Content processing method and related equipment

Abstract

This application discloses a content processing method and related equipment. The relevant embodiments can be applied to various scenarios such as cloud technology, artificial intelligence, and smart transportation; the target search content and heterogeneous search graph to be searched for the target object can be obtained; feature information transmission processing is performed on the associated objects and historical search results according to the heterogeneous search graph to obtain the search interest feature information of the target object and the interactive feature information of the historical search results; the target search results for the target search content are obtained, and the correlation strength of the target search results is calculated based on the interactive feature information; the target search results are sorted according to the target search content, correlation strength, and search interest feature information, and the sorted search results are output. This application can sort the search results by deeply mining the heterogeneous search graph of the target object, which is conducive to improving the accuracy of the search result sorting, thereby facilitating the target object to quickly find the required results and improving the search efficiency.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a content processing method and related equipment. Background Art

[0002] With the development of internet technology, online information is growing rapidly, and the web is flooded with redundant information. To find the information they need, users rely on search engines. A search engine is a software system used on the internet that uses specific strategies to collect and discover information online, processes the information, and then provides users with internet information search services. Search engines typically provide a web interface where users can submit search terms on the client. The search application then retrieves search results that match the user's search terms, ranks the results, and returns them to the user.

[0003] However, in the current related technologies, search applications arbitrarily sort search results. In the sorting of search results, usually only the relevance of the search results to the search content input by the user is considered. This method of sorting based only on similarity has the problem of low recall result sorting quality. It is easy for the search results at the top of the search result list to not be the results that the user is satisfied with, resulting in a low click-through rate for the search results at the top of the search result list. Summary of the Invention

[0004] An embodiment of the present application provides a content processing method and related equipment. The related equipment may include a content processing device, an electronic device, a computer-readable storage medium, and a computer program product, which can improve the accuracy of search result sorting.

[0005] The present invention provides a content processing method, including:

[0006] Obtaining a target search content currently to be searched for a target object and a heterogeneous search graph corresponding to the target object, wherein the heterogeneous search graph represents an interactive relationship between an associated object and historical search results, wherein the historical search results are search results of the associated object for the historical search content;

[0007] According to the heterogeneous search graph, feature information transfer processing is performed on the associated objects and the historical search results to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object;

[0008] Obtaining at least one target search result for the target search content, and calculating, based on the interaction feature information, the association strength of the historical search results of the target object with each target search result;

[0009] The target search results are sorted according to the target search content, the association strength, and the search interest feature information, and the sorted search results are output.

[0010] Accordingly, an embodiment of the present application provides a content processing device, including:

[0011] an acquisition unit, configured to acquire a target search content currently to be searched for a target object, and a heterogeneous search graph corresponding to the target object, wherein the heterogeneous search graph represents an interactive relationship between an associated object and historical search results, wherein the historical search results are search results of the associated object for the historical search content;

[0012] An information transfer unit is configured to perform feature information transfer processing on the associated objects and the historical search results according to the heterogeneous search graph to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object;

[0013] a calculation unit, configured to obtain at least one target search result for the target search content, and calculate, based on the interaction feature information, an association strength of the historical search results of the target object with respect to each target search result;

[0014] The sorting unit is used to sort the target search results according to the target search content, the association strength and the search interest feature information, and output the sorted search results.

[0015] Optionally, in some embodiments of the present application, the heterogeneous search graph includes at least one node and edges between nodes, the node including an object type node corresponding to the associated object and a result type node corresponding to the historical search result, and the edge represents an interactive relationship between the associated object and the historical search result;

[0016] The information transmission unit can be specifically used to perform feature information transmission processing on the edges between nodes in the heterogeneous search graph to perform feature information transmission processing on the associated objects and the historical search results, so as to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object.

[0017] Optionally, in some embodiments of the present application, the content processing apparatus may further include a search graph construction unit, the search graph construction unit being configured to construct a heterogeneous search graph for the target object; the search graph construction unit may include an acquisition subunit, a construction subunit, a selection subunit, and a deletion subunit, as follows:

[0018] The acquisition subunit is used to acquire historical search results of the reference object for historical search content;

[0019] A construction subunit, configured to construct an initial heterogeneous search graph for the target object based on an interactive relationship between the reference object and the historical search results corresponding to the reference object;

[0020] a selection subunit, configured to select, from the reference object, an associated object corresponding to the target object based on a similarity between the reference object and the target object;

[0021] The deletion subunit is used to perform node deletion processing on the initial heterogeneous search graph according to the associated object to obtain a heterogeneous search graph corresponding to the target object.

[0022] Optionally, in some embodiments of the present application, the selection sub-unit can be specifically used to determine, for each reference object, based on the initial heterogeneous search graph, the number of paths that meet preset meta-path conditions between the object type node corresponding to the reference object and the object type node corresponding to the target object; based on the number of paths, determine the similarity between the reference object and the target object; and based on the similarity, select the associated object corresponding to the target object from the reference object.

[0023] Optionally, in some embodiments of the present application, the deletion sub-unit can be specifically used to perform a node search in the initial heterogeneous search graph according to the associated object to determine the node to be deleted from the initial heterogeneous search graph, the node to be deleted does not belong to the object type node of the associated object, and the result type node corresponding to the historical search result of the associated object; the node to be deleted in the initial heterogeneous search graph is deleted to obtain the heterogeneous search graph corresponding to the target object.

[0024] Optionally, in some embodiments of the present application, the information delivery unit may include a first extraction subunit, a search subunit, a delivery subunit, and a first determination subunit, as follows:

[0025] The first extraction subunit is used to perform feature extraction on each node in the heterogeneous search graph to obtain node feature information corresponding to each node, wherein the node feature information includes node feature information corresponding to each object type node and node feature information corresponding to each result type node;

[0026] A search subunit, configured to perform a node search in the heterogeneous search graph based on each node in the heterogeneous search graph to determine a neighbor node corresponding to the node;

[0027] a transmission subunit, configured to perform feature information transmission processing on the node feature information of the node according to the node feature information of the neighboring nodes, to obtain the target node feature information of the node;

[0028] The first determining subunit is configured to determine the search interest feature information of the target object and the interaction feature information of the historical search results corresponding to the target object based on the target node feature information of the object type node corresponding to the target object.

[0029] Optionally, in some embodiments of the present application, the transmission sub-unit can be specifically used to perform attention processing on the node feature information of the neighbor node based on the edge type feature information corresponding to the edge between the node and its corresponding neighbor node, and the node feature information of the node, to obtain the attention weight corresponding to the neighbor node; based on the attention weight and the node feature information of the neighbor node, the node feature information of the node is updated to obtain the target node feature information of the node.

[0030] Optionally, in some embodiments of the present application, the step of “updating the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes to obtain the target node feature information of the node” may include:

[0031] Updating the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes;

[0032] Return to the step of performing attention processing on the node feature information of the neighboring node based on the edge type feature information corresponding to the edge between the node and its corresponding neighboring node, and the node feature information of the node, until the target node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

[0033] Optionally, in some embodiments of the present application, the calculation unit may include a second determination subunit and a first calculation subunit, as follows:

[0034] The second determining subunit is configured to determine time weight information corresponding to the historical search results of the target object based on the search time corresponding to the historical search results of the target object and the current time;

[0035] The first calculation subunit is configured to calculate, for each target search result, based on the interaction feature information and the time weight information, the association strength of the historical search results of the target object to the target search result.

[0036] Optionally, in some embodiments of the present application, the first calculation subunit can be specifically used to determine, for each target search result, a basic association strength of the target search result based on the similarity between the search interest feature information of the target object and the target search result; determine the content relevance of the historical search results of the target object to the target search result based on the correlation between the interaction feature information and the target search result, and the correlation between the interaction feature information and the target search content; and calculate the association strength of the historical search results of the target object to the target search result based on the basic association strength, the content relevance and the time weight information.

[0037] Optionally, in some embodiments of the present application, the sorting unit may include a second extraction subunit, a second calculation subunit, and a sorting subunit, as follows:

[0038] The second extraction subunit is used to extract attention features from the target search content and the historical search results of the target object to obtain the current search intention features corresponding to the target object;

[0039] The second calculation subunit is used to calculate the similarity between each target search result and the current search intention feature to obtain the intention relevance corresponding to each target search result;

[0040] The sorting subunit is used to sort the target search results based on the intention relevance, the association strength and the search interest feature information, and output the sorted search results.

[0041] An electronic device provided in an embodiment of the present application includes a processor and a memory, wherein the memory stores a plurality of instructions, and the processor loads the instructions to execute the steps in the content processing method provided in the embodiment of the present application.

[0042] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the content processing method provided in the embodiment of the present application.

[0043] In addition, an embodiment of the present application also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements the steps in the content processing method provided in the embodiment of the present application.

[0044] The embodiment of the present application provides a content processing method and related equipment, which can obtain the target search content to be searched by the target object currently, and the heterogeneous search graph corresponding to the target object, wherein the heterogeneous search graph represents the interactive relationship between the associated object and the historical search results, and the historical search results are the search results of the associated object for the historical search content; according to the heterogeneous search graph, the associated object and the historical search results are processed with feature information transmission to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object; obtain at least one target search result for the target search content, and calculate the association strength of the historical search results of the target object for each target search result based on the interactive feature information; sort the target search results according to the target search content, the association strength and the search interest feature information, and output the sorted search results. The present application can sort the search results by deeply mining the heterogeneous search graph of the target object, which is conducive to improving the accuracy of the search result sorting, thereby facilitating the target object to quickly find the required results and improving the search efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0046] Figure 1a This is a scenario diagram of the content processing method provided in an embodiment of the present application;

[0047] Figure 1b is a flowchart of the content processing method provided in an embodiment of the present application;

[0048] Figure 1c is an illustration of a content processing method provided in an embodiment of the present application;

[0049] Figure 1d This is a model structure diagram of the content processing method provided in the embodiment of the present application;

[0050] Figure 2 is another flow chart of the content processing method provided by an embodiment of the present application;

[0051] Figure 3 is a structural diagram of a content processing device provided in an embodiment of the present application;

[0052] Figure 4 It is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0054] The present invention provides a content processing method and related equipment, which may include a content processing device, an electronic device, a computer-readable storage medium, and a computer program product. The content processing device may be integrated into an electronic device, which may be a terminal or a server.

[0055] It is understandable that the content processing method of this embodiment can be executed on a terminal, on a server, or jointly by a terminal and a server. The above examples should not be construed as limiting the present application.

[0056] like Figure 1a As shown, a content processing method is performed jointly by a terminal and a server as an example. The content processing system provided in the embodiment of the present application includes a terminal 10 and a server 11, etc. The terminal 10 and the server 11 are connected via a network, such as a wired or wireless network connection, wherein the content processing device can be integrated into the server.

[0057] Among them, the server 11 can be used to: obtain the target search content currently to be searched for the target object, and the heterogeneous search graph corresponding to the target object, the heterogeneous search graph represents the interactive relationship between the associated object and the historical search results, and the historical search results are the search results of the associated object for the historical search content; according to the heterogeneous search graph, perform feature information transmission processing on the associated object and the historical search results to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object; obtain at least one target search result for the target search content, and calculate the association strength of the historical search results of the target object for each target search result based on the interactive feature information; sort the target search results according to the target search content, the association strength and the search interest feature information, and output the sorted search results to the terminal 10. Among them, the server 11 can be a single server, or a server cluster or cloud server composed of multiple servers. The content processing method or device disclosed in this application, wherein multiple servers can be composed of a blockchain, and the server is a node on the blockchain.

[0058] The terminal 10 can be used to receive the sorted search results sent by the server 11 and display the sorted search results to the target user on the corresponding search results page. The terminal 10 can include a mobile phone, a smart TV, a tablet computer, a laptop computer, or a personal computer (PC). The terminal 10 can also be configured with a client, which can be an application client or a browser client.

[0059] The step of the server 11 acquiring the sorted search results may also be performed by the terminal 10 .

[0060] The content processing method provided in the embodiments of the present application relates to computer vision technology and natural language processing in the field of artificial intelligence.

[0061] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive field of computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making. AI technology is an interdisciplinary discipline encompassing a wide range of fields, encompassing both hardware and software technologies. AI software technologies primarily include computer vision, speech processing, natural language processing, machine learning / deep learning, autonomous driving, and smart transportation.

[0062] Computer vision (CV) is the science of making machines "see." Specifically, it refers to machine vision, which uses cameras and computers to replace the human eye to identify and measure targets, and then further processes images to make them more suitable for human observation or transmission to instruments for detection. As a scientific discipline, computer vision studies related theories and technologies, attempting to build artificial intelligence systems that can obtain information from images or multidimensional data. Computer vision technology typically includes image processing, image recognition, image semantic understanding, image retrieval, optical character recognition (OCR), video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous positioning and mapping, autonomous driving, and smart transportation. It also includes common biometric recognition technologies such as facial recognition and fingerprint recognition.

[0063] Natural language processing (NLP) is a key area of ​​research in computer science and artificial intelligence. It studies the theories and methods that enable effective communication between humans and computers using natural language. Natural language processing (NLP) integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language we use in everyday life—and is closely linked to the study of linguistics. Natural language processing technologies typically include text processing, semantic understanding, machine translation, robotic question answering, and knowledge graphs.

[0064] It should be noted that the order of description of the following embodiments is not intended to limit the preferred order of the embodiments.

[0065] This embodiment will be described from the perspective of a content processing device. The content processing device may be integrated into an electronic device, which may be a server, a terminal or other device.

[0066] It is understandable that in the specific implementation of this application, user information, such as the user's historical search content and other related data, when the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0067] The content processing method of the embodiment of the present application can be applied to various content search scenarios. This embodiment can be applied to various scenarios such as cloud technology, artificial intelligence, smart transportation, and assisted driving.

[0068] like Figure 1b As shown, the specific process of the content processing method can be as follows:

[0069] 101. Obtain target search content currently to be searched for a target object and a heterogeneous search graph corresponding to the target object, wherein the heterogeneous search graph represents an interactive relationship between an associated object and historical search results, wherein the historical search results are search results of the associated object for historical search content.

[0070] Among them, the target object is the object currently performing content query. Specifically, the target object can perform content query through the target application. The target application can be a content search platform. The target object can realize information search through the search portal and built-in search engine provided by the target application. For example, the target application can be a browser.

[0071] The target search content specifically refers to the content currently to be searched, that is, the search content queried by the current target object; in a specific search scenario, the target search content may be the content entered by the current target object in the search content input box of the target application.

[0072] In this embodiment, the associated object may be an object that is related to the target object in terms of content interaction. It should be noted that the associated object may also include the target object itself. The historical search results of the associated object may be the search results obtained by the associated object when querying historical search content within a historical time period, specifically, the search results clicked by the associated object. The search results may include information in various modalities, such as video, audio, text, and images. Similarly, the search content may also include information in various modalities, such as video, audio, text, and images. This embodiment does not limit this.

[0073] Optionally, in this embodiment, the heterogeneous search graph includes at least one node and edges between nodes, the nodes include object type nodes corresponding to the associated objects and result type nodes corresponding to the historical search results, and the edges represent the interactive relationship between the associated objects and the historical search results.

[0074] Here, an edge is a connection between two nodes. Specifically, an edge can be used to connect an object type node and a result type node. That is, the node types at both ends of the connection are different.

[0075] In a specific embodiment, the heterogeneous search graph may also include query type nodes corresponding to the historical search content of the associated object. According to the node types connected at both ends of the edge, the edges in the heterogeneous search graph can be divided into three types, namely: edges connecting object type nodes and result type nodes, edges connecting object type nodes and query type nodes, and edges connecting query type nodes and result type nodes.

[0076] Optionally, this embodiment can generate object type nodes corresponding to each associated object, query type nodes corresponding to each historical search content, and result type nodes corresponding to each historical search result based on each associated object, the historical search content of each associated object, and the historical search results for the historical search content. Then, based on the interactive relationship between the associated objects corresponding to each object type node, the historical search content corresponding to each query type node, and the historical search results corresponding to each result type node, each node can be connected to obtain a heterogeneous search graph. For example, when a historical search result is a search result clicked by a certain associated object within a historical time period, the historical search result has an interactive relationship with the associated object, and the object type node of the associated object and the result type node corresponding to the historical search result can be connected. For another example, when a historical search content is the query content of a certain associated object within a historical time period, the historical search content has an interactive relationship with the associated object, and the object type node of the associated object and the query type node corresponding to the historical search content can be connected. For another example, if a historical search result is a search result corresponding to a historical search content, the historical search result and the historical search content have an interactive relationship, and the result type node corresponding to the historical search result and the query content node corresponding to the historical search content can be connected.

[0077] In a specific scenario, refer to Figure 1c , shows the heterogeneous search graph of the above embodiment, which includes the interactive relationships between the associated objects, historical search content, and historical search results. The associated objects here can include the target object itself and the similar objects corresponding to the target object. Among them, u2 represents the target object, u1 and u3 represent similar objects, q2 is the historical search content of the target object u2, q1 is the historical search content of the similar object u1, q3 is the historical search content of the similar object u3, the historical search results of the target object u2 for the historical search content q2 can include d2 and d3, the historical search results of the similar object u1 for the historical search content q1 can include d1 and d2, and the historical search results of the similar object u3 for the historical search content q3 can include d3 and d4. Figure 1c The heterogeneous search graph in [1] includes object type nodes, query type nodes, and result type nodes. Edges between object type nodes and query type nodes indicate that the object corresponding to the object type node has retrieved the historical search content corresponding to the query type node. Edges between object type nodes and result type nodes indicate that the object corresponding to the object type node has clicked on the historical search results corresponding to the result type node. Edges between query type nodes and result type nodes indicate that the historical search results corresponding to the result type node match the historical search content corresponding to the query type node. Specifically, the historical search results can be documents, denoted by D.

[0078] Optionally, in this embodiment, the content processing method may further include:

[0079] Obtain historical search results of the reference object for historical search content;

[0080] constructing an initial heterogeneous search graph for the target object based on an interactive relationship between the reference object and the historical search results corresponding to the reference object;

[0081] Based on the similarity between the reference object and the target object, selecting an associated object corresponding to the target object from the reference object;

[0082] Node deletion processing is performed on the initial heterogeneous search graph according to the associated object to obtain a heterogeneous search graph corresponding to the target object.

[0083] Specifically, the reference object may be an object that has clicked on the same search result as the target object. The similarity between the reference object and the target object may be determined based on the historical search content and clicked historical search results of the reference object and the similarity between the historical search content and clicked historical search results of the target object.

[0084] For example, the target object can be recorded as user1, and the historical search results clicked by the target object include item1. In addition to user1, there is also an object user2 that has an interactive relationship with item1, that is, user2 also clicked item1, so user2 can be regarded as a reference object of the target object user1.

[0085] This embodiment can create a heterogeneous search graph based on the target object's search log and filter out reference objects with similar search behavior to the target object and add them to the heterogeneous search graph to supplement the target object's search history. Specifically, considering that other objects that have clicked on the same document as the current target object may have certain similarities, adding all of these objects to the heterogeneous search graph may result in the graph being too large and difficult to calculate, this embodiment can further filter these reference objects.

[0086] Optionally, in this embodiment, the step of “selecting, from the reference object, an associated object corresponding to the target object based on the similarity between the reference object and the target object” may include:

[0087] For each reference object, determining, based on the initial heterogeneous search graph, the number of paths between the object type node corresponding to the reference object and the object type node corresponding to the target object that meet a preset meta-path condition;

[0088] determining a similarity between the reference object and the target object based on the number of paths;

[0089] Based on the similarity, an associated object corresponding to the target object is selected from the reference objects.

[0090] Among them, the meta-path is a path defined in the heterogeneous graph G, which is in the form of A1, A2, ... A k , where A1, A2, ...A k ∈A. The meta-path represents an A1A k A composite relationship between nodes, a node sequence that meets the definition of a meta-path is called an instance of the meta-path.

[0091] Among them, the heterogeneous graph can be a given directed graph G = {V, E, A, R}, where V is the set of nodes, E is the set of edges, A is the set of node types, and R is the set of edge types. There exists a node relationship mapping function τ(v): V→A and an edge relationship mapping function where v∈V and e∈E. Specifically, a heterogeneous graph G={V,E,A,R} is a heterogeneous search graph if and only if it satisfies the following conditions: A={U,Q,D} and E={I,C,R}, where U,Q,D represent object-type nodes, query-type nodes, and result-type nodes, respectively. I,C,R represent the edge types: object-type node-query-type node, object-type node-result-type node, and query-type node-result-type node, respectively.

[0092] Among them, the preset meta-path condition can be set according to actual conditions. For example, the preset meta-path condition can be the meta-path of object-search result-object (UDU). Then the number of paths that meet the preset meta-path condition between the object type node corresponding to the reference object and the object type node corresponding to the target object can be: the number of times the reference object and the target object have clicked on the same search result.

[0093] Optionally, in this embodiment, the step of “deleting nodes from the initial heterogeneous search graph according to the associated objects to obtain a heterogeneous search graph corresponding to the target object” may include:

[0094] Performing a node search in the initial heterogeneous search graph according to the associated object to determine a to-be-deleted node from the initial heterogeneous search graph, wherein the to-be-deleted node does not belong to an object type node of the associated object and a result type node corresponding to a historical search result of the associated object;

[0095] Deleting the nodes to be deleted in the initial heterogeneous search graph to obtain a heterogeneous search graph corresponding to the target object.

[0096] The node to be deleted may be an object type node of a non-associated object, or a result type node corresponding to a historical search result clicked by a non-associated object, or a query type node corresponding to a historical search content of a non-associated object.

[0097] In a specific embodiment, the specific steps of constructing a heterogeneous search graph for a target object are as follows:

[0098] Get the historical search set H of target object i i , where the historical search set can include the historical search content of the target object i and the historical search results that have been clicked. Then, based on H i Build the graph as follows: add the object type node of the target object to the graph, and add the historical search content of the target object and the search results clicked in each query as query type nodes and result type nodes to the graph respectively, and add edges between the object and historical search content, the object and historical search results, and the historical search content and historical search results.

[0099] Since objects with similar search behaviors play an important role in inferring the search interests of the target object, this embodiment can add i Reference objects with similar search behaviors and their behaviors are used as supplements. Considering that other objects that have clicked on the same document as the current target object may have certain similarities, but adding all of these objects to the heterogeneous search graph may cause the graph to be too large and difficult to calculate, this embodiment can screen these reference objects again. Specifically, the similarity between the reference object and the target object based on PathSim can be calculated, and related objects can be screened from the reference objects based on the similarity.

[0100] Specifically, this embodiment can first filter out all i Reference objects that have clicked on the same search result (such as documents) i ={u1, u2, ..., u m}, and add the target object, all reference objects and their historical search sets into the graph to obtain the initial heterogeneous search graph. Then, for U i Each reference object u in k ∈U i , calculate u i with u k PathSim similarity s based on meta-path object-document-object (UDU) k , as shown in formula (1):

[0101]

[0102] Among them, s k Can represent the reference object uk and target object u i The similarity between them, p ik :p ik ∈P is the number of connected nodes u i and u k , and satisfies the meta-path instance of meta-path P, specifically, p ik :p ik ∈P represents the reference object u k and target object u i The number of documents that have clicked on the same document, that is, the number of paths between the object type node corresponding to the reference object and the object type node corresponding to the target object that meet the preset meta-path conditions. ii :p ii ∈P represents the slave node u i The meta-path instance that starts and connects to itself and satisfies the meta-path P, p kk :p kk ∈P represents the slave node u k Departures a meta-path instance connected to itself that satisfies meta-path P.

[0103] In the calculation of the reference object u k and target object u i After determining the similarity between the two objects, the associated objects can be screened from the reference objects based on the similarity. Specifically, the reference objects can be sorted from largest to smallest according to the similarity, and the first n reference objects can be determined as associated objects. Alternatively, the reference objects with a similarity greater than a preset similarity can be determined as associated objects. The preset similarity can be set according to actual conditions.

[0104] 102. Perform feature information transfer processing on the associated objects and the historical search results according to the heterogeneous search graph to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object.

[0105] Optionally, in this embodiment, the heterogeneous search graph includes at least one node and edges between nodes, the node including an object type node corresponding to the associated object and a result type node corresponding to the historical search result, and the edge represents an interactive relationship between the associated object and the historical search result;

[0106] The step of “performing feature information transfer processing on the associated objects and the historical search results according to the heterogeneous search graph to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object” may include:

[0107] In the heterogeneous search graph, feature information is transferred between the edges of the nodes to transfer the feature information of the associated objects and the historical search results, thereby obtaining the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object.

[0108] Among them, the heterogeneous search graph can specifically be a heterogeneous graph neural network. This embodiment can use a heterogeneous graph neural network to learn the representation of each node based on the constructed heterogeneous search graph; specifically, the heterogeneous graph neural network can be used as the main architecture to integrate the interactive relationship between objects and search results into the feature embedding process.

[0109] The heterogeneous graph neural networks used here can include HAN (Heterogeneous Graph Attention Network, heterogeneous graph neural network based on attention mechanism), MAGNN (Meta-path Aggregation Graph Neural Network for Heterogeneous Graph Embedding), HGT (Heterogeneous Graph Transformer, heterogeneous graph transformation architecture), GTN (Graph Transformer Network, graph transformation network), and SimpleHGN (Simple Heterogeneous graph neural network) based on graph attention network and its improved version on heterogeneous graphs.

[0110] Optionally, in this embodiment, the step of “performing feature information transfer processing on edges between nodes in the heterogeneous search graph to perform feature information transfer processing on the associated objects and the historical search results to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object” may include:

[0111] Performing feature extraction on each node in the heterogeneous search graph to obtain node feature information corresponding to each node, wherein the node feature information includes node feature information corresponding to each object type node and node feature information corresponding to each result type node;

[0112] For each node in the heterogeneous search graph, performing a node search in the heterogeneous search graph according to the node to determine a neighbor node corresponding to the node;

[0113] performing feature information transfer processing on the node feature information of the node according to the node feature information of the neighboring node to obtain the target node feature information of the node;

[0114] Based on the target node feature information of the object type node corresponding to the target object, the search interest feature information of the target object and the interaction feature information of the historical search results corresponding to the target object are determined.

[0115] Among them, feature extraction is performed on each node, which may specifically include feature extraction on object type nodes and feature extraction on result type nodes; wherein, feature extraction is performed on object type nodes, which may specifically include feature extraction on associated objects corresponding to object type nodes; feature extraction is performed on result type nodes, which may specifically include feature extraction on historical search results corresponding to result type nodes.

[0116] Among them, in some embodiments, the heterogeneous search graph can also include query type nodes corresponding to the historical search content of the associated object. In the heterogeneous search graph, by performing feature information transmission processing on the edges between nodes, feature information transmission processing can be performed on the associated object, historical search results, and historical search content.

[0117] Specifically, in some embodiments, the initial node feature information of each node in the heterogeneous search graph can be generated by a semantic model. For query type nodes, if the historical search content is text, the average word vector of all words in the historical search content corresponding to the query type node can be used as the initial node feature information of the query type node; for result type nodes, if the historical search results are text, the average word vector of all words in the historical search result text corresponding to the result type node can be taken as the initial node feature information of the result type node; for object type nodes, the vector average of all queried historical search contents and clicked historical search results of the corresponding object is taken as the initial node feature information of the object type node. Among them, the semantic model can be a Word2vec (word to vector) word vector model.

[0118] The step of “extracting features from each node in the heterogeneous search graph to obtain node feature information corresponding to each node” may include:

[0119] For each node in the heterogeneous search graph, feature extraction is performed on the historical search content corresponding to the query type node to obtain node feature information corresponding to the query type node;

[0120] Extracting features of the historical search results corresponding to the result type node to obtain node feature information corresponding to the result type node;

[0121] Feature extraction is performed on historical search contents and historical search results of associated objects corresponding to the object type node to obtain node feature information corresponding to the object type node.

[0122] In some embodiments, the node feature information corresponding to the object type node may be obtained by extracting the object attribute information of the object corresponding to the object type node, wherein the object attribute information may include attribute sub-information on at least one dimension, such as attribute sub-information such as the object's own attributes, short-term features, and long-term features, and this embodiment does not impose any restrictions on this. Among them, the short-term features may specifically be features corresponding to historical search content and historical search results that are relatively close to the current time, such as short-term features may be obtained by extracting features from historical search content and historical search results within the past week; long-term features may include features corresponding to historical search content and historical search results that are far from the current time, such as long-term features may be obtained by extracting features from historical search content and historical search results within six months.

[0123] In this embodiment, for each node in the heterogeneous search graph, the heterogeneous search graph is searched for neighbor nodes corresponding to the node. Specifically, the nodes in the heterogeneous search graph that are directly connected to the node are determined as neighbor nodes.

[0124] Optionally, in this embodiment, the step of “performing feature information transfer processing on the node feature information of the node according to the node feature information of the neighboring node to obtain the target node feature information of the node” may include:

[0125] Performing attention processing on the node feature information of the neighbor node according to the edge type feature information corresponding to the edge between the node and its corresponding neighbor node, and the node feature information of the node, to obtain the attention weight corresponding to the neighbor node;

[0126] Based on the attention weight and the node feature information of the neighboring nodes, the node feature information of the node is updated to obtain the target node feature information of the node.

[0127] This embodiment adds edge type feature information when calculating attention weights. This allows for consideration of edge types and better modeling of heterogeneous graphs. Specifically, edge types may include edges connecting object-type nodes and result-type nodes, edges connecting object-type nodes and query-type nodes, and edges connecting query-type nodes and result-type nodes. Different edge types may correspond to different edge type feature information, which may specifically be a parameter vector obtained through training.

[0128] The step of “performing attention processing on the node feature information of the neighbor node according to the edge type feature information corresponding to the edge between the node and its corresponding neighbor node, and the node feature information of the node, to obtain the attention weight corresponding to the neighbor node” may include:

[0129] Concatenate edge type feature information corresponding to the edge between the node and its corresponding neighbor node, and the node feature information of the node, with the node feature information of the neighbor node to obtain processed feature information corresponding to the neighbor node;

[0130] According to the processed feature information corresponding to the neighbor node, attention processing is performed on the node feature information of the neighbor node to obtain the attention weight corresponding to the neighbor node.

[0131] Specifically, the process of paying attention to the node feature information of neighboring nodes can be expressed by the following equations (2), (3) and (4):

[0132]

[0133] z i =Wh i (3)

[0134]

[0135] Among them, a ij is the attention weight corresponding to the edge between node i and node j, which can also be regarded as the attention weight for node i and node j as its neighbor node. Neighbor(i) represents the neighbor node of node i, LeakyRelu is the activation function, and a T is a trainable parameter vector. e τ(i，k) Represents the edge type feature information corresponding to the edge between node i and node k, W r It is for e τ(i，k) The parameter matrix for linear transformation, h i It is the previous layer vector representation of node i, specifically the current node feature information of node i.

[0136] After calculating the attention weight corresponding to the edge, the node feature information of the node can be updated according to the attention weight and the node feature information of the neighboring nodes. The update is shown in formula (5):

[0137]

[0138] Where W is the linear transformation matrix of node features, j∈Neighbor(i) represents the neighbor node j of node i, h j Represents the current node feature information of neighbor node j, represents the updated node feature information of node i, and σ is a nonlinear activation function. Since residual connections can avoid the difficulty of training due to excessively deep layers, this embodiment can also add residual connections to the node feature information and edge attention weights, as shown in equations (6) and (7), respectively:

[0139]

[0140]

[0141] in, is the attention weight corresponding to the edge between node i and node j, is the updated attention weight corresponding to the edge between node i and node j, and θ is a trainable parameter.

[0142] In some embodiments, since the multi-head attention mechanism can increase the stability of training, the multi-head attention mechanism can be used in the calculation of node feature information. The specific calculation is shown in formula (8):

[0143]

[0144] Here, || represents concatenation, and K = 8 is the number of heads in the multi-head attention. Equation (8) represents concatenating the vectors output by each head to obtain the updated node feature information. It should be noted that to ensure dimensionality consistency, the last layer needs to average the vectors output by each head. In addition, for simplicity, the above formula omits some superscripts indicating the number of layers.

[0145] Here, l and l+1 represent the number of layers in the heterogeneous neural network, which can also be considered as the number of updates to the node feature information of the nodes in the heterogeneous search graph. Specifically, the number of layers in the heterogeneous neural network can be set to 2. In this way, only two rounds of node feature information updates are required to obtain the target node feature information that meets the preset information transfer conditions. Ultimately, a matrix consisting of the target node feature information of all nodes in the network output is obtained.

[0146] Optionally, in this embodiment, the step of “updating the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes to obtain the target node feature information of the node” may include:

[0147] Updating the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes;

[0148] Return to the step of performing attention processing on the node feature information of the neighboring node based on the edge type feature information corresponding to the edge between the node and its corresponding neighboring node, and the node feature information of the node, until the target node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

[0149] The step of "updating the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes" may include: fusing the attention weight corresponding to each neighboring node and the node feature information of each neighboring node to obtain fused feature information, and updating the node feature information of the node based on the fused feature information. Specifically, the fused feature information may be determined as the updated node feature information corresponding to the node. There are various fusion methods, which are not limited in this embodiment. For example, the fusion method may be weighted fusion.

[0150] Among them, this embodiment can perform a new round of attention processing based on the updated node feature information until the node feature information corresponding to the node that meets the preset information transmission conditions is obtained, and based on the node feature information corresponding to the node that meets the preset information transmission conditions, determine the target node feature information corresponding to the node.

[0151] The preset information transmission condition can be set according to actual conditions and is not limited in this embodiment. For example, the preset information transmission condition can be that the number of attention processing does not exceed a preset number. For example, if the preset number is 2, only two rounds of attention processing are performed.

[0152] Optionally, in this embodiment, the step of “determining search interest feature information of the target object and interaction feature information of historical search results corresponding to the target object based on target node feature information of the object type node corresponding to the target object” may include:

[0153] Determining search interest feature information of the target object based on target node feature information of the object type node corresponding to the target object;

[0154] For each historical search result of the target object, the interaction feature information corresponding to the historical search result is determined based on the target node feature information of the result type node corresponding to the historical search result, the target node feature information of the object type node corresponding to the target object, and the edge type feature information corresponding to the edge between the object type node and the result type node.

[0155] Among them, this embodiment can take the final vector of the node corresponding to the target object as the user's search interest feature information, specifically, the target node feature information of the object type node corresponding to the target object is determined as the search interest feature information of the target object, which can be recorded as the long-term object interest representation

[0156] In some embodiments, the heterogeneous search graph includes query type nodes. For each historical search content of the target object, the interaction feature information corresponding to the historical search content can be determined based on the target node feature information of the query type node corresponding to the historical search content, the target node feature information of the object type node corresponding to the target object, and the edge type feature information corresponding to the edge between the object type node and the query type node.

[0157] In a specific embodiment, feature information can be used to represent the search behavior of the target object. For the interactive feature information corresponding to the operation of the target object retrieving a certain content (i.e., historical search content) in a historical time period, that is, the interactive feature information corresponding to the historical search content, it can be obtained by adding the target node feature information corresponding to the target object, the target node feature information of the node corresponding to the historical search content, and the object-retrieval content edge type feature information and averaging them. For the interactive feature information corresponding to the operation of the target object clicking on a certain search result (i.e., historical search result) in a historical time period, that is, the interactive feature information corresponding to the historical search result, it can be obtained by adding the target node feature information corresponding to the target object, the target node feature information of the node corresponding to the historical search result, and the object-search result edge type feature information and averaging them. In this way, the historical search set corresponding to the target object can be obtained. The interactive feature information of the target object i is shown in FIG1 , wherein the historical search set may include the historical search content of the target object i and the historical search results that have been clicked.

[0158] 103. Obtain at least one target search result for the target search content, and calculate, based on the interaction feature information, the association strength of the historical search results of the target object with each target search result.

[0159] The target search results may be search results retrieved for the target search content currently being retrieved. In a specific scenario, each target search result may be a candidate document in a search result list retrieved for the target search content.

[0160] Optionally, in this embodiment, the step of “calculating, based on the interaction feature information, the association strength of the historical search results of the target object with each target search result” may include:

[0161] Determining time weight information corresponding to the historical search results of the target object based on the search time corresponding to the historical search results of the target object and the current time;

[0162] For each target search result, the association strength of the historical search results of the target object to the target search result is calculated based on the interaction feature information and the time weight information.

[0163] The time weight information corresponding to the historical search results can be used as an influencing factor for the current search scenario. The relevance strength of the historical search results to the target search results can also be seen as the motivation strength of the historical search results to the target search results.

[0164] Specifically, the time weight information corresponding to the historical search results can be determined based on the time difference between the search time corresponding to the historical search results and the current time; the larger the time difference, the smaller the time weight information, and vice versa, the smaller the time difference, the larger the time weight information.

[0165] In a specific embodiment, the target object u can be obtained i Corresponding historical search collection Historical Search Collection It includes at least one historical interaction content, where the historical interaction content may refer to the historical search content retrieved by the target object, or the historical search results clicked; this embodiment may determine the time weight information corresponding to each historical interaction content based on the interaction time (specifically, the retrieval time or the search time) and the current time corresponding to the interaction feature information of each historical interaction content; and then, for each target search result, calculate the association strength of the target search result based on the interaction feature information and the time weight information corresponding to each historical interaction content.

[0166] Specifically, this embodiment may use Hawkes Process to model the impact of the target object's historical behavior on the current search intent.

[0167] The Hawkes process is a self-excitation point process among the time series point processes. Its conditional intensity function is defined as shown in formula (9):

[0168]

[0169] Among them, λ(e) represents the conditional strength of historical events on current events, μ is the basic strength, and e h is the historical event that occurred before the current time t, t his the historical time point when the historical event occurred, and κ is a kernel function that decays over time. In a specific search scenario, historical events can be understood as the historical search content retrieved by the user within the historical time period and the historical search results clicked.

[0170] Optionally, in this embodiment, the step of “calculating, for each target search result, based on the interaction feature information and the time weight information, the association strength of the historical search results of the target object with the target search result” may include:

[0171] For each target search result, determining a basic association strength of the target search result based on the similarity between the search interest feature information of the target object and the target search result;

[0172] Determining content relevance between the historical search results of the target object and the target search results based on the relevance between the interaction feature information and the target search results, and the relevance between the interaction feature information and the target search content;

[0173] Based on the basic association strength, the content relevance, and the time weight information, the association strength of the historical search results of the target object to the target search results is calculated.

[0174] The calculation of the similarity between the search interest feature information of the target object and the target search result may adopt cosine similarity or the like, which is not limited in this embodiment.

[0175] The correlation between the interaction feature information and the target search results, as well as the correlation between the interaction feature information and the target search content, can also be determined using cosine similarity. Specifically, after calculating the correlation between the interaction feature information and the target search results, as well as the correlation between the interaction feature information and the target search content, these two correlations can be fused to obtain the content correlation between the historical search results of the target object and the target search results. The fusion processing method can specifically be multiplication or weighted operation.

[0176] The basic association strength, content relevance, and time weight information can be fused to obtain the association strength of the target search result. Specifically, the content relevance of each historical search result to the target search result and the time weight information corresponding to each historical search result can be weighted and fused to obtain the fused association strength. The fused association strength is then added to the basic association strength to obtain the association strength of the target search result.

[0177] In a specific embodiment, the target object queries the target search content and recalls the search result list for the target search content. For each candidate document d (i.e., the target search result) in the search result list, the calculation process of all historical behaviors of the target object for its incentive intensity can be shown as formula (10):

[0178]

[0179] Among them, λ d|u (t) represents the incentive intensity (i.e., the association intensity in the above embodiment) of the historical search behavior of the target object u on the candidate document d in the search result list returned by the current search. u，d represents the basic incentive strength of the historical search behavior of the target object u for the candidate document d, is the historical search set corresponding to the target object, h represents the historical interaction content of the target object, α h，d represents the relevance between the historical interaction content of the target object and the content of the candidate document d, t represents the current time, t h Indicates the interaction time of historical interaction content.

[0180] The basic incentive intensity in this embodiment can be specifically defined as the long-term object interest expression The cosine similarity with the candidate document d is calculated using formula (11):

[0181]

[0182] Among them, μ u，d It represents the basic incentive strength of the historical search behavior of the target object u for the candidate document d, that is, the basic association strength in the above embodiment.

[0183] Among them, in this embodiment, α h，d It can be the product of the relevance of the historical interaction content h to the candidate document d and the relevance q of the historical interaction content h to the target search content of the current query, as shown in formula (12):

[0184] α h，d =cos(h,d)·cos(h,q) (12)

[0185] In this embodiment, the kernel function κ may use an exponential kernel function, referring to formula (13):

[0186] K(t)=e -γt (13)

[0187] Where γ is a learnable parameter. Calculate the incentive intensity λ of all candidate documents in the candidate document set D. k|uAfter (t), since the excitation intensity may be negative, this embodiment can use softmax (normalized exponential function) to convert it into a probability distribution, as shown in formula (14):

[0188]

[0189] 104. Sorting the target search results according to the target search content, the association strength, and the search interest feature information, and outputting the sorted search results.

[0190] In some embodiments, the target search content, association strength, and search interest feature information can be fused to obtain the recommendation relevance corresponding to each target search result. The fusion method can be a weighted operation or a splicing process, etc., which is not limited in this embodiment; the target search results are sorted based on the recommendation relevance, and the sorted search results are output.

[0191] Optionally, in this embodiment, the step of “ranking the target search results according to the target search content, the association strength, and the search interest feature information, and outputting the ranked search results” may include:

[0192] Extracting attention features from the target search content and historical search results of the target object to obtain current search intent features corresponding to the target object;

[0193] Calculate the similarity between each target search result and the current search intent feature to obtain the intent relevance corresponding to each target search result;

[0194] Based on the intention relevance, the association strength and the search interest feature information, the target search results are sorted and the sorted search results are output.

[0195] The recommendation relevance of each target search result can be calculated based on the intent relevance, the association strength, and the search interest feature information, and the target search results can be sorted based on the recommendation relevance, and the sorted search results can be output. Specifically, the target search results can be sorted from largest to smallest according to the recommendation relevance to obtain a sorted search result. Optionally, the top n target search results in the sorted search results can be returned to the target object.

[0196] Optionally, in this embodiment, target historical interaction content whose interaction time satisfies a preset time condition may be selected from the target object's historical interaction content; attention features are then extracted from the target search content and the target historical interaction content to obtain the current search intent features corresponding to the target object. The preset time condition may be set based on actual circumstances; specifically, the preset time condition may be that the time difference between the interaction time and the current time is no greater than a preset value.

[0197] Specifically, in some embodiments, a current query intent representation can be obtained by combining recent searches and the target search content currently retrieved, i.e., the current search intent feature in the above embodiments. First, the historical search set H of the target object i can be obtained. i (Specifically, it may only include the historical search results that the target object has clicked) intercept the most recent m historical search results to form a recent historical search result sequence Then combine the current search target content q and recent historical search results Input into the Transformer to calculate the search intent representation of the current query combined with recent search results As shown in formula (15):

[0198]

[0199] in, Indicates the current search intent features corresponding to the target object.

[0200] Then, each candidate document d and the current search intent representation are calculated The similarity p(d, q) between them is shown in formula (16):

[0201]

[0202] Among them, p(d, q) represents the intention relevance corresponding to the target search result (specifically, candidate document d).

[0203] In some embodiments, the search history characteristics of the target object calculated based on traditional statistical methods, including the diversity of target object queries, the diversity of interests, etc., can also be recorded as P r The final recommendation relevance of each target search result (i.e., candidate document) can be calculated by concatenating these features in the concatenation layer and inputting them into the multi-layer perceptron (scoring layer). The calculation process is shown in formula (17):

[0204]

[0205] Where p represents the recommendation relevance of candidate document d. Tanh is an activation function, and MLP stands for Multilayer Perceptron. represents the association strength of candidate document d, Search interest feature information of the target object.

[0206] After calculating the recommendation relevance of each target search result, the target search results can be sorted according to the recommendation relevance to obtain a final re-sorted document list and return it to the target object.

[0207] Optionally, in this embodiment, the step of "performing feature information transfer processing on the associated objects and the historical search results according to the heterogeneous search graph to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object" may include:

[0208] Through the content processing model, according to the heterogeneous search graph, feature information transmission processing is performed on the associated objects and the historical search results to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object.

[0209] The content processing model can be a neural network model. This embodiment can also train the entire model using LambdaRank (a ranking algorithm). The training data consists of document pairs including positive examples (historical search results clicked by the sample subject) and negative examples (historical search results not clicked by the sample subject). Δ is the change in Mean Average Precision (MAP) when the order of the two documents is swapped. Indicates whether it is true that document i ranks higher than document j, p ij represents the probability that document i is ranked higher than document j, Indicates whether it is true that document j ranks higher than document i, p ji represents the probability that document j ranks higher than document i. The loss function is defined as shown in formula (18):

[0210]

[0211] After defining the loss function, this embodiment can use the Adam (Adaptive Moment Estimation) SGD (stochastic gradient descent) optimizer to optimize the model parameters according to the loss function. After iteration, the model finally converges. After adjusting the parameters on the validation set, the model with the best performance is used for ranking in the actual search scenario.

[0212] like Figure 1d As shown, based on the content processing method provided by this application, the search results can be sorted according to the following steps, which are specifically described as follows:

[0213] 1. Establish a heterogeneous search graph based on the historical search results of the associated objects for the historical search content;

[0214] 2. Use heterogeneous graph neural networks to learn the search interest feature information P of the target object L , and obtain the target node feature information corresponding to each node, thereby obtaining the interactive feature information of the historical search results corresponding to the target object; wherein, in some embodiments, the search interest feature information P L Specifically, we can search the edge type feature information e in the heterogeneous graph u The cosine similarity (cos) with the target search result d is determined;

[0215] 3. Use the Hawkes Process Layer to model the historical search results (h1, h2, ..., h) corresponding to the target object using the time series process n ) The incentive intensity λ for the current search behavior d|u ; Among them, historical search results h1, h2, ..., h n The corresponding search times are t1, t2, ..., t n ;

[0216] 4. Use cosine similarity to calculate the intent relevance P of candidate document d based on the target search content q of the current query q , and obtain other features of the target object's search history through a multi-layer perceptron, such as the diversity of target object queries, the diversity of interests, etc., denoted as P r ; Thus according to P L 、P r ,λ d|u and P q Reorder each candidate document and return the reordered document list.

[0217] This embodiment provides a search technology based on a heterogeneous graph neural network and a temporal point process. This technology can construct a heterogeneous search graph by leveraging the relationships between objects and query content, objects and search results, and query content and search results in the search scenario. Based on the established heterogeneous search graph, a heterogeneous graph neural network is used to learn node feature information corresponding to each node. The feature information corresponding to query-type nodes and result-type nodes is then represented as interactive feature information of the search sequence. The Hawkes process is then used to model the temporal information in the search sequence. Finally, the target node feature information corresponding to the target object learned by the heterogeneous graph neural network is used as the target object's long-term object interest representation, and the condition strength obtained by the Hawkes process is used as the short-term interest representation. This allows the target object's preference for each candidate document to be calculated, thereby performing personalized re-ranking of the candidate documents.

[0218] Specifically, the content processing method provided by this application can deeply mine the relationships between objects and query content, objects and search results, and objects in the search scene. By using heterogeneous graph modeling to search for the relationships between entities in the scene, it can mine data such as user retrieval information and reading documents. Moreover, this application can achieve collaborative filtering by searching for objects with similar behaviors, alleviating the problem of insufficient search records for some objects. In addition, by using a time-series point process combined with a heterogeneous graph neural network modeling to model the impact of historical object search operations on current search operations, it is possible to dynamically capture the search interest of an object, thereby improving the accuracy of search result sorting.

[0219] As can be seen from the above, this embodiment can obtain the target search content to be searched for the target object currently, and the heterogeneous search graph corresponding to the target object, the heterogeneous search graph represents the interactive relationship between the associated object and the historical search results, and the historical search results are the search results of the associated object for the historical search content; according to the heterogeneous search graph, the associated object and the historical search results are processed with feature information transmission to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object; obtain at least one target search result for the target search content, and calculate the association strength of the historical search results of the target object for each target search result based on the interactive feature information; sort the target search results according to the target search content, the association strength and the search interest feature information, and output the sorted search results. The present application can sort the search results by deeply mining the heterogeneous search graph of the target object, which is conducive to improving the accuracy of the search result sorting, thereby facilitating the target object to quickly find the required results and improving the search efficiency.

[0220] According to the method described in the previous embodiment, the following will be further described in detail by taking the example of the content processing device being specifically integrated into the server.

[0221] The present application embodiment provides a content processing method, such as Figure 2 As shown, the specific process of the content processing method can be as follows:

[0222] 201. The server obtains target search content currently to be searched for a target object and a heterogeneous search graph corresponding to the target object, wherein the heterogeneous search graph represents an interactive relationship between an associated object and historical search results, wherein the historical search results are search results of the associated object for the historical search content.

[0223] The target search content specifically refers to the content currently to be searched, that is, the search content queried by the current target object; in a specific search scenario, the target search content may be the content entered by the current target object in the search content input box of the target application.

[0224] In this embodiment, the associated object may be an object that is related to the target object in terms of content interaction. It should be noted that the associated object may also include the target object itself. The historical search results of the associated object may be the search results obtained by the associated object when querying historical search content within a historical time period, specifically, the search results clicked by the associated object. The search results may include information in various modalities, such as video, audio, text, and images. Similarly, the search content may also include information in various modalities, such as video, audio, text, and images. This embodiment does not limit this.

[0225] Optionally, in this embodiment, the heterogeneous search graph includes at least one node and edges between nodes, the nodes include object type nodes corresponding to the associated objects and result type nodes corresponding to the historical search results, and the edges represent the interactive relationship between the associated objects and the historical search results.

[0226] Here, an edge is a connection between two nodes. Specifically, an edge can be used to connect an object type node and a result type node. That is, the node types at both ends of the connection are different.

[0227] In a specific embodiment, the heterogeneous search graph may also include query type nodes corresponding to the historical search content of the associated object. According to the node types connected at both ends of the edge, the edges in the heterogeneous search graph can be divided into three types, namely: edges connecting object type nodes and result type nodes, edges connecting object type nodes and query type nodes, and edges connecting query type nodes and result type nodes.

[0228] Optionally, this embodiment can generate object type nodes corresponding to each associated object, query type nodes corresponding to each historical search content, and result type nodes corresponding to each historical search result based on each associated object, the historical search content of each associated object, and the historical search results for the historical search content, and then connect each node based on the interactive relationship between the associated objects corresponding to each object type node, the historical search content corresponding to each query type node, and the historical search results corresponding to each result type node, thereby obtaining a heterogeneous search graph.

[0229] Optionally, in this embodiment, the content processing method may further include:

[0230] Obtain historical search results of the reference object for historical search content;

[0231] constructing an initial heterogeneous search graph for the target object based on an interactive relationship between the reference object and the historical search results corresponding to the reference object;

[0232] Based on the similarity between the reference object and the target object, selecting an associated object corresponding to the target object from the reference object;

[0233] Node deletion processing is performed on the initial heterogeneous search graph according to the associated object to obtain a heterogeneous search graph corresponding to the target object.

[0234] Specifically, the reference object may be an object that has clicked on the same search result as the target object. The similarity between the reference object and the target object may be determined based on the historical search content and clicked historical search results of the reference object and the similarity between the historical search content and clicked historical search results of the target object.

[0235] Optionally, in this embodiment, the step of “selecting, from the reference object, an associated object corresponding to the target object based on the similarity between the reference object and the target object” may include:

[0236] For each reference object, determining, based on the initial heterogeneous search graph, the number of paths between the object type node corresponding to the reference object and the object type node corresponding to the target object that meet a preset meta-path condition;

[0237] determining a similarity between the reference object and the target object based on the number of paths;

[0238] Based on the similarity, an associated object corresponding to the target object is selected from the reference objects.

[0239] Among them, the preset meta-path condition can be set according to actual conditions. For example, the preset meta-path condition can be the meta-path of object-search result-object (UDU). Then the number of paths that meet the preset meta-path condition between the object type node corresponding to the reference object and the object type node corresponding to the target object can be: the number of times the reference object and the target object have clicked on the same search result.

[0240] Optionally, in this embodiment, the step of “deleting nodes from the initial heterogeneous search graph according to the associated objects to obtain a heterogeneous search graph corresponding to the target object” may include:

[0241] Performing a node search in the initial heterogeneous search graph according to the associated object to determine a to-be-deleted node from the initial heterogeneous search graph, wherein the to-be-deleted node does not belong to an object type node of the associated object and a result type node corresponding to a historical search result of the associated object;

[0242] Deleting the nodes to be deleted in the initial heterogeneous search graph to obtain a heterogeneous search graph corresponding to the target object.

[0243] The node to be deleted may be an object type node of a non-associated object, or a result type node corresponding to a historical search result clicked by a non-associated object, or a query type node corresponding to a historical search content of a non-associated object.

[0244] 202. The server performs feature information transfer processing on the associated objects and the historical search results according to the heterogeneous search graph to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object.

[0245] Optionally, in this embodiment, the heterogeneous search graph includes at least one node and edges between nodes, the node including an object type node corresponding to the associated object and a result type node corresponding to the historical search result, and the edge represents an interactive relationship between the associated object and the historical search result;

[0246] The step of “performing feature information transfer processing on the associated objects and the historical search results according to the heterogeneous search graph to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object” may include:

[0247] In the heterogeneous search graph, feature information is transferred between the edges of the nodes to transfer the feature information of the associated objects and the historical search results, thereby obtaining the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object.

[0248] Optionally, in this embodiment, the step of “performing feature information transfer processing on edges between nodes in the heterogeneous search graph to perform feature information transfer processing on the associated objects and the historical search results to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object” may include:

[0249] Performing feature extraction on each node in the heterogeneous search graph to obtain node feature information corresponding to each node, wherein the node feature information includes node feature information corresponding to each object type node and node feature information corresponding to each result type node;

[0250] For each node in the heterogeneous search graph, performing a node search in the heterogeneous search graph according to the node to determine a neighbor node corresponding to the node;

[0251] performing feature information transfer processing on the node feature information of the node according to the node feature information of the neighboring node to obtain the target node feature information of the node;

[0252] Based on the target node feature information of the object type node corresponding to the target object, the search interest feature information of the target object and the interaction feature information of the historical search results corresponding to the target object are determined.

[0253] Among them, in some embodiments, the heterogeneous search graph can also include query type nodes corresponding to the historical search content of the associated object. In the heterogeneous search graph, by performing feature information transmission processing on the edges between nodes, feature information transmission processing can be performed on the associated object, historical search results, and historical search content.

[0254] In this embodiment, for each node in the heterogeneous search graph, the heterogeneous search graph is searched for neighbor nodes corresponding to the node. Specifically, the nodes in the heterogeneous search graph that are directly connected to the node are determined as neighbor nodes.

[0255] Optionally, in this embodiment, the step of “performing feature information transfer processing on the node feature information of the node according to the node feature information of the neighboring node to obtain the target node feature information of the node” may include:

[0256] Performing attention processing on the node feature information of the neighbor node according to the edge type feature information corresponding to the edge between the node and its corresponding neighbor node, and the node feature information of the node, to obtain the attention weight corresponding to the neighbor node;

[0257] Based on the attention weight and the node feature information of the neighboring nodes, the node feature information of the node is updated to obtain the target node feature information of the node.

[0258] This embodiment adds edge type feature information when calculating attention weights. This allows for consideration of edge types and better modeling of heterogeneous graphs. Specifically, edge types may include edges connecting object-type nodes and result-type nodes, edges connecting object-type nodes and query-type nodes, and edges connecting query-type nodes and result-type nodes. Different edge types may correspond to different edge type feature information, which may specifically be a parameter vector obtained through training.

[0259] The step of “performing attention processing on the node feature information of the neighbor node according to the edge type feature information corresponding to the edge between the node and its corresponding neighbor node, and the node feature information of the node, to obtain the attention weight corresponding to the neighbor node” may include:

[0260] Concatenate edge type feature information corresponding to the edge between the node and its corresponding neighbor node, and the node feature information of the node, with the node feature information of the neighbor node to obtain processed feature information corresponding to the neighbor node;

[0261] According to the processed feature information corresponding to the neighbor node, attention processing is performed on the node feature information of the neighbor node to obtain the attention weight corresponding to the neighbor node.

[0262] Optionally, in this embodiment, the step of “updating the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes to obtain the target node feature information of the node” may include:

[0263] Updating the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes;

[0264] Return to the step of performing attention processing on the node feature information of the neighboring node based on the edge type feature information corresponding to the edge between the node and its corresponding neighboring node, and the node feature information of the node, until the target node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

[0265] The step of "updating the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes" may include: fusing the attention weight corresponding to each neighboring node and the node feature information of each neighboring node to obtain fused feature information, and updating the node feature information of the node based on the fused feature information. Specifically, the fused feature information may be determined as the updated node feature information corresponding to the node. There are various fusion methods, which are not limited in this embodiment. For example, the fusion method may be weighted fusion.

[0266] Among them, this embodiment can perform a new round of attention processing based on the updated node feature information until the node feature information corresponding to the node that meets the preset information transmission conditions is obtained, and based on the node feature information corresponding to the node that meets the preset information transmission conditions, determine the target node feature information corresponding to the node.

[0267] The preset information transmission condition can be set according to actual conditions and is not limited in this embodiment. For example, the preset information transmission condition can be that the number of attention processing does not exceed a preset number. For example, if the preset number is 2, only two rounds of attention processing are performed.

[0268] Optionally, in this embodiment, the step of “determining search interest feature information of the target object and interaction feature information of historical search results corresponding to the target object based on target node feature information of the object type node corresponding to the target object” may include:

[0269] Determining search interest feature information of the target object based on target node feature information of the object type node corresponding to the target object;

[0270] For each historical search result of the target object, the interaction feature information corresponding to the historical search result is determined based on the target node feature information of the result type node corresponding to the historical search result, the target node feature information of the object type node corresponding to the target object, and the edge type feature information corresponding to the edge between the object type node and the result type node.

[0271] Among them, this embodiment can take the final vector of the node corresponding to the target object as the user's search interest feature information, specifically, the target node feature information of the object type node corresponding to the target object is determined as the search interest feature information of the target object, which can be recorded as the long-term object interest representation

[0272] In a specific embodiment, feature information can be used to represent the search behavior of the target object. For the interactive feature information corresponding to the operation of the target object retrieving a certain content (i.e., historical search content) in a historical time period, that is, the interactive feature information corresponding to the historical search content, it can be obtained by adding the target node feature information corresponding to the target object, the target node feature information of the node corresponding to the historical search content, and the object-retrieval content edge type feature information and averaging them. For the interactive feature information corresponding to the operation of the target object clicking on a certain search result (i.e., historical search result) in a historical time period, that is, the interactive feature information corresponding to the historical search result, it can be obtained by adding the target node feature information corresponding to the target object, the target node feature information of the node corresponding to the historical search result, and the object-search result edge type feature information and averaging them. In this way, the historical search set corresponding to the target object can be obtained. The interactive feature information of the target object i is shown in FIG1 , wherein the historical search set may include the historical search content of the target object i and the historical search results that have been clicked.

[0273] 203. The server obtains at least one target search result for the target search content, and calculates, based on the interaction feature information, the association strength of the historical search results of the target object with each target search result.

[0274] The target search results may be search results retrieved for the target search content currently being retrieved. In a specific scenario, each target search result may be a candidate document in a search result list retrieved for the target search content.

[0275] Optionally, in this embodiment, the step of “calculating, based on the interaction feature information, the association strength of the historical search results of the target object with each target search result” may include:

[0276] Determining time weight information corresponding to the historical search results of the target object based on the search time corresponding to the historical search results of the target object and the current time;

[0277] For each target search result, the association strength of the historical search results of the target object to the target search result is calculated based on the interaction feature information and the time weight information.

[0278] Specifically, the time weight information corresponding to the historical search results can be determined based on the time difference between the search time corresponding to the historical search results and the current time; the larger the time difference, the smaller the time weight information, and vice versa, the smaller the time difference, the larger the time weight information.

[0279] Optionally, in this embodiment, the step of “calculating, for each target search result, based on the interaction feature information and the time weight information, the association strength of the historical search results of the target object with the target search result” may include:

[0280] For each target search result, determining a basic association strength of the target search result based on the similarity between the search interest feature information of the target object and the target search result;

[0281] Determining content relevance between the historical search results of the target object and the target search results based on the relevance between the interaction feature information and the target search results, and the relevance between the interaction feature information and the target search content;

[0282] Based on the basic association strength, the content relevance, and the time weight information, the association strength of the historical search results of the target object to the target search results is calculated.

[0283] The calculation of the similarity between the search interest feature information of the target object and the target search result may adopt cosine similarity or the like, which is not limited in this embodiment.

[0284] The correlation between the interaction feature information and the target search results, as well as the correlation between the interaction feature information and the target search content, can also be determined using cosine similarity. Specifically, after calculating the correlation between the interaction feature information and the target search results, as well as the correlation between the interaction feature information and the target search content, these two correlations can be fused to obtain the content correlation between the historical search results of the target object and the target search results. The fusion processing method can specifically be multiplication or weighted operation.

[0285] The basic association strength, content relevance, and time weight information can be fused to obtain the association strength of the target search result. Specifically, the content relevance of each historical search result to the target search result and the time weight information corresponding to each historical search result can be weighted and fused to obtain the fused association strength. The fused association strength is then added to the basic association strength to obtain the association strength of the target search result.

[0286] 204. The server sorts the target search results according to the target search content, the association strength, and the search interest feature information, and outputs the sorted search results.

[0287] In some embodiments, the target search content, association strength, and search interest feature information can be fused to obtain the recommendation relevance corresponding to each target search result. The fusion method can be a weighted operation or a splicing process, etc., which is not limited in this embodiment; the target search results are sorted based on the recommendation relevance, and the sorted search results are output.

[0288] Optionally, in this embodiment, the step of “ranking the target search results according to the target search content, the association strength, and the search interest feature information, and outputting the ranked search results” may include:

[0289] Extracting attention features from the target search content and historical search results of the target object to obtain current search intent features corresponding to the target object;

[0290] Calculate the similarity between each target search result and the current search intent feature to obtain the intent relevance corresponding to each target search result;

[0291] Based on the intention relevance, the association strength and the search interest feature information, the target search results are sorted and the sorted search results are output.

[0292] The recommendation relevance of each target search result can be calculated based on the intent relevance, the association strength, and the search interest feature information, and the target search results can be sorted based on the recommendation relevance, and the sorted search results can be output. Specifically, the target search results can be sorted from largest to smallest according to the recommendation relevance to obtain a sorted search result. Optionally, the top n target search results in the sorted search results can be returned to the target object.

[0293] Optionally, in this embodiment, target historical interaction content whose interaction time satisfies a preset time condition may be selected from the target object's historical interaction content; attention features are then extracted from the target search content and the target historical interaction content to obtain the current search intent features corresponding to the target object. The preset time condition may be set based on actual circumstances; specifically, the preset time condition may be that the time difference between the interaction time and the current time is no greater than a preset value.

[0294] As can be seen from the above, this embodiment can obtain the target search content currently to be searched for the target object and the heterogeneous search graph corresponding to the target object through the server, the heterogeneous search graph represents the interactive relationship between the associated object and the historical search results, and the historical search results are the search results of the associated object for the historical search content; according to the heterogeneous search graph, the associated object and the historical search results are processed with feature information transmission to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object; obtain at least one target search result for the target search content, and calculate the association strength of the historical search results of the target object for each target search result based on the interactive feature information; sort the target search results according to the target search content, the association strength and the search interest feature information, and output the sorted search results. The present application can sort the search results by deeply mining the heterogeneous search graph of the target object, which is conducive to improving the accuracy of the search result sorting, thereby facilitating the target object to quickly find the required results and improving the search efficiency.

[0295] In order to better implement the above method, the embodiment of the present application also provides a content processing device, such as Figure 3 As shown, the content processing device may include an acquisition unit 301, an information transmission unit 302, a calculation unit 303, and a sorting unit 304, as follows:

[0296] (1) Acquisition unit 301;

[0297] An acquisition unit is used to acquire the target search content currently to be searched for the target object and the heterogeneous search graph corresponding to the target object, wherein the heterogeneous search graph represents the interactive relationship between the associated object and the historical search results, and the historical search results are the search results of the associated object for the historical search content.

[0298] Optionally, in some embodiments of the present application, the content processing apparatus may further include a search graph construction unit, the search graph construction unit being configured to construct a heterogeneous search graph for the target object; the search graph construction unit may include an acquisition subunit, a construction subunit, a selection subunit, and a deletion subunit, as follows:

[0299] The acquisition subunit is used to acquire historical search results of the reference object for historical search content;

[0300] A construction subunit, configured to construct an initial heterogeneous search graph for the target object based on an interactive relationship between the reference object and the historical search results corresponding to the reference object;

[0301] a selection subunit, configured to select, from the reference object, an associated object corresponding to the target object based on a similarity between the reference object and the target object;

[0302] The deletion subunit is used to perform node deletion processing on the initial heterogeneous search graph according to the associated object to obtain a heterogeneous search graph corresponding to the target object.

[0303] Optionally, in some embodiments of the present application, the selection sub-unit can be specifically used to determine, for each reference object, based on the initial heterogeneous search graph, the number of paths that meet preset meta-path conditions between the object type node corresponding to the reference object and the object type node corresponding to the target object; based on the number of paths, determine the similarity between the reference object and the target object; and based on the similarity, select the associated object corresponding to the target object from the reference object.

[0304] Optionally, in some embodiments of the present application, the deletion sub-unit can be specifically used to perform a node search in the initial heterogeneous search graph according to the associated object to determine the node to be deleted from the initial heterogeneous search graph, the node to be deleted does not belong to the object type node of the associated object, and the result type node corresponding to the historical search result of the associated object; the node to be deleted in the initial heterogeneous search graph is deleted to obtain the heterogeneous search graph corresponding to the target object.

[0305] (2) Information transmission unit 302;

[0306] An information transfer unit is used to perform feature information transfer processing on the associated objects and the historical search results according to the heterogeneous search graph to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object.

[0307] Optionally, in some embodiments of the present application, the heterogeneous search graph includes at least one node and edges between nodes, the node including an object type node corresponding to the associated object and a result type node corresponding to the historical search result, and the edge represents an interactive relationship between the associated object and the historical search result;

[0308] The information transmission unit can be specifically used to perform feature information transmission processing on the edges between nodes in the heterogeneous search graph to perform feature information transmission processing on the associated objects and the historical search results, so as to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object.

[0309] Optionally, in some embodiments of the present application, the information delivery unit may include a first extraction subunit, a search subunit, a delivery subunit, and a first determination subunit, as follows:

[0310] The first extraction subunit is used to perform feature extraction on each node in the heterogeneous search graph to obtain node feature information corresponding to each node, wherein the node feature information includes node feature information corresponding to each object type node and node feature information corresponding to each result type node;

[0311] A search subunit, configured to perform a node search in the heterogeneous search graph based on each node in the heterogeneous search graph to determine a neighbor node corresponding to the node;

[0312] a transmission subunit, configured to perform feature information transmission processing on the node feature information of the node according to the node feature information of the neighboring nodes, to obtain the target node feature information of the node;

[0313] The first determining subunit is configured to determine the search interest feature information of the target object and the interaction feature information of the historical search results corresponding to the target object based on the target node feature information of the object type node corresponding to the target object.

[0314] Optionally, in some embodiments of the present application, the transmission sub-unit can be specifically used to perform attention processing on the node feature information of the neighbor node based on the edge type feature information corresponding to the edge between the node and its corresponding neighbor node, and the node feature information of the node, to obtain the attention weight corresponding to the neighbor node; based on the attention weight and the node feature information of the neighbor node, the node feature information of the node is updated to obtain the target node feature information of the node.

[0315] Optionally, in some embodiments of the present application, the step of “updating the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes to obtain the target node feature information of the node” may include:

[0316] Updating the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes;

[0317] Return to the step of performing attention processing on the node feature information of the neighboring node based on the edge type feature information corresponding to the edge between the node and its corresponding neighboring node, and the node feature information of the node, until the target node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

[0318] (3) calculation unit 303;

[0319] A calculation unit is configured to obtain at least one target search result for the target search content, and calculate, based on the interaction feature information, the association strength of the historical search results of the target object with respect to each target search result.

[0320] Optionally, in some embodiments of the present application, the calculation unit may include a second determination subunit and a first calculation subunit, as follows:

[0321] The second determining subunit is configured to determine time weight information corresponding to the historical search results of the target object based on the search time corresponding to the historical search results of the target object and the current time;

[0322] The first calculation subunit is configured to calculate, for each target search result, based on the interaction feature information and the time weight information, the association strength of the historical search results of the target object to the target search result.

[0323] Optionally, in some embodiments of the present application, the first calculation subunit can be specifically used to determine, for each target search result, a basic association strength of the target search result based on the similarity between the search interest feature information of the target object and the target search result; determine the content relevance of the historical search results of the target object to the target search result based on the correlation between the interaction feature information and the target search result, and the correlation between the interaction feature information and the target search content; and calculate the association strength of the historical search results of the target object to the target search result based on the basic association strength, the content relevance and the time weight information.

[0324] (4) sorting unit 304;

[0325] The sorting unit is used to sort the target search results according to the target search content, the association strength and the search interest feature information, and output the sorted search results.

[0326] Optionally, in some embodiments of the present application, the sorting unit may include a second extraction subunit, a second calculation subunit, and a sorting subunit, as follows:

[0327] The second extraction subunit is used to extract attention features from the target search content and the historical search results of the target object to obtain the current search intention features corresponding to the target object;

[0328] The second calculation subunit is used to calculate the similarity between each target search result and the current search intention feature to obtain the intention relevance corresponding to each target search result;

[0329] The sorting subunit is used to sort the target search results based on the intention relevance, the association strength and the search interest feature information, and output the sorted search results.

[0330] As can be seen from the above, in this embodiment, the acquisition unit 301 can acquire the target search content currently to be searched for the target object, and the heterogeneous search graph corresponding to the target object, wherein the heterogeneous search graph represents the interactive relationship between the associated object and the historical search results, and the historical search results are the search results of the associated object for the historical search content; the information transmission unit 302 performs feature information transmission processing on the associated object and the historical search results according to the heterogeneous search graph to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object; the calculation unit 303 acquires at least one target search result for the target search content, and calculates the association strength of the historical search results of the target object for each target search result based on the interactive feature information; the sorting unit 304 sorts the target search results according to the target search content, the association strength and the search interest feature information, and outputs the sorted search results. The present application can sort the search results by deeply mining the heterogeneous search graph of the target object, which is conducive to improving the accuracy of the search result sorting, thereby facilitating the target object to quickly find the required results and improving the search efficiency.

[0331] The present application also provides an electronic device, such as Figure 4 , which shows a schematic diagram of the structure of an electronic device involved in an embodiment of the present application. The electronic device may be a terminal or a server, etc. Specifically:

[0332] The electronic device may include one or more processing core processors 401, one or more computer-readable storage media memories 402, a power supply 403, an input unit 404 and other components. Those skilled in the art will understand that Figure 4 The electronic device structure shown in the figure does not constitute a limitation of the electronic device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange components differently.

[0333] Processor 401 is the control center of the electronic device. It connects all parts of the electronic device using various interfaces and circuits. It performs various functions of the electronic device and processes data by running or executing software programs and / or modules stored in memory 402 and accessing data stored in memory 402. Optionally, processor 401 may include one or more processing cores. Preferably, processor 401 may integrate an application processor and a modem processor. The application processor primarily handles the operating system, user interface, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into processor 401.

[0334] 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, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 402 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0335] The electronic device also includes a power supply 403 for supplying power to various components. Preferably, the power supply 403 can be logically connected to the processor 401 via a power management system, thereby enabling the power management system to manage charging, discharging, and power consumption. The power supply 403 can also include one or more DC or AC power supplies, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and other arbitrary components.

[0336] The electronic device may further include an input unit 404, which may be configured to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0337] Although not shown, the electronic device may further include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 401 in the electronic device 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 to implement various functions as follows:

[0338] Obtain the target search content to be searched for the target object currently, and the heterogeneous search graph corresponding to the target object, wherein the heterogeneous search graph represents the interactive relationship between the associated object and the historical search results, and the historical search results are the search results of the associated object for the historical search content; according to the heterogeneous search graph, perform feature information transmission processing on the associated object and the historical search results to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object; obtain at least one target search result for the target search content, and based on the interactive feature information, calculate the association strength of the historical search results of the target object to each target search result; according to the target search content, the association strength and the search interest feature information, sort the target search results and output the sorted search results.

[0339] The specific implementation of the above operations can be found in the previous embodiments and will not be repeated here.

[0340] As can be seen from the above, this embodiment can obtain the target search content to be searched for the target object currently, and the heterogeneous search graph corresponding to the target object, the heterogeneous search graph represents the interactive relationship between the associated object and the historical search results, and the historical search results are the search results of the associated object for the historical search content; according to the heterogeneous search graph, the associated object and the historical search results are processed with feature information transmission to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object; obtain at least one target search result for the target search content, and calculate the association strength of the historical search results of the target object for each target search result based on the interactive feature information; sort the target search results according to the target search content, the association strength and the search interest feature information, and output the sorted search results. The present application can sort the search results by deeply mining the heterogeneous search graph of the target object, which is conducive to improving the accuracy of the search result sorting, thereby facilitating the target object to quickly find the required results and improving the search efficiency.

[0341] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be accomplished by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0342] To this end, an embodiment of the present application provides a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute the steps of any of the content processing methods provided in the embodiments of the present application. For example, the instructions can execute the following steps:

[0343] Obtain the target search content to be searched for the target object currently, and the heterogeneous search graph corresponding to the target object, wherein the heterogeneous search graph represents the interactive relationship between the associated object and the historical search results, and the historical search results are the search results of the associated object for the historical search content; according to the heterogeneous search graph, perform feature information transmission processing on the associated object and the historical search results to obtain the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object; obtain at least one target search result for the target search content, and based on the interactive feature information, calculate the association strength of the historical search results of the target object to each target search result; according to the target search content, the association strength and the search interest feature information, sort the target search results and output the sorted search results.

[0344] The specific implementation of the above operations can be found in the previous embodiments and will not be repeated here.

[0345] The computer-readable storage medium may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0346] Since the instructions stored in the computer-readable storage medium can execute the steps in any content processing method provided in the embodiments of the present application, the beneficial effects that can be achieved by any content processing method provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0347] According to one aspect of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional implementations of the aforementioned content processing aspects.

[0348] The above is a detailed introduction to a content processing method and related equipment provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for technical personnel in this field, based on the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A content processing method, characterized in that: include: Obtaining target search content currently to be searched for a target object and a heterogeneous search graph corresponding to the target object, wherein the heterogeneous search graph represents an interactive relationship between associated objects and historical search results, wherein the historical search results are search results of the associated objects for the historical search content, and the associated objects include the target object and objects that are related to the target object in terms of content interaction; According to the heterogeneous search graph, feature information transfer processing is performed on the associated objects and the historical search results to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object; Obtaining at least one target search result for the target search content; Determining time weight information corresponding to the historical search results of the target object based on the search time corresponding to the historical search results of the target object and the current time; For each target search result, based on the interaction feature information and the time weight information, calculating the association strength of the historical search results of the target object to the target search result; The target search results are sorted according to the target search content, the association strength, and the search interest feature information, and the sorted search results are output.

2. The method according to claim 1, characterized in that The heterogeneous search graph includes at least one node and edges between nodes, the node including an object type node corresponding to the associated object and a result type node corresponding to the historical search result, and the edge represents an interactive relationship between the associated object and the historical search result; The step of performing feature information transfer processing on the associated objects and the historical search results according to the heterogeneous search graph to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object includes: In the heterogeneous search graph, feature information is transferred between the edges of the nodes to transfer the feature information of the associated objects and the historical search results, thereby obtaining the search interest feature information of the target object and the interactive feature information of the historical search results corresponding to the target object.

3. The method according to claim 2, characterized in that The method further comprises: Obtain historical search results of the reference object for historical search content; constructing an initial heterogeneous search graph for the target object based on an interactive relationship between the reference object and the historical search results corresponding to the reference object; Based on the similarity between the reference object and the target object, selecting an associated object corresponding to the target object from the reference object; Node deletion processing is performed on the initial heterogeneous search graph according to the associated object to obtain a heterogeneous search graph corresponding to the target object.

4. The method according to claim 3, characterized in that The selecting, from the reference object, an associated object corresponding to the target object based on the similarity between the reference object and the target object includes: For each reference object, determining, based on the initial heterogeneous search graph, the number of paths between the object type node corresponding to the reference object and the object type node corresponding to the target object that meet a preset meta-path condition; determining a similarity between the reference object and the target object based on the number of paths; Based on the similarity, an associated object corresponding to the target object is selected from the reference objects.

5. The method according to claim 3, characterized in that The performing node deletion processing on the initial heterogeneous search graph according to the associated object to obtain the heterogeneous search graph corresponding to the target object includes: Performing a node search in the initial heterogeneous search graph according to the associated object to determine a to-be-deleted node from the initial heterogeneous search graph, wherein the to-be-deleted node does not belong to an object type node of the associated object and a result type node corresponding to a historical search result of the associated object; Deleting the nodes to be deleted in the initial heterogeneous search graph to obtain a heterogeneous search graph corresponding to the target object.

6. The method according to claim 2, characterized in that The process of performing feature information transfer processing on edges between nodes in the heterogeneous search graph to transfer feature information on the associated objects and the historical search results to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object includes: Performing feature extraction on each node in the heterogeneous search graph to obtain node feature information corresponding to each node, wherein the node feature information includes node feature information corresponding to each object type node and node feature information corresponding to each result type node; For each node in the heterogeneous search graph, performing a node search in the heterogeneous search graph according to the node to determine a neighbor node corresponding to the node; performing feature information transfer processing on the node feature information of the node according to the node feature information of the neighboring node to obtain the target node feature information of the node; Based on the target node feature information of the object type node corresponding to the target object, the search interest feature information of the target object and the interaction feature information of the historical search results corresponding to the target object are determined.

7. The method according to claim 6, characterized in that The performing feature information transfer processing on the node feature information of the node according to the node feature information of the neighboring node to obtain the target node feature information of the node includes: Performing attention processing on the node feature information of the neighbor node according to the edge type feature information corresponding to the edge between the node and its corresponding neighbor node, and the node feature information of the node, to obtain the attention weight corresponding to the neighbor node; Based on the attention weight and the node feature information of the neighboring nodes, the node feature information of the node is updated to obtain the target node feature information of the node.

8. The method according to claim 7, characterized in that The updating of the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes to obtain the target node feature information of the node includes: Updating the node feature information of the node based on the attention weight and the node feature information of the neighboring nodes; Return to the step of performing attention processing on the node feature information of the neighboring node based on the edge type feature information corresponding to the edge between the node and its corresponding neighboring node, and the node feature information of the node, until the target node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

9. The method according to claim 1, characterized in that The step of calculating, for each target search result, the association strength of the historical search results of the target object with the target search result based on the interaction feature information and the time weight information includes: For each target search result, determining a basic association strength of the target search result based on the similarity between the search interest feature information of the target object and the target search result; Determining content relevance between the historical search results of the target object and the target search results based on the relevance between the interaction feature information and the target search results, and the relevance between the interaction feature information and the target search content; Based on the basic association strength, the content relevance, and the time weight information, the association strength of the historical search results of the target object to the target search results is calculated.

10. The method according to claim 1, characterized in that The step of sorting the target search results according to the target search content, the association strength, and the search interest feature information, and outputting the sorted search results includes: Extracting attention features from the target search content and historical search results of the target object to obtain current search intent features corresponding to the target object; Calculate the similarity between each target search result and the current search intent feature to obtain the intent relevance corresponding to each target search result; Based on the intention relevance, the association strength and the search interest feature information, the target search results are sorted and the sorted search results are output.

11. A content processing device, characterized in that: include: an acquisition unit, configured to acquire target search content currently to be searched for a target object, and a heterogeneous search graph corresponding to the target object, wherein the heterogeneous search graph represents an interactive relationship between associated objects and historical search results, wherein the historical search results are search results of the associated objects for the historical search content, and the associated objects include the target object and objects that are related to the target object in terms of content interaction; An information transfer unit is configured to perform feature information transfer processing on the associated objects and the historical search results according to the heterogeneous search graph to obtain search interest feature information of the target object and interactive feature information of the historical search results corresponding to the target object; a computing unit, configured to obtain at least one target search result for the target search content; Determining time weight information corresponding to the historical search results of the target object based on the search time corresponding to the historical search results of the target object and the current time; For each target search result, based on the interaction feature information and the time weight information, calculating the association strength of the historical search results of the target object to the target search result; The sorting unit is used to sort the target search results according to the target search content, the association strength and the search interest feature information, and output the sorted search results.

12. An electronic device, characterized in that: The invention comprises a memory and a processor; the memory stores an application program, and the processor is used to run the application program in the memory to perform the operations in the content processing method according to any one of claims 1 to 10.

13. 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 content processing method according to any one of claims 1 to 10.

14. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the content processing method according to any one of claims 1 to 10 are implemented.

Images