiple similar events centered around the same entity or theme and having logical associations are aggregated into a coarser-grained "big event". Establish the parent-child relationship between the "big event" and the events it contains.

[0151] The event aggregation by the event aggregation agent can be to aggregate the incremental sub-events into new events and to aggregate the incremental sub-events into existing events, where the existing events are the events recorded in the event graph.

[0152] (I). The process of aggregating incremental sub-events into new events: 1). Select the incremental sub-events e i and e j from the multiple sub-events obtained by the event deduplication agent, and obtain the event-based description information and corresponding media content of these two incremental sub-events; 2). Calculate the vector representation e i of the incremental sub-event e i according to the event-based description information of the incremental sub-event e i , and calculate the vector representation e j of the incremental sub-event e j according to the event-based description information of the incremental sub-event e j ; 3). Calculate the event correlation between e i and e j ; 4). When the event correlation is the preset correlation, proceed to the next step, otherwise highlight the aggregation process; 6). According to the media content corresponding to the incremental sub-event e i and the media content corresponding to its related incremental sub-event e j , calculate the incremental sub-event e i and the related incremental sub-event e jThe sub-event aggregation score between them. The calculation formula for the sub-event aggregation score is as follows: , is the incremental sub-event e i corresponding media content, is the incremental sub-event e j corresponding media content; 7), when then, is the preset relevance, then the incremental sub-event e i and its related incremental sub-event e j are aggregated to form a new event , determine the association relationship between the new event and the incremental sub-event e i , and the association relationship between the new event and the related incremental sub-event e j .

[0153] (II) The process of aggregating incremental sub-events into existing events: 1) Select the incremental sub-event e i from the multiple sub-events obtained by the event deduplication agent, and the event in the event graph; 2) Calculate the vector representation e i of the incremental sub-event e i according to the event basic description information of the incremental sub-event e i , and calculate the vector representation of the event according to the event basic description information of the event ; 3) Calculate the event correlation between e i and ; 4) When the event correlation proceeds to the next step, otherwise exit the aggregation process; 5) Calculate the event aggregation score between the incremental sub-event e i and the related event . The calculation formula for the event aggregation score is as follows: _score(e i , ); 6) When , then aggregate the incremental sub-event e i to the related event , and establish the association relationship between the incremental sub-event and the related event, otherwise exit the aggregation process.

[0154] In some embodiments, an event graph is constructed based on the association relationships between each sub-event and each media content, and the association relationships between each event and each sub-event, including: An initial event graph is established based on the association relationships between sub-events and content, and the association relationships between events and sub-events. The initial event graph includes an initial event knowledge graph corresponding to each event. The initial event knowledge graph corresponding to each event includes a parent event node corresponding to the event, sub-event nodes corresponding to each sub-event of the event, content nodes of the media content corresponding to each sub-event, and edges between the nodes. The edges represent the association relationships between events and sub-events, or the association relationships between sub-events and media content; Representation learning is performed on each node and edge in the initial event graph to obtain the event graph.

[0155] In some embodiments, the graph understanding agent can perform representation learning on each node and edge in the initial event graph. The representation learning includes extracting node keywords and calculating node embedding vectors. Specifically, through natural language understanding algorithms (Natural Language Understanding, NLU) and graph neural network algorithms (Graph Neural Networks, GNN), etc., according to the node types in the event graph (such as parent event nodes, sub-event nodes, and content nodes), the edges representing the association relationships between events and sub-events, and the edges representing the association relationships between sub-events and media content, the node keywords and node embedding vectors are calculated.

[0156] In some embodiments, the method further includes: Calculating the event heat information of each event; Deleting the event knowledge graphs corresponding to each event in the event graph according to the event heat information.

[0157] For example, in order to improve resource utilization efficiency and query performance, the least recently used (LRU) mechanism can be considered to update the index. First, analyze the historical query logs, and count the query frequencies and the most recent query times of different events. Based on these data, a "heat" attribute that comprehensively considers the query frequency and time factors is set for each event. Next, the event index can be hierarchically managed, and the events with "high heat" are placed in the cache, while the remaining events are stored in the regular storage. An LRU cache eviction strategy can be designed. When the cache reaches the capacity limit, the event with the lowest "heat" is evicted. For newly occurred events, their initial "heat" can be set to a higher value to ensure that the newly occurred events can enter the cache. Through such index management, it can be ensured that the event graph will not keep expanding, and only the "high heat" and "high value" events need to be maintained.

[0158] 202. Retrieve the event background knowledge information of each sub - event corresponding to the target event through the event knowledge graph.

[0159] In some embodiments, the event knowledge graph includes multiple nodes and edges between the nodes. The nodes include the parent - event nodes corresponding to the target event, the sub - event nodes corresponding to the sub - events, and the content nodes corresponding to the content. The edges represent the association relationship between the target event and the sub - event, or the association relationship between the sub - event and the content. Retrieving the event background knowledge information of each sub - event corresponding to the target event through the event knowledge graph includes: Search for the sub - event nodes corresponding to each sub - event of the target event in the event knowledge graph. Determine the local sub - graphs corresponding to each sub - event node in the event knowledge graph. The local sub - graph of a sub - event node includes the sub - event node, the content nodes connected to the sub - event node, and the edges connecting the sub - event node and the content nodes. Determine the event background knowledge information of each sub - event node according to the association relationship between the sub - event and the content in the local sub - graph of each sub - event node.

[0160] In some embodiments, a query agent can be used to provide query functions for the nodes, relationships, and related information of the event graph. Users can retrieve events or detailed content (local sub - graph recall) related to the input information in the event graph based on conditions such as "Identity Document of the event", sub - event ID, keywords, etc., for detailed query or subsequent generation tasks. During the query process, based on the input information, through id hard - matching / keyword matching / vector matching, anchor nodes related to the input can be queried, and at the same time, based on this anchor node, depth - first or breadth - first search can be performed to obtain an N - hop local sub - graph.

[0161] Through the collaboration of the event recognition agent, event extraction agent, event re - ranking agent, event aggregation agent, and graph understanding agent, the construction of the event graph can be completed efficiently and flexibly. The event graph can support the query agent to perform explicit (such as: ID query, keyword query, depth search, etc.) or implicit (vector matching) retrieval of nodes and association relationships, and has high flexibility and reliability.

[0162] 203. Based on the event background knowledge information of each sub - event and the event basic description information of the target event, determine at least one key sub - event of the target event from each sub - event.

[0163] In some embodiments, the key sub-events can be identified from the sub-events of the target event through the context node recognition agent. Specifically, the query agent uses the event knowledge graph as a retriever (GraphRAG) to retrieve the sub-events of the target event and the corresponding media content of these sub-events, and uses them as the event background knowledge information of each sub-event for the large language model to generate. Using in-context learning, the event background knowledge information of each sub-event corresponding to the target event obtained by the retriever is input into the large language model for the identification process of key sub-nodes, and at least one key sub-node of the target event is returned. The key sub-nodes constitute the main line of the development of the target event. Compared with obtaining knowledge from a vast amount of text using a search engine based on keywords, the event background knowledge information obtained based on the event knowledge graph is more accurate and reliable.

[0164] , where represents the input node recognition task template of the large language model, that is, the key sub-event recognition model in the foregoing embodiments, which has three inputs: the target event 's event basic description information, the sub-event e of the target event i 's event basic description information and the event background knowledge information of the sub-event e i ( ).

[0165] Where Figure 2c the node to be confirmed in the event graph is the node corresponding to the new sub-event of the target event, and the new sub-event corresponding to this node has not passed through the context node recognition agent. The non-node is the node corresponding to the non-key sub-event of the target event, and the node is the node corresponding to the key event of the target event. Among them, the key sub-event to be confirmed can be understood as the new sub-event of the target event, which has not been analyzed by the context node recognition agent.

[0166] 204. According to the event background knowledge information and the event basic description information of the key sub-event, perform event structure analysis processing on the key sub-event to obtain the event element information of the key sub-event.

[0167] In some embodiments, as Figure 2c shown, through the context node understanding agent, using the retriever and in-context learning, and using the reasoning ability of the large language model, extract the event element information (such as the time, place, person, event description, summary, etc.) of the key sub-event from the input event background knowledge information of each key sub-event.

[0168] , where Represents the input node understanding task template of the large language model, that is, the event element recognition model in the foregoing embodiments, with 2 inputs: the input key sub-event e i 's event basic description information and the key sub-event e i 's event background knowledge information (knowledge enhanced by image retrieval ), returns event element information, and the event element information of the key sub-event e i 's event element information can specifically be as Figure 2d shown.

[0169] 205. Through the event knowledge graph, perform at least one-dimensional evaluation processing on the event element information of the key sub-event to obtain evaluation information; when the evaluation information indicates that the evaluation of the event element information of the key sub-event fails, based on the evaluation information, the event background knowledge information of the key sub-event, and the event basic description information of the key sub-event, perform correction processing on the event element information of the key sub-event to generate new event element information.

[0170] In some embodiments, as Figure 2c and Figure 2d shown, through the context node evaluation model, based on GraphRAG, evaluate the event element information of the key sub-event. The main evaluation dimensions are: node necessity (importance for the large event), node accuracy (accuracy of elements such as time, place, and person), and description accuracy (whether the event description quality of the node is accurate and does not violate facts).

[0171] This step is a key step in the system's self-optimization. Generating the key sub-node is not the end. The evaluation of the event element information is the basis and foundation for self-correction and optimization. This step still needs to rely on GraphRAG to retrieve all information related to this key sub-node as background knowledge and input it into the large language model to judge the accuracy and relevant basis of the event element information of the generated key sub-node.

[0172] , where represents the input node information evaluation task model of the large language model, with 3 inputs: the event basic description information of the key sub-event e i , the event element information of the key sub-event , and the knowledge enhanced by image retrieval , and returns whether the event element information of the key sub-node is accurate and a comment.

[0173] In some embodiments, as Figure 2cAs shown, by rewriting the agent through the context nodes, when the evaluation information indicates that the evaluation of the event element information of the key sub-event fails, it provides guidance on the rewriting direction for the rewriting of the event element information of the key sub-event. This step still needs to rely on GraphRAG to retrieve all the information related to this key sub-event as background knowledge, combine it with the evaluation information, and input it into the large language model to guide the large language model to correct the incorrect information.

[0174] , where represents the error correction task template for the input node information of the large language model, which has 4 inputs: the key sub-event e i , the event element information of the key sub-event obtained in the previous step , the evaluation information of , and the knowledge enhanced by image retrieval .

[0175] 206. Based on the event knowledge graph and the event-based description information of the target event, evaluate the logical relationship between the key sub-events of the target event to obtain event evaluation result information; and select the target key sub-event from the key sub-events according to the event evaluation result information.

[0176] In some embodiments, as Figure 2c and Figure 2e shown, the key sub-event evaluation agent can be used to evaluate each key sub-event of the target event. The evaluation dimensions mainly include: logical relationship problems between key sub-events, whether there are identical nodes, and whether there are mutually inclusive nodes. This step still needs to rely on GraphRAG to retrieve all the information related to this key sub-event as background knowledge and input it into the large language model to determine whether each key sub-event is reasonable.

[0177] , where represents the context list evaluation task model of the large language model. The context list is a list composed of the key sub-events of the target event, which has 2 inputs: the target event , and the knowledge enhanced by image retrieval , and returns the event evaluation result information (that is, whether each key sub-event is reasonable and related comments).

[0178] In some embodiments, as Figure 2cAs shown, the intelligent agent can be optimized through the context. Based on GraphRAG, combined with the event evaluation result information, and used as the input for in-context learning at the same time, each key sub-event of the target event is optimized to solve the logical problems between the key sub-events, and the duplicate key sub-events and non-essential key sub-events included are removed. This step still relies on GraphRAG to retrieve all information related to this key sub-event as background knowledge, combined with the event evaluation result information, and input into the large language model to obtain the final target key sub-events.

[0179] , where represents the refined task template of the context list input to the large language model, with 3 inputs: the target event , the event evaluation result information of the previous step , and the knowledge enhanced by image retrieval , and returns the target key sub-events of the target event.

[0180] 207. Generate the event context information of the target event based on the event element information of the target key sub-events.

[0181] The method for generating event context information proposed in this application effectively solves the following problems: 1. Hierarchical heterogeneous event graph: Based on multi-intelligent agent collaboration (event recognition, event extraction, event rearrangement, event aggregation, graph understanding, query intelligent agent), a heterogeneous event graph with a hierarchical structure is constructed. Compared with a homogeneous planar graph, it can depict more complex hierarchical relationships; at the same time, the application of the event graph is not only used for event context information generation alone. In fact, such a complex structure allows for a deeper insight into the event structure and progress. This hierarchical heterogeneous graph can model the mutual relationships between large events, sub-events, and individual contents.

[0182] 2. Flexible graph retrieval and query: Based on the heterogeneous event graph, id query, keyword query, and vector matching query of nodes in the event graph can be realized (the vector representations of nodes and edges are usually learned using GNN), and complex query requirements can be processed.

[0183] 3. Identification, understanding, and query system for key sub-events of events: A complete fully automatic system is constructed, covering the construction of the event graph and the identification, understanding (basic event description information), and query of key sub-events. The processing efficiency is high, the information is complete and clear, and the application layer can not only be implicitly used for service recommendation, but also be explicitly displayed on the client side for product operation.

[0184] 4. Automatic evaluation and correction optimization of key sub-events of events: Traditional solutions often only cover the process to the generation of context information, and there is no systematic automatic evaluation mechanism for complex evaluations such as whether the generated context information is accurate, whether the information is complete, whether the nodes are repeated, etc., which often rely on manual operations and have high operating costs. This application is based on the automatic evaluation and correction optimization mechanism of key sub-events of GraphRAG, which can automatically self-check the generated event context information, improve the reliability of the generated event context information, and reduce the cost of manual operations.

[0185] As can be seen from the above, the beneficial effects of this application include: 1. Reduce storage and computing costs.

[0186] By building an event graph with multiple agents, we aggregate tens of millions of articles into thousands of event granularities, greatly compressing redundant information and significantly reducing the amount of data that needs to be stored and processed. The index scale at the event level is usually much smaller than the scale of the original news, which will significantly reduce the storage and computing overhead of index construction and retrieval. Especially in the scenario of massive news data, the advantages of reducing costs and increasing efficiency will be more obvious.

[0187] 2. Improve retrieval efficiency and quality.

[0188] Based on the event graph, this application can use a combination of vector search and keyword inverted search to efficiently search for relevant knowledge. At the same time, the advantage of the graph is that the graph provides topological relationship information between nodes. In the graph, in addition to being able to retrieve relevant content and knowledge based on relevance, more effective and high-quality information can be recalled through the graph search algorithm. High-quality graphs provide GraphRAG with high-quality knowledge.

[0189] 3. Improve user experience.

[0190] The present invention utilizes a large language model to generate event context information based on GraphRAG retrieval knowledge enhancement content, and can fully automatically generate the ins and outs of an event and detailed information on each specific node, so that users can quickly and accurately grasp the overall picture of the event without having to read a large amount of redundant information.

[0191] 4. Empower content production and operations.

[0192] Based on GraphRAG, the event context is automatically generated, which can timely discover and generate the latest event context node progress. The latest progress will be pushed to the hot event operation so that the operation can carry out the operation of related events and topics, thereby improving labor efficiency.

[0193] To better implement the above method, an embodiment of the present application further provides a device for generating event context information. The device for generating event context information may be specifically integrated in an electronic device, which may be a device such as a terminal or a server. Among them, the terminal may be a device such as a mobile phone, a tablet computer, a smart Bluetooth device, a laptop computer, or a personal computer; the server may be a single server or a server cluster composed of multiple servers.

[0194] For example, in this embodiment, taking the device for generating event context information specifically integrated in an electronic device as an example, the method of the embodiment of the present application will be described in detail.

[0195] For example, as Figure 3 shown, the device for generating event context information may include a graph acquisition unit 301, an information retrieval unit 302, an event determination unit 303, an event analysis unit 304, and a context generation unit 305, as follows: (1). Graph acquisition unit 301.

[0196] The graph acquisition unit 301 is configured to acquire an event knowledge graph corresponding to a target event, where the event knowledge graph represents the association relationship between the target event and at least one sub-event, and the association relationship between the sub-event and at least one piece of content.

[0197] In some embodiments, the device further includes a content acquisition unit, an information aggregation unit, an event aggregation unit, and a graph construction unit: The content acquisition unit is configured to acquire a plurality of initial media contents, identify event information in the plurality of initial media contents, and select media contents containing event information from the plurality of initial media contents according to the identification results; The information aggregation unit is configured to perform aggregation processing on the event information of the media contents to obtain at least one sub-event, and establish an association relationship between each sub-event and each media content; The event aggregation unit is configured to perform aggregation processing on at least one sub-event to obtain at least one event, and establish an association relationship between each event and the sub-events; The graph construction unit is configured to construct an event graph according to the association relationship between each sub-event and each media content, and the association relationship between each event and the sub-events. The event graph includes event knowledge graphs corresponding to each event.

[0198] In some embodiments, the information aggregation unit is configured to perform information comparison on the event information of each media content to perform duplicate removal processing on the repeatedly occurring event information, and obtain sub-events corresponding to the repeatedly occurring event information; mount the media contents corresponding to the repeatedly occurring event information to the corresponding sub-events.

[0199] In some embodiments, a graph construction unit is configured to establish an initial event graph based on the association relationship between sub-events and content, and the association relationship between events and sub-events. The initial event graph includes initial event knowledge graphs corresponding to each event. The initial event knowledge graph corresponding to each event includes a parent event node corresponding to the event, sub-event nodes corresponding to each sub-event of the event, content nodes of the media content corresponding to each sub-event, and edges between the nodes. The edges represent the association relationship between an event and a sub-event, or the association relationship between a sub-event and media content. Perform representation learning on each node and edge in the initial event graph to obtain an event graph.

[0200] In some embodiments, the apparatus further includes a graph update unit, configured to: Obtain incremental sub-events; Based on the event basic description information of the incremental sub-events and the event basic description information of each event in the event graph, calculate the event correlation between the incremental sub-events and each event in the event graph; When the event correlation is greater than a preset correlation, calculate the event aggregation score between the incremental sub-events and the related events based on the media content corresponding to the incremental sub-events and the media content corresponding to each sub-event of the related events; According to the event aggregation score, establish an association relationship between the incremental sub-events and the related events, and update the event graph.

[0201] In some embodiments, the apparatus further includes a graph update unit, configured to: Determine multiple incremental sub-events and the media content corresponding to each incremental sub-event; Calculate the event correlation of the multiple incremental sub-events based on the event basic description information of each incremental sub-event; When the event correlation is greater than a preset correlation, calculate the sub-event aggregation score between the incremental sub-events and the related incremental sub-events according to the media content corresponding to the incremental sub-events and the media content corresponding to the related incremental sub-events thereof; When the sub-event aggregation score is greater than a preset aggregation score, perform aggregation processing on the incremental sub-events and the related incremental sub-events thereof to obtain a new event, and update the event graph based on the association relationship between the new event and the related incremental sub-events thereof, and the association relationship between the related incremental sub-events of the new event and the corresponding media content.

[0202] In some embodiments, the apparatus further includes a heat calculation unit and a graph management unit, configured to: Calculate the event heat information of each event; Delete the event knowledge graphs corresponding to each event in the event graph according to the event heat information.

[0203] In some embodiments, a heat calculation unit is configured to determine a first sub-heat of an event based on the query frequency of the event; determine a second sub-heat based on the query time information; determine a third sub-heat based on the event occurrence time information; and determine the event heat information based on the first sub-heat, the second sub-heat, and the third sub-heat.

[0204] (2) An information retrieval unit 302.

[0205] The information retrieval unit 302 is configured to retrieve the event background knowledge information of each sub-event corresponding to the target event through an event knowledge graph.

[0206] In some embodiments, the event knowledge graph includes multiple nodes and edges between the nodes. The nodes include a parent event node corresponding to the target event, a sub-event node corresponding to the sub-event, and a content node corresponding to the content. The edges represent the association relationship between the target event and the sub-event, or the association relationship between the sub-event and the content. The information retrieval unit is configured to: Search for the sub-event nodes corresponding to each sub-event of the target event in the event knowledge graph; Respectively determine the local sub-graphs corresponding to the sub-event nodes in the event knowledge graph. The local sub-graph of a sub-event node includes the sub-event node, the content nodes connected to the sub-event node, and the edges connecting the sub-event node and the content nodes; Determine the event background knowledge information of each sub-event node according to the association relationship between the sub-event and the content in the local sub-graph of each sub-event node.

[0207] (3) An event determination unit 303.

[0208] The event determination unit 303 is configured to determine at least one key sub-event of the target event from each sub-event based on the event background knowledge information of each sub-event and the event basic description information of the target event.

[0209] In some embodiments, the event determination unit is configured to: Through a key sub-event recognition model, respectively predict the probability information that each sub-event belongs to the key sub-event of the target event based on the event background knowledge information of each sub-event, the event basic description information of each sub-event, and the event basic description information of the target event; Determine at least one key sub-event of the target event according to the probability information.

[0210] (4) An event analysis unit 304.

[0211] An event analysis unit 304, configured to perform event structure analysis processing on a key sub - event according to the event background knowledge information and the event basic description information of the key sub - event, so as to obtain the event element information of the key sub - event.

[0212] In some embodiments, the apparatus further includes an element evaluation unit and an element correction unit: The element evaluation unit is configured to perform evaluation processing on the event element information of the key sub - event in at least one dimension through an event knowledge graph to obtain evaluation information; The element correction unit is configured to, when the evaluation information indicates that the evaluation of the event element information of the key sub - event fails, perform correction processing on the event element information of the key sub - event based on the evaluation information, the event background knowledge information of the key sub - event, and the event basic description information of the key sub - event, so as to generate new event element information.

[0213] In some embodiments, the element evaluation unit is configured to: Perform evaluation processing on the accuracy of the event element information of the key sub - event through an event knowledge graph to obtain accuracy evaluation information; Perform description quality evaluation processing on the event element information of the key sub - event through an event knowledge graph to obtain quality evaluation information; Evaluate the importance of the key sub - event to the target event through an event knowledge graph.

[0214] In some embodiments, the event analysis unit is configured to perform identification processing on at least one preset event element of the background knowledge information of the key sub - event and the event basic description information of the key sub - event through an event element identification model to obtain the element content of the key sub - event under each preset event element; based on the element priority of each preset event element, perform content sorting processing on the element content of the key sub - event under each preset event element to obtain the event element information of the key sub - event.

[0215] (5). A context generation unit 305.

[0216] The context generation unit 305 is configured to generate event context information of the target event based on the event element information of each key sub - event.

[0217] In some embodiments, the context generation unit is configured to: Evaluate the logical relationship between each key sub - event of the target event based on the event knowledge graph and the event basic description information of the target event to obtain event evaluation result information; Select target key sub - events from the key sub - events according to the event evaluation result information; Generate the event context information of the target event based on the event element information of the target key sub - events.

[0218] In some embodiments, the context generation unit is configured to, when the event evaluation result information indicates that there are key sub - events to be removed among the key sub - events, optimize each key sub - event based on the event evaluation result information, the basic event description information of the target event, and the event knowledge graph, so as to select the target key sub - events.

[0219] In specific implementation, each of the above units can be implemented as an independent entity, or can be combined arbitrarily and implemented as the same or several entities. For the specific implementation of each of the above units, reference can be made to the foregoing method embodiments, which will not be elaborated herein.

[0220] As can be seen from the above, the device for generating the event context information of this embodiment obtains the event knowledge graph corresponding to the target event through the graph acquisition unit. The event knowledge graph represents the association relationship between the target event and at least one sub - event, as well as the association relationship between the sub - event and at least one piece of content. The information retrieval unit retrieves the event background knowledge information of each sub - event corresponding to the target event through the event knowledge graph. The event determination unit determines at least one key sub - event of the target event from each sub - event based on the event background knowledge information of each sub - event and the basic event description information of the target event. The event analysis unit performs event structure analysis and processing on the key sub - events according to the event background knowledge information and the basic event description information of the key sub - events to obtain the event element information of the key sub - events. The context generation unit is configured to generate the event context information of the target event based on the event element information of each key sub - event.

[0221] Therefore, the embodiment of the present application can not only improve the acquisition efficiency of the event context information, but also avoid the deviation and omission that may occur in the manual sorting process, which is beneficial to ensuring the high reliability of the event context information in terms of accuracy and objectivity, and can truly and comprehensively present the evolution process of the target event.

[0222] The embodiment of the present application also provides an electronic device, which can be a device such as a terminal, a server, etc. Among them, the terminal can be a mobile phone, a tablet computer, a smart Bluetooth device, a notebook computer, a personal computer, etc.; the server can be a single server or a server cluster composed of multiple servers, etc.

[0223] In some embodiments, the device for generating the event context information can also be integrated in multiple electronic devices. For example, the device for generating the event context information can be integrated in multiple servers, and the method for generating the event context information of the present application is implemented by multiple servers.

[0224] In this embodiment, the electronic device in this embodiment is taken as an example of a server for detailed description. For example, as Figure 4 shown, it shows a schematic structural diagram of the server involved in the embodiments of the present application. Specifically: The server may include a processor 401 with one or more processing cores, a memory 402 of one or more computer-readable storage media, a power supply 403, an input module 404, a communication module 405 and other components. Those skilled in the art can understand that Figure 4 the server structure shown in does not constitute a limitation on the server, and may include more or fewer components than shown, or combine certain components, or arrange different components. Among them: The processor 401 is the control center of the server, connecting various parts of the entire server through various interfaces and lines, and executing various functions of the server and processing data by running or executing software programs and / or modules stored in the memory 402, and calling data stored in the memory 402. In some embodiments, the processor 401 may include one or more processing cores; in some embodiments, the processor 401 may integrate an application processor and a modem processor. Among them, the application processor mainly processes the operating system, user interface, application programs, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor may not be integrated into the processor 401.

[0225] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area. Among them, the program storage area may store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area may store data created according to the use of the server. In addition, the memory 402 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices. Correspondingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0226] The server also includes a power supply 403 for supplying power to each component. In some embodiments, the power supply 403 may be logically connected to the processor 401 through a power management system, so as to implement functions such as management of charging, discharging, and power consumption management through the power management system. The power supply 403 may also include any components such as one or more DC or AC power supplies, a recharge system, a power failure detection circuit, a power converter or inverter, and a power status indicator.

[0227] The server may further include an input module 404, which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function controls.

[0228] The server may further include a communication module 405. In some embodiments, the communication module 405 may include a wireless module. The server can perform short-range wireless transmission through the wireless module of the communication module 405, thereby providing users with wireless broadband Internet access. For example, the communication module 405 can be used to help users send and receive emails, browse web pages, and access streaming media, etc.

[0229] Although not shown, the server may further include a display unit, etc., which will not be elaborated here. Specifically, in this embodiment, the processor 401 in the server will load the executable files corresponding to the processes of one or more application programs into the memory 402 according to the following instructions, and the processor 401 will run the application programs stored in the memory 402, so as to implement the steps in the methods of the embodiments of the present application.

[0230] For the specific implementation of the above operations, reference can be made to the previous embodiments, which will not be elaborated here.

[0231] As can be seen from the above, since the event context information generation process is automated and completely requires no manual intervention, it not only greatly improves the acquisition efficiency of event context information, but also avoids the biases and omissions that may occur during the manual collation process. In this way, the generated event context information can be fully guaranteed in terms of accuracy and objectivity, ensuring that it can comprehensively and truthfully present the evolution process of the target event.

[0232] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by controlling relevant hardware through instructions. The instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0233] For this purpose, an embodiment of the present application provides a computer-readable storage medium, in which multiple instructions are stored, and the instructions can be loaded by a processor to execute the steps in any one of the event context information generation methods provided by the embodiments of the present application. For example, the instructions can execute the following steps: Obtain an event knowledge graph corresponding to the target event, where the event knowledge graph represents the association relationship between the target event and at least one sub-event, as well as the association relationship between the sub-event and at least one piece of content; Retrieve the event background knowledge information of each sub-event corresponding to the target event through the event knowledge graph; Based on the event background knowledge information of each sub - event and the event basic description information of the target event, at least one key sub - event of the target event is determined from each sub - event; According to the event background knowledge information of the key sub - event and the event basic description information of the key sub - event, event structure analysis and processing are performed on the key sub - event to obtain the event element information of the key sub - event; Based on the event element information of each key sub - event, the event context information of the target event is generated.

[0234] Among them, the storage medium may include: read - only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, etc.

[0235] According to one aspect of the present application, a computer program product or a computer program is provided. The computer program product or the computer program includes computer programs / instructions, and the computer programs / instructions are stored in a computer - readable storage medium. The processor of the electronic device reads the computer programs / instructions from the computer - readable storage medium, and the processor executes the computer programs / instructions, so that the electronic device executes the methods provided in various alternative implementations of the generation of the event context information in the above - mentioned embodiments.

[0236] Since the instructions stored in the storage medium can execute the steps in any of the event context information generation methods provided in the embodiments of the present application, the beneficial effects that can be achieved by any of the event context information generation methods provided in the embodiments of the present application can be realized. For details, see the previous embodiments and will not be elaborated here.

[0237] The above has introduced in detail a method for generating event context information and related devices provided in the embodiments of the present application. Specific examples are used in this article to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for generating event context information, characterized in that: include: Acquire an event knowledge graph corresponding to a target event, wherein the event knowledge graph represents an association relationship between the target event and at least one sub-event, and an association relationship between the sub-event and at least one content; Retrieving event background knowledge information of each sub-event corresponding to the target event through the event knowledge graph; Based on the event background knowledge information of each sub-event and the event basic description information of the target event, determining at least one key sub-event of the target event from the sub-events; According to the event background knowledge information of the key sub-event and the event basic description information of the key sub-event, the event structure analysis processing is performed on the key sub-event to obtain the event element information of the key sub-event; Based on the event element information of each key sub-event, the event context information of the target event is generated.

2. The method for generating event context information according to claim 1, characterized in that: The event knowledge graph includes a plurality of nodes and edges between the nodes, wherein the nodes include a parent event node corresponding to the target event, a sub-event node corresponding to the sub-event, and a content node corresponding to the content, and the edges represent the association relationship between the target event and the sub-event, or the association relationship between the sub-event and the content; The retrieving event background knowledge information of each sub-event corresponding to the target event through the event knowledge graph includes: Searching the event knowledge graph for sub-event nodes corresponding to each sub-event corresponding to the target event; Determine the local subgraph corresponding to each sub-event node in the event knowledge graph, wherein the local subgraph of the sub-event node includes the sub-event node, a content node connected to the sub-event node, and an edge connecting the sub-event node and the content node; The event background knowledge information of each sub-event node is determined according to the association relationship between the sub-event and the content in the local sub-graph of each sub-event node.

3. The method for generating event context information according to claim 1, characterized in that: Before generating the event context information of the target event based on the event element information of each key sub-event, the method further includes: Through the event knowledge graph, the event element information of the key sub-event is evaluated in at least one dimension to obtain evaluation information; When the evaluation information indicates that the event element information of the key sub-event fails the evaluation, the event element information of the key sub-event is corrected based on the evaluation information, the event background knowledge information of the key sub-event, and the event basic description information of the key sub-event to generate new event element information.

4. The method for generating event context information according to claim 3, wherein the event knowledge graph is used to evaluate the event element information of the key sub-event in at least one dimension to obtain evaluation information, including: Through the event knowledge graph, the event element information of the key sub-event is evaluated for element accuracy to obtain accuracy evaluation information; Through the event knowledge graph, the event element information of the key sub-event is subjected to description quality assessment processing to obtain quality assessment information; The importance of the key sub-events to the target event is evaluated through the event knowledge graph.

5. The method for generating event context information according to claim 1, characterized in that: The step of generating event context information of the target event based on event element information of each key sub-event includes: Based on the event knowledge graph and the event basic description information of the target event, the logical relationship between the key sub-events of the target event is evaluated to obtain event evaluation result information; Selecting a target key sub-event from the key sub-events according to the event evaluation result information; The event context information of the target event is generated based on the event element information of the target key sub-event.

6. The method for generating event context information according to claim 5, characterized in that: The step of selecting a target key sub-event from each key sub-event according to the event evaluation result information includes: When the event evaluation result information indicates that there are key sub-events to be removed in the key sub-events, the key sub-events are optimized based on the event evaluation result information, the basic event description information of the target event and the event knowledge graph to select the target key sub-event.

7. The method for generating event context information according to claim 1, characterized in that: The step of determining at least one key sub-event of the target event from the sub-events based on the event background knowledge information of the sub-events and the event basic description information of the target event includes: Using a key sub-event identification model, based on event background knowledge information of each sub-event, event basic description information of each sub-event, and event basic description information of the target event, respectively predicting probability information of each sub-event belonging to a key sub-event of the target event; At least one key sub-event of the target event is determined according to the probability information.

8. The method for generating event context information according to claim 1, characterized in that: Before obtaining the event knowledge graph corresponding to the target event, the method further includes: Acquire a plurality of initial media contents, and identify event information of the plurality of initial media contents, so as to select media contents containing event information from the plurality of initial media contents according to the identification result; Aggregating the event information of the media content to obtain at least one sub-event, and establishing an association relationship between each sub-event and each media content; Aggregating the at least one sub-event to obtain at least one event, and establishing an association relationship between each event and the sub-event; According to the association relationship between each sub-event and each media content, and the association relationship between each event and the sub-event, an event graph is constructed, and the event graph includes an event knowledge graph corresponding to each event.

9. The method for generating event context information according to claim 8, characterized in that: The aggregating the event information of the media content to obtain at least one sub-event includes: Comparing the event information of each media content to eliminate duplicates of the repeated event information and obtain sub-events corresponding to the repeated event information; The media content corresponding to the repeatedly occurring event information is mounted to the corresponding sub-event.

10. The method for generating event context information according to claim 8, wherein the event graph is constructed based on the association relationship between each sub-event and each media content, and the association relationship between each event and the sub-event, characterized in that: include: Based on the association relationship between sub-events and content, and the association relationship between events and sub-events, an initial event graph is established, wherein the initial event graph includes an initial event knowledge graph corresponding to each event, and the initial event knowledge graph corresponding to each event includes a parent event node of the corresponding event, sub-event nodes corresponding to each sub-event of the event, content nodes of media content corresponding to each sub-event, and edges between nodes, wherein the edges represent the association relationship between events and sub-events, or the association relationship between sub-events and media content; Representation learning is performed on each node and edge in the initial event graph to obtain an event graph.

11. The method for generating event context information according to claim 8, characterized in that: The method further comprises: Get incremental sub-events; Based on the event basic description information of the incremental sub-event and the event basic description information of each event in the event graph, calculating the event correlation between the incremental sub-event and each event in the event graph; When the event correlation is greater than a preset correlation, based on the media content corresponding to the incremental sub-event and the media content corresponding to each sub-event of the related event, calculating an event aggregation score between the incremental sub-event and the related event; According to the event aggregation score, an association relationship between the incremental sub-event and the related event is established, and the event graph is updated.

12. The method for generating event context information according to claim 8, characterized in that: The method further comprises: Determine a plurality of incremental sub-events and media content corresponding to each incremental sub-event; Calculating the event correlation of the plurality of incremental sub-events based on the event basic description information of each incremental sub-event; When the event correlation is greater than a preset correlation, calculating a sub-event aggregation score between the incremental sub-event and the related incremental sub-event according to the media content corresponding to the incremental sub-event and the media content corresponding to its related incremental sub-event; When the sub-event aggregation score is greater than the preset aggregation score, the incremental sub-event and its related incremental sub-events are aggregated to obtain a new event, and based on the association relationship between the new event and its related incremental sub-events, and the association relationship between the incremental sub-events related to the new event and the corresponding media content, the event graph is updated.

13. The method for generating event context information according to claim 8, characterized in that: The method further comprises: Calculate the event heat information of each event; According to the event heat information, the event knowledge graph corresponding to each event in the event graph is deleted.

14. The method for generating event context information according to claim 13, characterized in that: The calculating of the event heat information of each event includes: Determine a first heat sub-heat according to the query frequency of the event; Determine the second heat sub-heat according to the query time information; Determine the third heat sub-heat according to the event occurrence time information; Event heat information is determined according to the first heat sub-heat, the second heat sub-heat, and the third heat sub-heat.

15. The method for generating event context information according to claim 1, characterized in that: The step of performing event structure analysis on the key sub-event according to the event background knowledge information of the key sub-event and the event basic description information of the key sub-event to obtain event element information of the key sub-event includes: Using an event element recognition model, the background knowledge information of the key sub-event and the event basic description information of the key sub-event are processed to recognize at least one preset event element, so as to obtain the element content of the key sub-event under each preset event element; Based on the element priorities of the preset event elements, the element contents of the key sub-events under the preset event elements are sorted to obtain event element information of the key sub-events.

16. A device for generating event context information, characterized in that: include: A graph acquisition unit, configured to acquire an event knowledge graph corresponding to a target event, wherein the event knowledge graph represents an association relationship between the target event and at least one sub-event, and an association relationship between the sub-event and at least one content; An information retrieval unit, used to retrieve event background knowledge information of each sub-event corresponding to the target event through the event knowledge graph; An event determination unit, configured to determine at least one key sub-event of the target event from the sub-events based on the event background knowledge information of the sub-events and the event basic description information of the target event; An event analysis unit, configured to perform event structure analysis on the key sub-event according to the event background knowledge information of the key sub-event and the event basic description information of the key sub-event, so as to obtain event element information of the key sub-event; The context generating unit is used to generate event context information of the target event based on the event element information of each key sub-event.

17. An electronic device, characterized in that: It includes a processor and a memory, wherein the memory stores a plurality of instructions; the processor loads instructions from the memory to execute the steps in the method for generating event context information as described in any one of claims 1 to 15.

18. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the steps in the method for generating event context information as described in any one of claims 1 to 15.

19. A computer program product comprising a plurality of instructions, characterized in that: When the instructions are executed by the processor, the steps in the method for generating event context information described in any one of claims 1 to 15 are implemented.

Images