Text information recommendation method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a text information recommendation method and device, electronic equipment and a storage medium, and relates to the technical field of data analysis, the method comprises the following steps: obtaining a data table, the data table comprising a patent data table and a user data table; obtaining a heterogeneous graph convolution recommendation model and a natural language model, wherein the heterogeneous graph convolution recommendation model at least comprises a semantic convolution layer and a heterogeneous convolution layer; inputting the patent data table and the user data table into a semantic convolutional layer for semantic extraction to obtain a first semantic representation and a second semantic representation; inputting the first semantic representation and the second semantic representation into a heterogeneous convolutional layer for keyword recommendation to obtain a first keyword combination; inputting the first keyword combination into a natural language model for keyword recommendation to obtain a second keyword combination; and text retrieval is performed according to the first keyword combination and the second keyword combination, and the corresponding target patent text is output, so that the comprehensiveness and accuracy of retrieval are improved.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a text information recommendation method, a text information recommendation device, an electronic device, and a computer-readable storage medium. Background Art

[0002] As global economic and technological competition intensifies, patents, as a crucial manifestation and vehicle for scientific and technological innovation, have become a crucial lever for driving modern social progress and economic and technological development. Enterprises can acquire innovative technologies through two main paths: independent research and development (R&D) and technology introduction. Both approaches require obtaining relevant patents related to the innovative technology. For example, when choosing independent R&D, patent avoidance designs are necessary to circumvent the protection scope of relevant patents and prevent infringement. When choosing technology introduction, technical analysis of relevant patents is necessary to select the most appropriate and advanced patented technologies for targeted introduction.

[0003] However, obtaining a sufficient number of patent documents related to innovative technologies from massive patent databases is a complex and specialized task that most small and medium-sized enterprises lack the necessary skills. In related technologies, patent document retrieval methods often suffer from a high rate of missed detections, easily overlooking a large number of important patent documents with high technical similarity, making it difficult to achieve satisfactory patent recommendation results. Summary of the Invention

[0004] The embodiments of the present invention provide a method, device, electronic device, and computer-readable storage medium for recommending text information to solve or partially solve the problem of a high missed detection rate in the process of recommending patent texts.

[0005] An embodiment of the present invention discloses a method for recommending text information, including:

[0006] Obtaining a data table, wherein the data table includes at least a patent data table and a user data table;

[0007] Obtaining a heterogeneous graph convolutional recommendation model and a natural language model, wherein the heterogeneous graph convolutional recommendation model includes at least a semantic convolutional layer and a heterogeneous convolutional layer;

[0008] Inputting the patent data table and the user data table into a semantic convolution layer for semantic extraction to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table;

[0009] Inputting the first semantic representation and the second semantic representation into the heterogeneous convolutional layer for keyword recommendation to obtain at least one corresponding first keyword combination;

[0010] Inputting the first keyword combination into the natural language model for keyword recommendation to obtain at least one corresponding second keyword combination;

[0011] Perform text search according to the first keyword combination and the second keyword combination, and output the corresponding target patent text.

[0012] In some feasible implementations, the data table further includes a project data table, and inputting the patent data table and the user data table into a semantic convolution layer for semantic extraction to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table includes:

[0013] extracting employee information from the employee data table and using the employee information as semantic data;

[0014] Extracting project information from the project data table and patent information from the patent data table, and using the project information and the patent information as auxiliary semantic data corresponding to the semantic data;

[0015] extracting a plurality of patent keywords from the patent information;

[0016] The semantic data, the auxiliary semantic data and each of the patent keywords are trained to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table.

[0017] In some feasible implementations, the semantic convolution layer is a graph convolution network, which includes at least an input layer, multiple execution convolution layers, an activation function, a pooling layer, and an output layer; wherein the graph convolution network also includes a first user node and a first patent node, and the training of the semantic data, the auxiliary semantic data, and each of the patent keywords to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table includes:

[0018] Converting the semantic data and the auxiliary semantic data into first embedding information;

[0019] Converting the plurality of patent keywords into second embedded information;

[0020] Using the first embedding vector as the node feature of the first user node and the second embedding vector as the node feature of the first patent node, obtaining a first feature vector of the patent data table and a second feature vector of the user data table;

[0021] Aggregating information of neighboring nodes by executing the convolutional layer, iteratively updating the first feature vector and the second feature vector to obtain a node representation that integrates neighborhood semantics;

[0022] Performing a nonlinear transformation on the node representation of the fused neighborhood semantics according to the activation function to obtain a transformed node representation;

[0023] The transformed node representation is reduced in dimension by the pooling layer to obtain a compressed node representation, and the first semantic representation corresponding to the patent data table and the second semantic representation corresponding to the user data table are generated by the linear transformation and task adaptation processing of the output layer.

[0024] In some feasible implementations, the employee information includes at least one of user basic information, professional background information, job data, research direction and historical achievements; the project information includes at least one of project basic information, technical content and achievement data; the patent information includes at least one of title, abstract, claim text, patent type, classification information, technical feature text, conversion data, pledge information, financing information and text-derived information.

[0025] In some feasible implementations, inputting the first semantic representation and the second semantic representation into the heterogeneous convolutional layer for keyword recommendation to obtain at least one corresponding first keyword combination includes:

[0026] Constructing a heterogeneous graph for the first semantic representation and the second semantic representation, wherein the heterogeneous graph includes several different types of graph nodes, and the different types of graph nodes in the heterogeneous graph are connected through corresponding node relationships;

[0027] Obtaining node attributes corresponding to each of the graph nodes;

[0028] Obtaining weight values ​​corresponding to each of the node relationships;

[0029] The first semantic representation and the second semantic representation are input into the heterogeneous convolutional layer, and keyword recommendations are performed according to the node attributes, the node relationships, and the weight values ​​to obtain at least one corresponding first keyword combination.

[0030] In some feasible implementations, inputting the first semantic representation and the second semantic representation into the heterogeneous convolutional layer, performing keyword recommendation based on the node attributes, the node relationships, and the weight values, and obtaining the corresponding at least one first keyword combination includes:

[0031] Performing path search in the heterogeneous graph using the weight values ​​and the node relationships to obtain candidate edge sequences corresponding to the first semantic representation and the second semantic representation;

[0032] Performing a weight-sensitive random walk on the candidate edge sequence to obtain a node sequence for the candidate edge sequence, and calculating a total path weight corresponding to the node sequence according to the weight value;

[0033] The node sequence whose total path weight is greater than or equal to the preset threshold is taken as the initial meta-path;

[0034] enhancing the initial meta-path using the node attributes to obtain a semantically enhanced meta-path;

[0035] Performing stratified sampling on the semantically enhanced meta-path to obtain a meta-path neighbor dictionary corresponding to each of the graph nodes;

[0036] Keyword recommendation is performed on the meta-path neighbor dictionary according to the heterogeneous convolutional layer to obtain at least one first keyword combination corresponding to the first semantic representation and the second semantic representation.

[0037] In some feasible implementations, recommending keywords on the meta-path neighbor dictionary according to the heterogeneous convolutional layer to obtain at least one first keyword combination corresponding to the first semantic representation and the second semantic representation includes:

[0038] Recalling a parameter matrix for the meta-path neighbor dictionary from the heterogeneous convolutional layer;

[0039] Performing feature transformation on the meta-path neighbor dictionary using the parameter matrix and the weight value, performing weighted aggregation on the transformed output result, and outputting a grouped aggregation result corresponding to the meta-path neighbor dictionary;

[0040] Calculating the grouping aggregation results using trained attention mechanism parameters, outputting corresponding calculation results, performing feature weighted aggregation and residual connection on the calculation results, and outputting a first target semantic representation of the first semantic representation and a second target semantic representation corresponding to the second semantic representation;

[0041] Obtaining a similarity algorithm for the first target semantic representation and the second target semantic representation;

[0042] The similarity algorithm is used to perform keyword recommendation on the first target semantic representation and the second target semantic representation, and to generate at least one corresponding first keyword combination.

[0043] The embodiment of the present invention further discloses a device for recommending text information, comprising:

[0044] A data acquisition module, configured to acquire a data table, wherein the data table includes at least a patent data table and a user data table;

[0045] A model acquisition module is used to obtain a heterogeneous graph convolutional recommendation model and a natural language model, wherein the heterogeneous graph convolutional recommendation model includes at least a semantic convolutional layer and a heterogeneous convolutional layer;

[0046] a semantic extraction module, configured to input the patent data table and the user data table into a semantic convolution layer for semantic extraction, and obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table;

[0047] A first recommendation module, configured to input the first semantic representation and the second semantic representation into the heterogeneous convolutional layer for keyword recommendation to obtain at least one corresponding first keyword combination;

[0048] a second recommendation module, configured to input the first keyword combination into the natural language model for keyword recommendation, and obtain at least one corresponding second keyword combination;

[0049] A text retrieval module is used to perform text retrieval according to the first keyword combination and the second keyword combination, and output the corresponding target patent text.

[0050] In some feasible implementations, the data table further includes a project data table, and the semantic extraction module is specifically configured to:

[0051] extracting employee information from the employee data table and using the employee information as semantic data;

[0052] Extracting project information from the project data table and patent information from the patent data table, and using the project information and the patent information as auxiliary semantic data corresponding to the semantic data;

[0053] extracting a plurality of patent keywords from the patent information;

[0054] The semantic data, the auxiliary semantic data and each of the patent keywords are trained to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table.

[0055] In some feasible implementations, the semantic convolution layer is a graph convolution network, which includes at least an input layer, multiple execution convolution layers, an activation function, a pooling layer, and an output layer; wherein the graph convolution network also includes a first user node and a first patent node, and the semantic extraction module is specifically used to:

[0056] Converting the semantic data and the auxiliary semantic data into first embedding information;

[0057] Converting the plurality of patent keywords into second embedded information;

[0058] Using the first embedding vector as the node feature of the first user node and the second embedding vector as the node feature of the first patent node, obtaining a first feature vector of the patent data table and a second feature vector of the user data table;

[0059] Aggregating information of neighboring nodes by executing the convolutional layer, iteratively updating the first feature vector and the second feature vector to obtain a node representation that integrates neighborhood semantics;

[0060] Performing a nonlinear transformation on the node representation of the fused neighborhood semantics according to the activation function to obtain a transformed node representation;

[0061] The transformed node representation is reduced in dimension by the pooling layer to obtain a compressed node representation, and the first semantic representation corresponding to the patent data table and the second semantic representation corresponding to the user data table are generated by the linear transformation and task adaptation processing of the output layer.

[0062] In some feasible implementations, the employee information includes at least one of user basic information, professional background information, job data, research direction and historical achievements; the project information includes at least one of project basic information, technical content and achievement data; the patent information includes at least one of title, abstract, claim text, patent type, classification information, technical feature text, conversion data, pledge information, financing information and text-derived information.

[0063] In some feasible implementations, the first recommendation module is specifically configured to:

[0064] Constructing a heterogeneous graph for the first semantic representation and the second semantic representation, wherein the heterogeneous graph includes several different types of graph nodes, and the different types of graph nodes in the heterogeneous graph are connected through corresponding node relationships;

[0065] Obtaining node attributes corresponding to each of the graph nodes;

[0066] Obtaining weight values ​​corresponding to each of the node relationships;

[0067] The first semantic representation and the second semantic representation are input into the heterogeneous convolutional layer, and keyword recommendations are performed according to the node attributes, the node relationships, and the weight values ​​to obtain at least one corresponding first keyword combination.

[0068] In some feasible implementations, the first recommendation module is specifically configured to:

[0069] Performing path search in the heterogeneous graph using the weight values ​​and the node relationships to obtain candidate edge sequences corresponding to the first semantic representation and the second semantic representation;

[0070] Performing a weight-sensitive random walk on the candidate edge sequence to obtain a node sequence for the candidate edge sequence, and calculating a total path weight corresponding to the node sequence according to the weight value;

[0071] The node sequence whose total path weight is greater than or equal to the preset threshold is taken as the initial meta-path;

[0072] enhancing the initial meta-path using the node attributes to obtain a semantically enhanced meta-path;

[0073] Performing stratified sampling on the semantically enhanced meta-path to obtain a meta-path neighbor dictionary corresponding to each of the graph nodes;

[0074] Keyword recommendation is performed on the meta-path neighbor dictionary according to the heterogeneous convolutional layer to obtain at least one first keyword combination corresponding to the first semantic representation and the second semantic representation.

[0075] In some feasible implementations, the first recommendation module is specifically configured to:

[0076] Recalling a parameter matrix for the meta-path neighbor dictionary from the heterogeneous convolutional layer;

[0077] Performing feature transformation on the meta-path neighbor dictionary using the parameter matrix and the weight value, performing weighted aggregation on the transformed output result, and outputting a grouped aggregation result corresponding to the meta-path neighbor dictionary;

[0078] Calculating the grouping aggregation results using trained attention mechanism parameters, outputting corresponding calculation results, performing feature weighted aggregation and residual connection on the calculation results, and outputting a first target semantic representation of the first semantic representation and a second target semantic representation corresponding to the second semantic representation;

[0079] Obtaining a similarity algorithm for the first target semantic representation and the second target semantic representation;

[0080] The similarity algorithm is used to perform keyword recommendation on the first target semantic representation and the second target semantic representation, and to generate at least one corresponding first keyword combination.

[0081] An embodiment of the present invention further discloses an electronic device, comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0082] The memory is used to store computer programs;

[0083] The processor is configured to implement the method described in the embodiment of the present invention when executing the program stored in the memory.

[0084] An embodiment of the present invention further discloses a computer-readable storage medium having instructions stored thereon. When executed by one or more processors, the processors are enabled to execute the method according to the embodiment of the present invention.

[0085] The embodiments of the present invention include the following advantages:

[0086] In an embodiment of the present invention, when a user wants to search for relevant text, a data table is obtained, which includes at least a patent data table and a user data table. Then, a heterogeneous graph convolution recommendation model and a natural language model are obtained. The heterogeneous graph convolution recommendation model includes at least a semantic convolution layer and a heterogeneous convolution layer. Then, the patent data table and the user data table are input into the semantic convolution layer for semantic extraction, and a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table are obtained. The first semantic representation and the second semantic representation are input into the heterogeneous convolution layer for keyword recommendation, and at least one corresponding first keyword combination is obtained. The first keyword combination is input into the natural language model for keyword recommendation, and at least one corresponding second keyword combination is obtained. Finally, text retrieval is performed according to the first keyword combination and the second keyword combination, and the corresponding target patent text is output. Thus, keyword recommendation processing is performed through the semantic convolution layer and the heterogeneous convolution layer, making the information richer and the recommendation results more accurate. At the same time, by using the natural language model to generate new keyword combinations based on the recommended keyword combinations, more diverse choices are provided to users, reducing the probability of users falling into information cocoons, and text retrieval based on rich keyword combinations effectively ensures the comprehensiveness and accuracy of the retrieval results. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 This is a flowchart of a method for recommending text information provided in an embodiment of the present invention;

[0088] Figure 2 is a schematic diagram of a recommended process provided in an embodiment of the present invention;

[0089] Figure 3 is a schematic diagram of an application scenario provided in an embodiment of the present invention;

[0090] Figure 4 This is a structural block diagram of a text information recommendation device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0091] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0092] For example, obtaining a sufficient number of patent documents related to innovative technologies from massive patent databases is a complex and specialized task that most small and medium-sized enterprises lack the necessary skills. In related technologies, patent document retrieval methods often suffer from a high rate of missed detections, easily overlooking a large number of important patent documents with high technical similarity, making it difficult to achieve satisfactory patent recommendation results.

[0093] Reference Figure 1 , shows a flowchart of a method for recommending text information provided in an embodiment of the present invention, which may specifically include the following steps:

[0094] Step 101: obtaining a data table, wherein the data table includes at least a patent data table and a user data table;

[0095] For company employees, especially those responsible for technology research and development, the process of developing related technologies inevitably requires obtaining relevant patents related to the innovative technology. For example, when choosing an independent research and development path, patent avoidance designs are necessary to circumvent the protection scope of relevant patents and prevent infringement. When choosing a technology introduction path, technical analysis of relevant patents is required to select the most appropriate and advanced patented technologies for targeted introduction. Based on this, retrieving comprehensive and highly accurate patent texts can effectively assist R&D personnel in technological innovation.

[0096] When users (such as R&D personnel) search for relevant patent texts, they can construct corresponding data tables to extract related information from the data tables for patent text retrieval. Among them, the data tables can at least include patent data tables, user data tables, project data tables, and enterprise data tables. The patent data tables can at least include information such as patent abstracts, patent titles, and technical fields corresponding to different patent texts; user data tables can include information such as the professions, positions, and research directions of employees within the enterprise; project data tables can include information such as project titles, research contents, key technologies, and project difficulties corresponding to the enterprise's R&D projects; and enterprise data tables can include information such as the enterprise's main business information and product information. By constructing corresponding data tables, a good data foundation can be provided for subsequent text retrieval, ensuring the comprehensiveness and accuracy of text retrieval.

[0097] Step 102: Obtain a heterogeneous graph convolutional recommendation model and a natural language model, wherein the heterogeneous graph convolutional recommendation model includes at least a semantic convolutional layer and a heterogeneous convolutional layer;

[0098] After building the corresponding data table, you can also obtain the heterogeneous graph convolutional recommendation model and natural language model for text recommendation. Among them, the heterogeneous graph convolutional recommendation model can be a convolutional model for content recommendation based on heterogeneous graphs. It is a model designed for heterogeneous graphs (including graphs of multiple types of nodes and edges) based on the graph convolutional network. The heterogeneous graph convolutional recommendation model can effectively handle heterogeneity and retain multi-relationship information, while effectively performing feature fusion and representation learning.

[0099] For example, in heterogeneous graphs, nodes and edges of various types represent different semantic information. Therefore, the heterogeneous graph convolutional model can fully utilize this heterogeneous information to distinguish and process different types of nodes and edges. Furthermore, heterogeneous graphs contain multiple relationships, which reflect the interactions and connections between different entities. The heterogeneous graph convolutional recommendation model can preserve this multi-relational information and consider the impact of different relationships during information transmission. In the feature fusion process, the heterogeneous graph convolutional recommendation model can effectively fuse the features of different types of nodes. In heterogeneous graphs, different types of nodes may have features with different dimensions and meanings. The heterogeneous graph convolutional recommendation model can integrate these features through convolution operations to generate more expressive node representations. In the representation learning process, the heterogeneous graph convolutional recommendation model can learn low-dimensional vector representations of nodes and edges in heterogeneous graphs. These representations can capture the structure and semantic information of the graph. By performing convolution operations on heterogeneous graphs, the model can automatically learn feature representations for different types of nodes and edges, making nodes with similar structures and semantics closer in vector space. Therefore, the heterogeneous graph convolutional recommendation model can effectively perform content recommendation.

[0100] In some feasible implementations, the heterogeneous graph convolutional recommendation model may include a semantic convolution layer and a heterogeneous convolution layer. The semantic convolution layer can be used to extract the corresponding semantic representation from the data table and input the semantic representation into the heterogeneous convolution layer for further data processing (such as collaborative processing, etc.) to enrich the information and improve the comprehensiveness and accuracy of the recommendation results.

[0101] Furthermore, the natural language model can be an artificial intelligence model for processing and generating natural language. It learns and models the patterns, structures, and semantics of natural language to understand, generate, and process human language. In this embodiment of the present invention, by inputting a corresponding keyword combination into the natural language model, the model can output another related but different keyword combination, thereby providing users with additional recommendation options and offering them more diverse choices.

[0102] Step 103: Input the patent data table and the user data table into a semantic convolution layer for semantic extraction to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table;

[0103] In an embodiment of the present invention, after the data table and the heterogeneous convolutional recommendation model are determined, the patent data table and the user data table in the data table can be input into the semantic convolution layer for semantic extraction to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table. By extracting the corresponding semantic representations, user information, patent information, etc. can be better understood so as to recommend more suitable content to users, thereby improving the accuracy and personalization of the recommendations.

[0104] In some feasible implementations, in the process of extracting semantic representations, the data table also includes a project data table. Employee information can be extracted from the employee data table and used as semantic data. Project information can be extracted from the project data table, and patent information can be extracted from the patent data table. Project information and patent information can be used as auxiliary semantic data corresponding to the semantic data. Multiple patent keywords can be extracted from the patent information. The semantic data, auxiliary semantic data, and each patent keyword are then trained to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table.

[0105] Among them, employee information includes at least one of user basic information, professional background information, job data, research direction and historical achievements; project information includes at least one of project basic information, technical content and achievement data; patent information includes at least one of title, abstract, claim text, patent type, classification information, technical feature text, conversion data, pledge information, financing information and text-derived information, so as to extract more characteristic information from the data table and input it into the semantic convolution layer for semantic understanding, and obtain the corresponding semantic representation, so as to recommend more suitable content to users, thereby improving the accuracy and personalization of the recommendation.

[0106] Optionally, the semantic convolution layer is a graph convolution network, which includes at least an input layer, multiple execution convolution layers, an activation function, a pooling layer, and an output layer; wherein the graph convolution network also includes a first user node and a first patent node, then the semantic data and the auxiliary semantic data can be converted into the first embedding information, and the multiple patent keywords can be converted into the second embedding information, and then the first embedding vector is used as the node feature of the first user node, and the second embedding vector is used as the node feature of the first patent node to obtain the first feature vector of the patent data table and the second feature vector of the user data table, and then the information of the neighboring nodes is aggregated by the execution convolution layer, and the first feature vector is iteratively updated. The vector and the second eigenvector are used to obtain the node representation of the fused neighborhood semantics, and then the node representation of the fused neighborhood semantics is nonlinearly transformed according to the activation function to obtain the transformed node representation. Finally, the transformed node representation is reduced in dimension through the pooling layer to obtain the compressed node representation, and the first semantic representation corresponding to the patent data table and the second semantic representation corresponding to the user data table are generated through the linear transformation and task adaptation processing of the output layer. Therefore, by extracting more characteristic information from the data table and inputting it into the semantic convolution layer for semantic understanding, the corresponding semantic representation is obtained, so as to recommend more suitable content to users, thereby improving the accuracy and personalization of the recommendation.

[0107] In some examples, the corresponding processing process for semantic understanding through the semantic convolution layer can be as follows:

[0108] First, user information is extracted from the user data table as its semantic data, and project information, patent information and other information associated with the user are used as its auxiliary semantic information.

[0109] Next, we extract information such as the patent's abstract, title, and technical field from the patent data table. Using a supervised learning algorithm assisted by manual labeling, such as LDA or TF-IDF, we extract multiple keywords from each patent and label them. Keywords can be categorized into three types: technical keywords, scenario keywords, and application keywords, though this is not a limitation of the present invention.

[0110] For example, there is a patent about smart home, and its abstract and title are as follows:

[0111] Title: A smart home control system based on the Internet of Things

[0112] Abstract: This paper proposes an IoT-based smart home control system. This system uses sensors to collect home environment data and transmits it to a central controller via a wireless network for processing. The system automatically adjusts the operating status of home appliances based on user-defined rules and home environment data, achieving automated control of the smart home.

[0113] The extracted keywords and their categories can be as follows: technology keywords: Internet of Things, sensor, wireless network, central controller; scenario keywords: smart home; application keywords: automated control, environmental data collection, device status adjustment.

[0114] Then, BERT or other encoders can be used to further train user and keyword information to obtain the initial embedding information corresponding to the first user node u and the first keyword node i;

[0115] The initial embedding information of the first user node u and the first keyword node i is used as node features, and a multi-layer graph convolutional model is used to further enrich the semantic information of the nodes. The corresponding process may include:

[0116] Input layer: takes the node’s feature vector as the input of the graph convolutional network;

[0117] Convolutional layer: Apply convolution operations to the graph structure to extract neighborhood information of the node. Graph convolutional networks update the representation of the current node by aggregating information from neighboring nodes. This typically involves multiple convolutional layers, with each layer further fusing information from neighboring nodes.

[0118] Activation function: Apply a nonlinear activation function (such as ReLU) after the convolution operation to increase the nonlinear expression ability of the model;

[0119] Pooling layer: Use pooling operations to reduce the size of the graph while retaining important information;

[0120] Output layer: Obtain the final semantic representation of each node, that is, output the first semantic representation corresponding to the user data table and the second semantic representation corresponding to the patent data table.

[0121] Step 104: Input the first semantic representation and the second semantic representation into the heterogeneous convolutional layer to perform keyword recommendation, and obtain at least one corresponding first keyword combination;

[0122] After obtaining the corresponding semantic representation through the semantic convolution layer, the corresponding first semantic representation and the second semantic representation can be input into the heterogeneous convolution layer for keyword recommendation to obtain the corresponding keyword combination, so that the user can perform text retrieval based on the recommended keyword combination to obtain the corresponding retrieval results.

[0123] In some feasible implementations, in the process of keyword recommendation based on semantic representation, a heterogeneous graph can be constructed for the first semantic representation and the second semantic representation, wherein the heterogeneous graph has several different types of graph nodes, and the different types of graph nodes in the heterogeneous graph are connected through corresponding node relationships. Then, the node attributes corresponding to each graph node are obtained, and the weight values ​​corresponding to each node relationship are obtained. Finally, the first semantic representation and the second semantic representation are input into the heterogeneous convolutional layer, and keyword recommendation is performed based on the node attributes, node relationships and weight values ​​to obtain at least one corresponding first keyword combination.

[0124] For a heterogeneous graph, it can be composed of four different types of nodes, including a second user node, a project node, a second patent node, and a keyword node. Different types of nodes together constitute a heterogeneous graph. At the same time, different types of nodes can be connected to each other through different node relationships. For example, there are connection relationships between users and projects, projects and patents, and patents and keywords. These relationships can be constructed based on the projects in which users participate, the patents included in the projects, the keywords mentioned in the patents, and other content. The present invention does not limit this.

[0125] After constructing the corresponding heterogeneous graph, weight values ​​and node relationships can be used to perform path search in the heterogeneous graph to obtain candidate edge sequences corresponding to the first semantic representation and the second semantic representation. Then, a weight-sensitive random walk is performed on the candidate edge sequence to obtain a node sequence for the candidate edge sequence, and the total path weight corresponding to the node sequence is calculated according to the weight value. The node sequence whose total path weight is greater than or equal to a preset threshold is used as the initial meta-path, and the initial meta-path is enhanced using node attributes to obtain a semantically enhanced meta-path. The semantically enhanced meta-path is layered sampled to obtain a meta-path neighbor dictionary corresponding to each graph node. Finally, keywords are recommended for the meta-path neighbor dictionary based on the heterogeneous convolutional layer to obtain at least one first keyword combination corresponding to the first semantic representation and the second semantic representation.

[0126] It should be noted that weight-sensitive random walk refers to an algorithm that selects a walk path based on edge weights (i.e., the weight values ​​corresponding to node relationships). The higher the weight of an edge, the greater the possibility of being selected. Meta-path refers to a path model of semantic relationships between different nodes in a heterogeneous graph, such as "patent-keyword-user-patent"; semantically enhanced meta-path refers to expanding the meta-path by combining node attributes (node ​​attributes can be static feature data carried by the node itself and unrelated to the network topology. These static feature data can be used to enhance the semantic information represented by the node, and can effectively help the model better understand the intrinsic characteristics of the node during the embedding learning process) to enhance its semantic expression ability; stratified sampling refers to stratified sampling of neighbor nodes according to the meta-path type, balancing computational efficiency and semantic coverage; attention mechanism parameters can be used to learn the importance weights of different meta-paths or neighbor nodes; similarity algorithms can be used to calculate the correlation between node representations, such as cosine similarity algorithm, inner product algorithm, etc., and the present invention does not impose any restrictions on this.

[0127] Among them, in the keyword recommendation process, the parameter matrix for the meta-path neighbor dictionary can be called from the heterogeneous convolutional layer first, and then the parameter matrix and weight value are used to perform feature transformation on the meta-path neighbor dictionary, and the transformed output results are weighted aggregated to output the group aggregation results corresponding to the meta-path neighbor dictionary, and then the group aggregation results are calculated by the trained attention mechanism parameters, and the corresponding calculation results are output. The calculation results are feature-weighted aggregated and residually connected, and the first target semantic representation of the first semantic representation and the second target semantic representation corresponding to the second semantic representation are output. Then, a similarity algorithm for the first target semantic representation and the second target semantic representation is obtained, and finally, the similarity algorithm is used to recommend keywords for the first target semantic representation and the second target semantic representation to generate at least one corresponding first keyword combination.

[0128] In the specific implementation, during the process of path discovery and meta-path generation, in the heterogeneous graph, possible paths are searched based on edge weights and node relationships to obtain a candidate edge sequence connecting the "first semantic representation" (such as patents) and the "second semantic representation" (such as users). Then, the candidate edge sequence can be converted into a node sequence through weight-sensitive random walks, and the total path weight can be calculated (such as by accumulating edge weights). After obtaining the total path weight, paths with weights greater than or equal to a preset threshold can be screened out as initial meta-paths. Then, the initial meta-path is enhanced with node attributes (such as patent titles, user information, etc.) to generate a more semantically rich meta-path (such as "Patent A → Keyword 'AI' → User B → Patent C").

[0129] In the process of neighbor sampling and feature aggregation, semantically enhanced meta-paths are hierarchically sampled, and a meta-path neighbor dictionary is constructed for each node (such as "Patent A's neighbors: {'keyword':['AI', 'blockchain'], 'user':['user B']}"). The parameter matrix is ​​called through the heterogeneous convolutional layer, and the neighbor dictionary is subjected to feature transformation (such as linear projection) in combination with the edge weights. Then, weighted aggregation is used to obtain the grouping results (such as "keyword group aggregation results" and "user group aggregation results").

[0130] In the process of attention mechanism and keyword recommendation, the pre-trained attention mechanism can be used to calculate the weights of different meta-paths or neighbors (such as "the attention score of the keyword 'AI' is higher than that of 'blockchain'"), superimpose residual connections after weighted aggregation, output the enhanced target semantic representation, and calculate the similarity of the target representations of "patent" and "user" (such as cosine value, etc.). Based on the similarity calculation results, the keyword combination with the highest similarity (such as "AI, machine learning") is recommended.

[0131] In some examples, the corresponding processing process for keyword recommendation through heterogeneous convolutional layers can be as follows:

[0132] 1. Constructing a heterogeneous graph:

[0133] First, we define four types of nodes: users, projects, patents, and keywords. These nodes together form a heterogeneous graph, where different types of nodes are connected through different relationships. Connections exist between users and projects, projects and patents, and patents and keywords. These relationships can be based on projects in which users participate, patents included in projects, keywords mentioned in patents, and so on.

[0134] 2. Initial embedding acquisition:

[0135] The user and keyword embeddings obtained by the semantic convolution layer are used as the initial embeddings of the collaborative convolution layer to better reflect the position and relationship of nodes in the heterogeneous graph. At the same time, the attribute information of the node is input into the model as a feature so that this additional information can be considered in the embedding process.

[0136] 3. Introducing weighted edges:

[0137] In order to distinguish the significance of different edges, a weight is assigned to each edge according to specific weight rules. These rules are based on considerations at the company level, project level, and individual level. Company level: For example, if a company's patent layout strategy is a patent wall of the same type, then the weight of keywords similar to the user's existing keywords will be increased. Project level: If a keyword is mentioned in the project's attribute project difficulty, and the keyword also exists in the patent associated with the project, the weight of this path will be increased. Individual level: If a user's patent has completed the transformation of results because of a certain keyword, the weight of this path will be increased. In a heterogeneous graph, weights are automatically assigned to each edge according to these rules, and eventually a weighted heterogeneous graph will be formed.

[0138] For the above-mentioned weight rules, the edge weight (i.e., the weight value corresponding to the node relationship) can be dynamically adjusted according to the degree of association between the keywords and corporate information, project information, personal information, etc. At the same time, it can also be adjusted according to time decay (i.e., the longer the time is from the current time, the lower the weight), the result conversion multiplier (for example, if the keyword exists in a commercialized patent, the weight value is doubled; if the keyword does not exist in a commercialized patent, the weight value remains unchanged) and strategic matching (i.e., when matching the research direction of the enterprise, the weight value can be multiplied by 1.5 times, etc.), etc. In actual applications, it can be adjusted according to actual needs, and the present invention does not impose any restrictions on this.

[0139] Automatically learning and defining meta-paths: In heterogeneous graphs, a meta-path is a sequence connecting nodes of different types. For example, user-project-patent-keyword could be a meaningful meta-path. Based on weighting rules and node attribute information, the model automatically learns and defines meaningful meta-paths. Heterogeneous graph GCN models such as R-GCNh or HetGNN are then used to learn representations of user and keyword nodes that combine semantic and collaborative information.

[0140] 4. Recommendation:

[0141] Finally, the top n keyword combinations are recommended to each user through cosine similarity or other similarity algorithms. Each combination can contain at least one technical keyword, scenario keyword, and application keyword.

[0142] For example, assuming that we want to recommend multi-dimensional related keywords for "AI-based medical image analysis patents" based on the data table, the corresponding process may include:

[0143] 1. Heterogeneous graph construction and path discovery:

[0144] Node Type:

[0145] Patent P: A method for detecting lung cancer in CT images based on deep learning

[0146] Technical Keywords: K1 ("convolutional neural network"), K2 ("attention mechanism")

[0147] Scenario keywords: S1 (“medical imaging”), S2 (“tertiary hospitals”)

[0148] Application keywords: A1 (“early screening for lung cancer”), A2 (“remote diagnosis”)

[0149] User U: R&D director of a medical AI company

[0150] The edge relationships and corresponding weights in the heterogeneous graph can be:

[0151] P→K1 (technology, weight = 0.9), P→S1 (scenario, weight = 0.8), P→A1 (application, weight = 0.85), K1→U (user focus on technology, weight = 0.7), S1→U (user demand scenario, weight = 0.75)

[0152] The process of discovering candidate paths: The following paths can be obtained through random walks:

[0153] P→K1→U→S1→P (Technology→User→Scenario→Patent, total weight = 2.15)

[0154] P→A1→U→K2→P (application→user→technology→patent, total weight = 2.0)

[0155] 2. Semantic enhancement and meta-path generation:

[0156] Initial meta path:

[0157] Path 1: Patent → Technology → User → Scenario → Patent

[0158] Path 2: Patent → Application → User → Technology → Patent

[0159] Attribute enhancement:

[0160] Extending path semantics with patent attributes ("deep learning", "CT imaging"):

[0161] Enhancement Path 1:

[0162] “Patent (deep learning + CT imaging) → Technology (CNN) → User (medical AI company) → Scenario (tertiary hospitals + medical imaging) → Patent”

[0163] Enhancement Path 2:

[0164] “Patent (lung cancer detection) → Application (early screening) → User (R&D manager) → Technology (attention mechanism) → Patent”

[0165] 3. Neighbor aggregation and attention calculation:

[0166] Patent P's metapath neighbor dictionary:

[0167] {

[0168] "Technology": ["Convolutional Neural Network", "Attention Mechanism"],

[0169] "Scene": ["Medical Imaging", "Grade 3 Hospital"],

[0170] "Application": ["Early screening for lung cancer", "Remote diagnosis"]

[0171] }

[0172] Heterogeneous convolutional layer processing:

[0173] Technology group aggregation: CNN (weight 0.6) + attention mechanism (weight 0.4)

[0174] Scenario group aggregation: medical imaging (weight 0.7) + tertiary hospitals (weight 0.3)

[0175] Application group aggregation: lung cancer screening (weight 0.8) + remote diagnosis (weight 0.2)

[0176] Attention mechanism output:

[0177] Importance of technical keywords: CNN > Attention mechanism

[0178] Importance of scenario keywords: Medical imaging > Grade 3 hospitals

[0179] Application keyword importance: lung cancer screening > remote diagnosis

[0180] 4. Multi-dimensional keyword recommendation

[0181] Target semantic representation similarity calculation:

[0182] The similarity between the representation of patent P and user U is:

[0183] Technology matching degree: 0.9 (CNN)

[0184] Scene matching degree: 0.85 (medical imaging)

[0185] Application matching degree: 0.88 (lung cancer screening)

[0186] Final recommended keyword combination:

[0187] Technical keywords: Convolutional Neural Network (CNN), Attention Mechanism

[0188] Scenario keywords: medical imaging, tertiary hospitals

[0189] Application keywords: Early screening of lung cancer, remote diagnosis

[0190] In the above process, a weighted heterogeneous graph is used. By defining different rules to assign weights to the edges in the heterogeneous graph, the relationship strength between different nodes can be described more accurately to reflect multiple factors such as the similarity, association strength, and interaction frequency between nodes, thereby improving the accuracy of the recommendation results.

[0191] Step 105: Input the first keyword combination into the natural language model for keyword recommendation to obtain at least one corresponding second keyword combination;

[0192] After obtaining the corresponding first keyword combination through the heterogeneous convolutional recommendation model, the first keyword combination can be further processed using a natural language model to obtain at least one related and different second keyword combination, thereby generating a new keyword combination based on the recommended keyword combination by using a natural language model, providing users with more diverse choices and reducing the probability of users falling into an information cocoon.

[0193] Among them, the natural language model can be a large model obtained by training using data such as patent titles and abstracts as a corpus. By inputting the first keyword combination into the natural language model, the natural language model can generate a new set of keyword combinations based on learning experience, thereby obtaining two different sets of keyword combinations.

[0194] For example, in the process of keyword recommendation, on the one hand, historical data can be processed according to the heterogeneous convolutional recommendation model to obtain the first keyword combination, and on the other hand, the first keyword combination can be input into the natural language model to obtain the second keyword combination. Through generative recommendation, new keywords that are related to but not exactly the same as the original list are generated, and these new keywords are used as additional recommendation options to provide users with more diverse choices and reduce the probability of users falling into information cocoons.

[0195] Step 106: Perform text search according to the first keyword combination and the second keyword combination, and output the corresponding target patent text.

[0196] After obtaining the first keyword combination and the second keyword combination through the above process, text retrieval can be performed based on the first keyword combination and the second keyword combination, and the corresponding target patent text can be output based on the retrieval results, thereby performing keyword recommendation processing through the semantic convolution layer and the heterogeneous convolution layer, making the information richer and the recommendation results more accurate. At the same time, by using the natural language model to generate new keyword combinations based on the recommended keyword combinations, more diverse choices are provided to users, reducing the probability of users falling into information cocoons, and text retrieval based on rich keyword combinations effectively ensures the comprehensiveness and accuracy of the retrieval results.

[0197] In some examples, various keywords involved in the first keyword combination and the second keyword combination recommended to the user (such as technical keywords, scenario keywords, and application keywords, etc.) can be input into the language model as prompt words to assist the model in generating relevant patent titles and abstracts that combine technical information, scenario information, and application information. Based on the corresponding text recommendations, it can effectively provide users with richer, more comprehensive, and accurate text content, ensuring the recommendation effect, so that users can obtain technical content related to the research direction based on the generated patent title and abstract and other text information, thereby assisting users to better understand the technical content of the relevant technical direction.

[0198] It should be noted that the embodiments of the present invention include but are not limited to the above examples. It is understandable that those skilled in the art can also make settings according to actual needs under the guidance of the ideas of the embodiments of the present invention, and the present invention does not limit this.

[0199] In an embodiment of the present invention, when a user wants to search for relevant text, a data table is obtained, which includes at least a patent data table and a user data table. Then, a heterogeneous graph convolution recommendation model and a natural language model are obtained. The heterogeneous graph convolution recommendation model includes at least a semantic convolution layer and a heterogeneous convolution layer. Then, the patent data table and the user data table are input into the semantic convolution layer for semantic extraction, and a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table are obtained. The first semantic representation and the second semantic representation are input into the heterogeneous convolution layer for keyword recommendation, and at least one corresponding first keyword combination is obtained. The first keyword combination is input into the natural language model for keyword recommendation, and at least one corresponding second keyword combination is obtained. Finally, text retrieval is performed according to the first keyword combination and the second keyword combination, and the corresponding target patent text is output. Thus, keyword recommendation processing is performed through the semantic convolution layer and the heterogeneous convolution layer, making the information richer and the recommendation results more accurate. At the same time, by using the natural language model to generate new keyword combinations based on the recommended keyword combinations, more diverse choices are provided to users, reducing the probability of users falling into information cocoons, and text retrieval based on rich keyword combinations effectively ensures the comprehensiveness and accuracy of the retrieval results.

[0200] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the following examples are provided for illustrative purposes:

[0201] As an example, see Figure 2 , which shows a schematic diagram of the recommendation process provided in an embodiment of the present invention, which may include at least the following processes:

[0202] For the process of data collection and data storage, patent data can be obtained, and information such as abstracts, titles, technical fields, and conversion status can be broken down and stored in the patent data table; information such as titles, research contents, key technologies, and project difficulties of projects within the enterprise can be obtained and stored in the project data table; information such as the professions, job positions, research directions, and results information of users within the enterprise can be obtained and stored in the user data table; information such as the main business and product layout of the enterprise can be obtained and stored in the enterprise data table.

[0203] After obtaining the corresponding data content, on the one hand, the natural language model can be trained based on data such as the patent title and abstract as a corpus; on the other hand, it can be used for data analysis and keyword recommendations can be made through a recommendation algorithm (a heterogeneous graph convolutional recommendation model combining semantic structure and collaborative structure) to obtain a list of top n ternary keyword candidates recommended based on historical data.

[0204] After obtaining the top n triple keyword candidate lists, they can be input into a natural language model for keyword expansion to generate a recommended top n triple keyword candidate list.

[0205] Patent texts are searched using the keywords of the two groups of topn ternary keyword candidate lists obtained above, and corresponding patent titles and abstract lists are output based on the search results, thereby recommending keywords through semantic structures and heterogeneous structures, making the information richer and the recommendation results more accurate. At the same time, new keyword combinations are generated based on the recommended keyword combinations using natural language models, providing users with more diverse choices and reducing the probability of users falling into information cocoons. Text retrieval is performed based on rich keyword combinations, effectively ensuring the comprehensiveness and accuracy of the search results.

[0206] For example, refer to Figure 3 , which shows a schematic diagram of an application scenario provided in an embodiment of the present invention. The heterogeneous convolutional recommendation model may include a semantic module (graph convolution aggregation) and a collaborative module (weighted heterogeneous graph convolution), etc. The heterogeneous convolutional recommendation model can output a corresponding keyword candidate list based on the constructed data table, such as:

[0207] 1. Smart home, Internet of Things technology, and intelligent control;

[0208] 2. Online education, cloud computing platforms, and distance learning;

[0209] 3. Autonomous driving, deep learning algorithms, and vehicle navigation;

[0210] 4. Smart agriculture, IoT sensors, precision planting, etc.

[0211] Next, the top n keyword combinations in these keyword candidate lists can be input into the natural language model for keyword expansion to obtain related and different keyword candidate lists, such as:

[0212] 1. Smart home appliances, sensor interconnection technology, and scene linkage;

[0213] 2. Online classroom, cloud service architecture, asynchronous teaching;

[0214] 3. Unmanned driving, neural network algorithms, and path planning;

[0215] 4. Digital farmland, environmental sensing equipment, crop optimization, etc.

[0216] Furthermore, based on the obtained keyword candidate list, recommendations for patent titles and patent abstracts are made, such as:

[0217] Recommended list 1:

[0218] Patent title: XXXXXXXXX;

[0219] Patent Abstract: XXXXXXXXXX;

[0220] Recommended list 2:

[0221] Patent Title: YYYYYYY;

[0222] Patent Abstract: YYYYYYYYYYYYY;

[0223] ……wait.

[0224] In the above process, keyword recommendation processing is performed through semantic convolution layers and heterogeneous convolution layers, making the information richer and the recommendation results more accurate. At the same time, by using natural language models to generate new keyword combinations based on recommended keyword combinations, users are provided with more diverse choices, reducing the probability of users falling into information cocoons. In addition, text retrieval based on rich keyword combinations effectively ensures the comprehensiveness and accuracy of the retrieval results.

[0225] It should be noted that for the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0226] Reference Figure 4 , shows a structural block diagram of a text information recommendation device provided in an embodiment of the present invention, which may specifically include the following modules:

[0227] The data acquisition module 401 is used to acquire data tables, which include at least a patent data table and a user data table;

[0228] A model acquisition module 402 is configured to acquire a heterogeneous graph convolutional recommendation model and a natural language model, wherein the heterogeneous graph convolutional recommendation model includes at least a semantic convolutional layer and a heterogeneous convolutional layer;

[0229] Semantic extraction module 403, configured to input the patent data table and the user data table into a semantic convolution layer for semantic extraction, and obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table;

[0230] A first recommendation module 404 is configured to input the first semantic representation and the second semantic representation into the heterogeneous convolutional layer for keyword recommendation to obtain at least one corresponding first keyword combination;

[0231] A second recommendation module 405 is configured to input the first keyword combination into the natural language model for keyword recommendation, and obtain at least one corresponding second keyword combination;

[0232] The text search module 406 is used to perform text search according to the first keyword combination and the second keyword combination, and output the corresponding target patent text.

[0233] In some feasible implementations, the data table further includes a project data table, and the semantic extraction module is specifically configured to:

[0234] extracting employee information from the employee data table and using the employee information as semantic data;

[0235] Extracting project information from the project data table and patent information from the patent data table, and using the project information and the patent information as auxiliary semantic data corresponding to the semantic data;

[0236] extracting a plurality of patent keywords from the patent information;

[0237] The semantic data, the auxiliary semantic data and each of the patent keywords are trained to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table.

[0238] In some feasible implementations, the semantic convolution layer is a graph convolution network, which includes at least an input layer, multiple execution convolution layers, an activation function, a pooling layer, and an output layer; wherein the graph convolution network also includes a first user node and a first patent node, and the semantic extraction module is specifically used to:

[0239] Converting the semantic data and the auxiliary semantic data into first embedding information;

[0240] Converting the plurality of patent keywords into second embedded information;

[0241] Using the first embedding vector as the node feature of the first user node and the second embedding vector as the node feature of the first patent node, obtaining a first feature vector of the patent data table and a second feature vector of the user data table;

[0242] Aggregating information of neighboring nodes by executing the convolutional layer, iteratively updating the first feature vector and the second feature vector to obtain a node representation that integrates neighborhood semantics;

[0243] Performing a nonlinear transformation on the node representation of the fused neighborhood semantics according to the activation function to obtain a transformed node representation;

[0244] The transformed node representation is reduced in dimension by the pooling layer to obtain a compressed node representation, and the first semantic representation corresponding to the patent data table and the second semantic representation corresponding to the user data table are generated by the linear transformation and task adaptation processing of the output layer.

[0245] In some feasible implementations, the employee information includes at least one of user basic information, professional background information, job data, research direction and historical achievements; the project information includes at least one of project basic information, technical content and achievement data; the patent information includes at least one of title, abstract, claim text, patent type, classification information, technical feature text, conversion data, pledge information, financing information and text-derived information.

[0246] In some feasible implementations, the first recommendation module 404 is specifically configured to:

[0247] Constructing a heterogeneous graph for the first semantic representation and the second semantic representation, wherein the heterogeneous graph includes several different types of graph nodes, and the different types of graph nodes in the heterogeneous graph are connected through corresponding node relationships;

[0248] Obtaining node attributes corresponding to each of the graph nodes;

[0249] Obtaining weight values ​​corresponding to each of the node relationships;

[0250] The first semantic representation and the second semantic representation are input into the heterogeneous convolutional layer, and keyword recommendations are performed according to the node attributes, the node relationships, and the weight values ​​to obtain at least one corresponding first keyword combination.

[0251] In some feasible implementations, the first recommendation module 404 is specifically configured to:

[0252] Performing path search in the heterogeneous graph using the weight values ​​and the node relationships to obtain candidate edge sequences corresponding to the first semantic representation and the second semantic representation;

[0253] Performing a weight-sensitive random walk on the candidate edge sequence to obtain a node sequence for the candidate edge sequence, and calculating a total path weight corresponding to the node sequence according to the weight value;

[0254] The node sequence whose total path weight is greater than or equal to the preset threshold is taken as the initial meta-path;

[0255] enhancing the initial meta-path using the node attributes to obtain a semantically enhanced meta-path;

[0256] Performing stratified sampling on the semantically enhanced meta-path to obtain a meta-path neighbor dictionary corresponding to each of the graph nodes;

[0257] Keyword recommendation is performed on the meta-path neighbor dictionary according to the heterogeneous convolutional layer to obtain at least one first keyword combination corresponding to the first semantic representation and the second semantic representation.

[0258] In some feasible implementations, the first recommendation module 404 is specifically configured to:

[0259] Recalling a parameter matrix for the meta-path neighbor dictionary from the heterogeneous convolutional layer;

[0260] Performing feature transformation on the meta-path neighbor dictionary using the parameter matrix and the weight value, performing weighted aggregation on the transformed output result, and outputting a grouped aggregation result corresponding to the meta-path neighbor dictionary;

[0261] Calculating the grouping aggregation results using trained attention mechanism parameters, outputting corresponding calculation results, performing feature weighted aggregation and residual connection on the calculation results, and outputting a first target semantic representation of the first semantic representation and a second target semantic representation corresponding to the second semantic representation;

[0262] Obtaining a similarity algorithm for the first target semantic representation and the second target semantic representation;

[0263] The similarity algorithm is used to perform keyword recommendation on the first target semantic representation and the second target semantic representation, and to generate at least one corresponding first keyword combination.

[0264] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0265] In addition, an embodiment of the present invention further provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the various processes of the above-mentioned text information recommendation method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, they will not be described here.

[0266] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program implements the various processes of the aforementioned text information recommendation method embodiment and achieves the same technical effects. To avoid repetition, the description is omitted here. The computer-readable storage medium may be, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0267] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0268] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, apparatuses, or computer program products. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, EEPROM, Flash, and eMMC, etc.) containing computer-usable program code.

[0269] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0270] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0271] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0272] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic creative concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0273] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0274] The above is a detailed introduction to a method for recommending text information and a device for recommending text information provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only intended to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A method for recommending text information, characterized in that: include: Obtaining a data table, wherein the data table includes at least a patent data table and a user data table; Obtaining a heterogeneous graph convolutional recommendation model and a natural language model, wherein the heterogeneous graph convolutional recommendation model includes at least a semantic convolutional layer and a heterogeneous convolutional layer; Inputting the patent data table and the user data table into a semantic convolution layer for semantic extraction to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table; Inputting the first semantic representation and the second semantic representation into the heterogeneous convolutional layer for keyword recommendation to obtain at least one corresponding first keyword combination; Inputting the first keyword combination into the natural language model for keyword recommendation to obtain at least one corresponding second keyword combination; Perform text search according to the first keyword combination and the second keyword combination, and output the corresponding target patent text.

2. The method according to claim 1, characterized in that The data table further includes a project data table. Inputting the patent data table and the user data table into a semantic convolution layer for semantic extraction to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table includes: extracting employee information from the employee data table and using the employee information as semantic data; Extracting project information from the project data table and patent information from the patent data table, and using the project information and the patent information as auxiliary semantic data corresponding to the semantic data; extracting a plurality of patent keywords from the patent information; The semantic data, the auxiliary semantic data and each of the patent keywords are trained to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table.

3. The method according to claim 2, characterized in that The semantic convolution layer is a graph convolution network, which includes at least an input layer, multiple execution convolution layers, an activation function, a pooling layer, and an output layer; wherein the graph convolution network also includes a first user node and a first patent node, and the training of the semantic data, the auxiliary semantic data, and each of the patent keywords to obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table includes: Converting the semantic data and the auxiliary semantic data into first embedding information; Converting the plurality of patent keywords into second embedded information; Using the first embedding vector as the node feature of the first user node and the second embedding vector as the node feature of the first patent node, obtaining a first feature vector of the patent data table and a second feature vector of the user data table; Aggregating information of neighboring nodes by executing the convolutional layer, iteratively updating the first feature vector and the second feature vector to obtain a node representation that integrates neighborhood semantics; Performing a nonlinear transformation on the node representation of the fused neighborhood semantics according to the activation function to obtain a transformed node representation; The transformed node representation is reduced in dimension by the pooling layer to obtain a compressed node representation, and the first semantic representation corresponding to the patent data table and the second semantic representation corresponding to the user data table are generated by the linear transformation and task adaptation processing of the output layer.

4. The method according to claim 2 or 3, characterized in that The employee information includes at least one of user basic information, professional background information, job data, research direction and historical achievements; the project information includes at least one of project basic information, technical content and achievement data; the patent information includes at least one of title, abstract, claim text, patent type, classification information, technical feature text, conversion data, pledge information, financing information and text-derived information.

5. The method according to claim 1, characterized in that Inputting the first semantic representation and the second semantic representation into the heterogeneous convolutional layer for keyword recommendation to obtain at least one corresponding first keyword combination includes: Constructing a heterogeneous graph for the first semantic representation and the second semantic representation, wherein the heterogeneous graph includes several different types of graph nodes, and the different types of graph nodes in the heterogeneous graph are connected through corresponding node relationships; Obtaining node attributes corresponding to each of the graph nodes; Obtaining weight values ​​corresponding to each of the node relationships; The first semantic representation and the second semantic representation are input into the heterogeneous convolutional layer, and keyword recommendations are performed according to the node attributes, the node relationships, and the weight values ​​to obtain at least one corresponding first keyword combination.

6. The method according to claim 5, characterized in that The step of inputting the first semantic representation and the second semantic representation into the heterogeneous convolutional layer, performing keyword recommendation based on the node attributes, the node relationships, and the weight values, and obtaining at least one corresponding first keyword combination includes: Performing path search in the heterogeneous graph using the weight values ​​and the node relationships to obtain candidate edge sequences corresponding to the first semantic representation and the second semantic representation; Performing a weight-sensitive random walk on the candidate edge sequence to obtain a node sequence for the candidate edge sequence, and calculating a total path weight corresponding to the node sequence according to the weight value; The node sequence whose total path weight is greater than or equal to the preset threshold is taken as the initial meta-path; enhancing the initial meta-path using the node attributes to obtain a semantically enhanced meta-path; Performing stratified sampling on the semantically enhanced meta-path to obtain a meta-path neighbor dictionary corresponding to each of the graph nodes; Keyword recommendation is performed on the meta-path neighbor dictionary according to the heterogeneous convolutional layer to obtain at least one first keyword combination corresponding to the first semantic representation and the second semantic representation.

7. The method according to claim 6, characterized in that The performing keyword recommendation on the meta-path neighbor dictionary according to the heterogeneous convolutional layer to obtain at least one first keyword combination corresponding to the first semantic representation and the second semantic representation includes: Recalling a parameter matrix for the meta-path neighbor dictionary from the heterogeneous convolutional layer; Performing feature transformation on the meta-path neighbor dictionary using the parameter matrix and the weight value, performing weighted aggregation on the transformed output result, and outputting a grouped aggregation result corresponding to the meta-path neighbor dictionary; Calculating the grouping aggregation results using trained attention mechanism parameters, outputting corresponding calculation results, performing feature weighted aggregation and residual connection on the calculation results, and outputting a first target semantic representation of the first semantic representation and a second target semantic representation corresponding to the second semantic representation; Obtaining a similarity algorithm for the first target semantic representation and the second target semantic representation; The similarity algorithm is used to perform keyword recommendation on the first target semantic representation and the second target semantic representation, and to generate at least one corresponding first keyword combination.

8. A device for recommending text information, characterized in that: include: A data acquisition module, configured to acquire a data table, wherein the data table includes at least a patent data table and a user data table; A model acquisition module is used to obtain a heterogeneous graph convolutional recommendation model and a natural language model, wherein the heterogeneous graph convolutional recommendation model includes at least a semantic convolutional layer and a heterogeneous convolutional layer; a semantic extraction module, configured to input the patent data table and the user data table into a semantic convolution layer for semantic extraction, and obtain a first semantic representation corresponding to the patent data table and a second semantic representation corresponding to the user data table; A first recommendation module, configured to input the first semantic representation and the second semantic representation into the heterogeneous convolutional layer for keyword recommendation to obtain at least one corresponding first keyword combination; a second recommendation module, configured to input the first keyword combination into the natural language model for keyword recommendation, and obtain at least one corresponding second keyword combination; A text retrieval module is used to perform text retrieval according to the first keyword combination and the second keyword combination, and output the corresponding target patent text.

9. An electronic device, characterized in that: comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; The memory is used to store computer programs; The processor is configured to implement the method according to any one of claims 1 to 7 when executing a program stored in the memory.

10. A computer-readable storage medium having instructions stored thereon, which, when executed by one or more processors, cause the processors to perform the method according to any one of claims 1 to 7.

Images