Technical opportunity recommendation method and device, equipment, storage medium and computer product

Abstract

The invention discloses a technology opportunity recommendation method and device, equipment, a storage medium and a computer product, and relates to the technical field of technology opportunity recognition, and the method comprises the steps: obtaining a target domain patent data set and a classification number co-occurrence network, constructing a reference sample pool, and constructing a candidate sample pool, respectively carrying out feature extraction processing on the reference technical sub-graph and the candidate technical sub-graph to obtain reference technical representation information corresponding to the reference technical sub-graph and candidate technical representation information corresponding to the candidate technical sub-graph, determining a comprehensive similarity between the candidate technical representation information and the reference sample pool for each piece of candidate technical representation information, and carrying out feature extraction processing on the reference technical sub-graph and the candidate technical representation information; and determining a technical opportunity recommendation result based on a comprehensive similarity comparison condition among all the candidate technical representation information. According to the method, potential frontier technology fusion schemes can be more comprehensively mined, and the perspectiveness of technology opportunity prediction and the interpretability of recommendation results are enhanced.

Description

Technical Field

[0001] The present application relates to the technical field of technology opportunity identification, and in particular to a technology opportunity recommendation method, apparatus, device, storage medium, and computer product. Background Art

[0002] Technological opportunities serve as a compass for businesses to adapt to technological change and gain market advantage. Technology Opportunity Discovery (TOD) is a systematic process for identifying potential technological opportunities. It extracts or mines valuable information from objective data to identify technological opportunities, thus safeguarding the planning and development of innovative themes and strategies. In practical applications, the application of technology opportunity identification provides concrete support for the design of technical solutions. For example, in the field of biomedical engineering, by identifying opportunities for the integration of 5G communication and sensor technologies, wearable medical devices have been developed, enabling real-time health data transmission and telemedicine capabilities. These applications demonstrate that technology opportunity identification can provide practical solutions to engineering problems through the integration of cross-disciplinary technologies.

[0003] In related technologies, on the one hand, the use of natural language processing (NLP) technology to process patent titles, abstracts, and claims, and the analysis of technological trends or potential technological opportunities from patent texts utilizes semantic information but ignores some other potential information contained in the patent.

[0004] On the other hand, methods based on relational data also have limitations in TOD research centered around the Cooperative Patent Classification (CPC) co-occurrence network. These methods focus on predicting future connections between CPC node pairs. While they can monitor trends in technological convergence, they are limited to revealing the innovation patterns of binary elements. However, patent data shows that technological convergence often involves the synergy of multiple CPC subclasses. For example, according to personal statistics, 60.36% of US artificial intelligence patents between 2013 and 2023 included three or more CPC subclasses. Furthermore, previous methods relied on the static properties of global co-occurrence networks, ignoring the dynamic semantics of CPC nodes in different technological contexts.

[0005] Therefore, there is an urgent need to develop new methods that break through the binary correlation paradigm and achieve more comprehensive technology opportunity prediction by identifying multi-factor technology fusion solutions and integrating the local and global representations of CPC nodes. Summary of the Invention

[0006] The main purpose of this application is to provide a technology opportunity recommendation method, device, equipment, storage medium and computer product, aiming to solve the technical problems of insufficient comprehensiveness and accuracy in the identification of technology opportunities in related technologies.

[0007] To achieve the above objectives, this application proposes a technology opportunity recommendation method, comprising: Obtain the target field patent dataset and classification code co-occurrence network; the classification code co-occurrence network is constructed by the classification code co-occurrence relationship corresponding to each patent in the comprehensive field patent dataset; For each target field patent in the target field patent dataset, based on the classification code co-occurrence network, the first connected subgraph containing the target field patent classification code is determined to construct a reference sample pool; the reference sample pool includes a reference technology subgraph obtained by reconstructing the first connected subgraph through a spanning tree; A classification code is designated as a candidate classification code from the target field patent dataset, and a second connected subgraph containing the candidate classification code is determined based on the classification code co-occurrence network to construct a candidate sample pool; the candidate sample pool includes a candidate technology subgraph obtained by reconstructing the second connected subgraph through a spanning tree. Performing feature extraction processing on the reference technology subgraph and the candidate technology subgraph respectively to obtain reference technology representation information corresponding to the reference technology subgraph and candidate technology representation information corresponding to the candidate technology subgraph; For each candidate technology representation information, determine the comprehensive similarity between the candidate technology representation information and the reference sample pool; Based on the comprehensive similarity comparison between all candidate technology representation information, the technology opportunity recommendation results are determined.

[0008] In one embodiment, for each target field patent in the target field patent dataset, extracting a first connected subgraph containing the target field patent classification number from the classification number co-occurrence network, and constructing a reference sample pool includes: Abstracting the target patent classification number contained in the target field patents into a first network node; the first network node includes node feature information composed of node embeddings of the corresponding classification number in the classification number co-occurrence network; Add edges between any two network node pairs to obtain a first connected subgraph; Based on the node feature information, edge weight information of all edges of the first connected subgraph is calculated to determine the association strength between the first network nodes in the first connected subgraph; wherein the association strength is negatively correlated with the edge weight information; Based on the association strength, the first connected subgraph is reconstructed using the minimum spanning tree algorithm to obtain the reference technology subgraph; A reference sample pool is constructed based on the reference technology subgraphs corresponding to all patents in the target field.

[0009] In one embodiment, a classification code is designated as a candidate classification code from a target field patent dataset, and a second connected subgraph containing the candidate classification code is extracted from a classification code co-occurrence network. The steps of constructing a candidate sample pool include: A depth-first search algorithm is used in the classification code co-occurrence network to obtain a sample sequence of a specified length containing candidate classification codes as a second network node sequence; Adding an edge between any pair of network nodes in the second network node sequence to obtain a second connected subgraph; Calculating edge weight information of all edges of the second connected subgraph based on the node feature information of the second network nodes, and determining the association strength between the second network nodes in the second connected subgraph; Based on the association strength, the second connected subgraph is reconstructed using the minimum spanning tree algorithm to obtain the candidate technology subgraph; A candidate sample pool is constructed based on the candidate technology subgraphs corresponding to all patents in the target field.

[0010] In one embodiment, the steps of performing feature extraction processing on the reference technology subgraph and the candidate technology subgraph respectively to obtain reference technology characterization information corresponding to the reference technology subgraph and candidate technology characterization information corresponding to the candidate technology subgraph include: The node feature information and node position information of the reference technology subgraph and the candidate technology subgraph are respectively input into the technology representation extraction model to obtain the reference technology representation information corresponding to the reference technology subgraph and the candidate technology representation information corresponding to the candidate technology subgraph.

[0011] Among them, the technical representation extraction model is obtained by training positive samples composed of technical sub-graphs corresponding to existing patents and negative samples composed of randomly generated technical sub-graphs; the original feature extraction model includes parallel graph neural network units and Transformer units, as well as image-text feature fusion units in series with the graph neural network units and Transformer units.

[0012] In one embodiment, for each candidate technology characterization information, the step of determining the comprehensive similarity between the candidate technology characterization information and the reference sample pool includes: For each candidate technology representation information, respectively confirm the cosine similarity between the candidate technology representation information and each reference representation information in the reference sample pool; The maximum cosine similarity is taken as the comprehensive similarity between the candidate technology representation information and the reference sample pool.

[0013] In one embodiment, the step of determining a technology opportunity recommendation result based on a comprehensive similarity comparison between all candidate technology representation information includes: Determine the comprehensive similarity between all candidate technology characterization information in the candidate sample pool and the reference sample pool; Based on all comprehensive similarities, a preset number of candidate technology subgraphs are selected from high to low to determine as technology opportunity recommendation results.

[0014] In addition, to achieve the above-mentioned purpose, the present application also provides a technology opportunity recommendation device, which includes: An acquisition module is used to obtain a target field patent dataset and a classification code co-occurrence network; wherein the classification code co-occurrence network is constructed by the classification code co-occurrence relationship corresponding to each patent in the comprehensive field patent dataset; A reference sample pool construction module is used to determine, for each target field patent in the target field patent dataset, based on the classification code co-occurrence network, a first connected subgraph containing the target field patent classification code, and construct a reference sample pool; the reference sample pool includes a reference technology subgraph obtained by performing spanning tree reconstruction on the first connected subgraph; The candidate sample pool construction module is used to designate a classification code from the target field patent dataset as a candidate classification code, determine a second connected subgraph containing the candidate classification code based on the classification code co-occurrence network, and construct a candidate sample pool; the candidate sample pool includes a candidate technology subgraph obtained by reconstructing the second connected subgraph through a spanning tree; A feature extraction module is used to perform feature extraction processing on the reference technology subgraph and the candidate technology subgraph respectively to obtain reference technology representation information corresponding to the reference technology subgraph and candidate technology representation information corresponding to the candidate technology subgraph; A similarity calculation module is used to determine the comprehensive similarity between each candidate technology representation information and the reference sample pool; The recommendation module is used to determine the technology opportunity recommendation results based on the comprehensive similarity comparison between the representation information of all candidate technologies.

[0015] In addition, to achieve the above-mentioned purpose, the present application also provides a technology opportunity recommendation device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the above-mentioned technology opportunity recommendation method.

[0016] In addition, to achieve the above-mentioned purpose, the present application also provides a storage medium, which is a computer-readable storage medium. A computer program is stored on the storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned technology opportunity recommendation method are implemented.

[0017] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the above-mentioned technology opportunity recommendation method.

[0018] One or more technical solutions proposed in this application have at least the following technical effects: This application proposes a technology opportunity recommendation method that breaks through the limitations of traditional methods based only on CPC binary co-occurrence relationships. It can identify complex technology fusion patterns containing multiple CPC subclasses, and integrates technical means such as node embedding features, minimum spanning tree construction and graph neural networks to achieve joint modeling of the local structure and global semantics of CPC nodes. By evaluating the similarity between candidate technology subgraphs and reference technology subgraphs, it improves the accuracy and practicality of multi-factor technology opportunity identification, helps to more comprehensively explore potential cutting-edge technology fusion solutions, and enhances the foresight of technology opportunity prediction and the interpretability of recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0021] Figure 1 This is a flowchart of the first embodiment of the technical opportunity recommendation method of this application.

[0022] Figure 2 Schematic diagram of the framework for reconstructing the technical subgraph.

[0023] Figure 3 Schematic diagram of the framework extracted for technology characterization.

[0024] Figure 4 Schematic diagram of the visualization process for data partitioning and negative sample generation.

[0025] Figure 5 Recommend a process flow diagram for the technology opportunity.

[0026] Figure 6 Recommend device diagrams for technology opportunities.

[0027] Figure 7 Recommend equipment diagrams for technology opportunities.

[0028] The purpose, features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0029] It should be understood that the specific embodiments described herein are merely used to explain the technical solutions of the present application and are not intended to limit the present application.

[0030] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0031] Based on the problems mentioned in the background technology, the embodiment of the present application provides a method for recommending technology opportunities. Figure 1 , Figure 1 This is a flowchart of the first embodiment of the technical opportunity recommendation method of this application.

[0032] In the embodiment of the present application, the design idea of ​​the technology opportunity recommendation method is: by extracting classification numbers from the classification number co-occurrence network to construct a connected subgraph, and using the minimum spanning tree algorithm to reconstruct the reference and candidate technology subgraphs, the knowledge inheritance of the co-occurrence network and the graph reconstruction based on node correlation are realized, and the representation information of the technology subgraph is further used to calculate the comprehensive similarity between the candidate sample and the reference sample pool, so as to identify the candidate technology subgraph that is close to the existing technology fusion trend as the technology opportunity recommendation result, and realize the multi-factor driven intelligent recommendation of technology opportunities.

[0033] Specifically, in this embodiment, the technology opportunity recommendation method includes steps S10 to S60: Step S10: Obtain the target field patent dataset and the classification code co-occurrence network; wherein the classification code co-occurrence network is constructed by the classification code co-occurrence relationship corresponding to each patent in the comprehensive field patent dataset.

[0034] It should be noted that the target field patent data set includes patent data in the main fields involved in the recommended technological opportunities. Taking artificial intelligence as the target field as an example, the target field patent data set can be some or all patents involving the field of artificial intelligence screened from the comprehensive field patent data set. Preferably, in order to prevent data leakage, the patent data involved in the target field may also be excluded from the comprehensive field patent data set.

[0035] The comprehensive field patent dataset contains patent data from multiple fields. The patent data in the comprehensive field patent dataset can be used to construct a classification code co-occurrence network and serve as a model training sample set.

[0036] The classification number may be any type of patent classification number, for example, IPC, EPC, CPC, etc. For ease of understanding, this embodiment uses the CPC classification number as an example to illustrate the technology opportunity recommendation method of this application.

[0037] For example, we can extract US authorized patents from 2013-2023 from IncoPat, construct a search formula using CPC departmental classification codes, and select the top 2000 high-value patents with the highest citation frequency each year from each field to construct an initial comprehensive dataset. After cleaning and deduplication, the initial comprehensive dataset retains the remaining patents to form the comprehensive field patent dataset. Using "artificial intelligence" as the target field, we remove patent data related to the target field from the comprehensive patent dataset. We then collect patents from IncoPat that contain "artificial intelligence" in their titles, abstracts, and claims as the target field dataset.

[0038] The classification code co-occurrence network is a global co-occurrence network based on the CPC co-occurrence relationship of all patents. CPC co-occurrence reflects the situation where a pair of classification code codes appear at the same time in any patent. In the classification code co-occurrence network, the nodes are CPC classification codes. There are edges between node pairs with co-occurrence relationships, and vice versa.

[0039] Step S20: For each target field patent in the target field patent dataset, based on the classification code co-occurrence network, determine the first connected subgraph containing the target field patent classification code, and construct a reference sample pool; the reference sample pool includes a reference technology subgraph obtained by reconstructing the first connected subgraph through a spanning tree.

[0040] Step S30, designate a classification code from the target field patent dataset as a candidate classification code, determine a second connected subgraph containing the candidate classification code based on the classification code co-occurrence network, and construct a candidate sample pool; the candidate sample pool includes a candidate technology subgraph obtained by reconstructing the second connected subgraph through a spanning tree.

[0041] It should be noted that, considering the problems encountered in the background technology: the current relational data method is only limited to revealing the innovation rules of binary elements. This embodiment proposes a composition method to model technology opportunities containing multiple CPC classification numbers as a graph with a clear structure - a technology subgraph. Each node in the technology subgraph represents a CPC classification feature, and technology subgraphs containing different CPC nodes represent different technology opportunities.

[0042] The reference technology subgraph, based on the classification codes of patents in a single target field, serves as a comparison benchmark for potential technology opportunities. The reference sample pool, a collection of multiple reference technology subgraphs in the target field, is used to compare with the candidate technology subgraph to assess the potential of technology opportunities.

[0043] Based on a classification number involved in the target field as a candidate classification number, the candidate technology subgraph is composed of a classification number sample sequence containing the candidate classification number extracted from the classification number co-occurrence network. Each candidate technology subgraph can represent a potential new technology opportunity. For example, in order to recommend the most suitable technology opportunity to the target field, the CPC subclass with the highest frequency in the target field (such as G05D) can be counted as a candidate classification number, and a depth-first search algorithm can be performed on the classification number co-occurrence network with this subclass (such as G05D) as the starting point to extract a node sequence of a specified length to construct a candidate technology subgraph. The candidate sample pool represents a collection of candidate technology subgraphs, which contains potential technology opportunities.

[0044] It should also be noted that the technical subgraph is obtained by reconstructing the spanning tree of the connected subgraph. The connected subgraph uses the classification numbers in the co-occurrence network as nodes, and each node pair is connected by an edge. For the first connected subgraph, the classification numbers of its nodes are the same as the classification numbers contained in the corresponding single target field patent. For the second connected subgraph, the classification numbers of its nodes include candidate classification numbers specified from the target field classification numbers, and each node pair is connected by an edge. It can be understood that the classification numbers other than the candidate classification numbers in the second connected subgraph are searched from the classification number co-occurrence network based on the depth-first search algorithm. Therefore, the classification numbers other than the candidate classification numbers in the second connected subgraph may also not belong to the target field.

[0045] The so-called spanning tree reconstruction is to reconstruct the connected graph into a spanning tree, that is, the structure of the technology subgraph (reference technology subgraph or candidate technology subgraph) is a spanning tree structure.

[0046] Specifically, based on the characteristics of the classification code co-occurrence network, the connected subgraph is reconstructed into a technology subgraph, which enables the technology subgraph TS to inherit knowledge from the global network. The candidate technology subgraphs constructed based on the classification code co-occurrence network are regarded as candidate technology opportunities. The candidate technology subgraphs are classified and evaluated based on the reference sample pool to recommend high-quality technology opportunities.

[0047] In a feasible real-time manner, step S20 includes steps A10-A50: Step A10: abstract the target patent classification number contained in the target field patent into a first network node; the first network node includes node feature information composed of node embeddings of the corresponding classification number in the classification number co-occurrence network.

[0048] Step A20: adding edges between any two network node pairs to obtain a first connected subgraph.

[0049] Step A30: Calculate edge weight information of all edges of the first connected subgraph based on the node feature information, and determine the association strength between the first network nodes in the first connected subgraph; wherein the association strength is negatively correlated with the edge weight information.

[0050] Step A40 : reconstructing the first connected subgraph using a minimum spanning tree algorithm based on the association strength to obtain a reference technology subgraph.

[0051] Step A50: construct a reference sample pool based on the reference technology subgraphs corresponding to all patents in the target field.

[0052] Step S30 includes steps B10-B50: Step B10: Using a depth-first search algorithm in the classification code co-occurrence network, obtain a sample sequence of a specified length containing candidate classification codes as a second network node sequence.

[0053] Step B20: adding an edge between any pair of network nodes in the second network node sequence to obtain a second connected subgraph.

[0054] Step B30: Calculate edge weight information of all edges of the second connected subgraph based on the node feature information of the second network nodes, and determine the association strength between the second network nodes in the second connected subgraph.

[0055] Step B40: Based on the association strength, the second connected subgraph is reconstructed by a minimum spanning tree algorithm to obtain a candidate technology subgraph.

[0056] Step B50: construct a candidate sample pool based on the candidate technology subgraphs corresponding to all patents in the target field.

[0057] It should be noted that the node embeddings for the corresponding classification number in the classification number co-occurrence network can include node feature information from the classification number co-occurrence network. Inheriting these node features into the technology subgraph can further improve the effectiveness of technology opportunity recommendations. Regarding the method for obtaining node embeddings in the classification number network, for example, the Node2vec embedding algorithm can be used to learn CPC node embeddings for the classification number co-occurrence network.

[0058] It should also be noted that previous studies extracted node-related metrics or custom features based on the global co-occurrence network, and used global static attributes as one of the measurement factors for technological opportunities. This type of method ignores the dynamic semantics of CPC nodes in different technological context scenarios. Similar to the concept that individuals in social networks play different roles in different sub-networks, the characteristics of CPC nodes, in addition to reflecting the macro-technical association patterns of the global co-occurrence network, also need to be adaptively and dynamically adjusted according to the context of specific technological scenarios. Therefore, this embodiment proposes a method for constructing a technology subgraph based on the classification number co-occurrence network to achieve knowledge inheritance of the global co-occurrence network, and to use technology subgraphs to represent technological opportunities in the target field. For the construction method of the technology subgraph, refer to Figure 2 , Figure 2 Schematic diagram of the framework for reconstructing the technical subgraph.

[0059] Specifically, to construct the reference technology subgraph, we first identify the classification codes involved in the target field patent documents corresponding to the reference technology subgraph. Using the classification codes as nodes, we add edges between each node pair to form a first connected subgraph. We then obtain the node embeddings corresponding to the classification codes in the classification code co-occurrence network. The first connected subgraph and node embeddings are used as the basis for calculating edge weights. Edge weights are calculated using Formula 1: Formula 1: in, and Represents the node feature information of the two nodes corresponding to a pair of nodes in the connected subgraph, represents the cosine similarity between two nodes in a node pair, It is not difficult to see that the edge weight reflects the association strength between two nodes through cosine similarity, which maps the high similarity of the node pair to low weight, that is, the association strength is negatively correlated with the edge weight information, that is, ( , ), the higher the degree of correlation, the lower the edge weight.

[0060] It is not difficult to understand that the edge weights of each edge of the first connected subgraph can be calculated by formula 1. Furthermore, the first connected subgraph can be converted into a minimum spanning tree using a minimum spanning tree algorithm, such as the Kruska algorithm, to obtain a reference technology subgraph. After converting multiple target field patents into reference technology subgraphs, a reference sample pool is formed.

[0061] As for the construction of the candidate technology subgraph, it can be understood that it is similar to the construction method of the reference technology subgraph. After confirming the second connected subgraph containing the candidate classification number, the edge weight of the second connected subgraph is calculated according to each second connected subgraph and the corresponding node features, and then the candidate technology subgraph is obtained through the minimum spanning tree algorithm to form a candidate sample pool. I will not go into details here.

[0062] It can be understood that the technology subgraphs can be connected into a strongly correlated graph in a way that maximizes correlation under connectivity conditions, realizing the knowledge inheritance of the global co-occurrence network.

[0063] Step S40 : performing feature extraction processing on the reference technology subgraph and the candidate technology subgraph respectively to obtain reference technology characterization information corresponding to the reference technology subgraph and candidate technology characterization information corresponding to the candidate technology subgraph.

[0064] Specifically, in order to better reflect the essence of technology opportunities, this embodiment further performs characterization learning on the technology subgraph, extracts the technology representation information of the technology subgraph, and compares the candidate technology representation information with the reference technology representation information to achieve a more accurate technology opportunity identification effect.

[0065] In a feasible implementation, step S40 may include step S41: The node feature information and node position information of the reference technology subgraph and the candidate technology subgraph are respectively input into the technology representation extraction model to obtain the reference technology representation information corresponding to the reference technology subgraph and the candidate technology representation information corresponding to the candidate technology subgraph.

[0066] Among them, the technical representation extraction model is obtained by training positive samples composed of technical sub-graphs corresponding to existing patents and negative samples composed of randomly generated technical sub-graphs; the original feature extraction model includes parallel graph neural network units and Transformer units, as well as image-text feature fusion units in series with the graph neural network units and Transformer units.

[0067] It should be noted that the transformer unit is used to extract the global semantic dependency features of the reference technology subgraph or candidate technology subgraph. The graph neural network unit is used to extract the local structural features of the reference technology subgraph or candidate technology subgraph. The graph-text feature fusion module is used to fuse topological structure information and node attribute information. The topological structure information is composed of global semantic dependency features and local structural features. The node attribute information is determined based on the text attribute features of the classification number nodes contained in the reference technology subgraph or candidate technology subgraph.

[0068] It should also be noted that, in this embodiment, the technical representation extraction model is essentially a graph classification model, which is used to implement graph classification of technical subgraphs. In this process, technical representation information of technical subgraphs can be extracted, such as Figure 3 As shown, Figure 3 This is a schematic diagram of the framework for technical representation extraction. The technical representation extraction model includes parallel graph neural network units GIN and Transformer units, and a graphic-text feature fusion unit that is serial with the graph neural network unit and Transformer unit.

[0069] like Figure 3 As shown, the node features and node position encodings of the technology subgraph are summed as the initial input feature data. Each TS sample is processed through a parallel graph isomorphism network (GIN) and transformer model to extract the local structural features and global semantic dependency features of the technology subgraph, respectively. The fusion output of the parallel modules is used as the node representation of the topological structure information. The text attribute features of the nodes are then combined with the graph-text feature fusion module (GTF) to obtain a graph-level representation that fuses the two types of information. A prediction head is then used to determine the TS category. The high-dimensional feature vector of the technology subgraph before passing through the final prediction head can be used as the technology representation information of the technology subgraph.

[0070] Regarding the feature extraction part of the Transformer unit, considering that the traditional Transformer model cannot fully characterize the structural characteristics of the graph when combining token features and their position encoding, this embodiment designs an improved Transformer unit with edge-weighted attention mechanism (EGA) as its core. By strengthening the model's edge perception ability, it improves the node representation effect. This makes up for the traditional Transformer's lack of perception of graph structure.

[0071] Specifically, graph inductive bias is a priori knowledge or assumption related to structure. Considering graph inductive bias in the Transformer model helps improve the model's ability to understand graph structure. For example, in the process of learning node representation, it can increase the attention paid to strongly associated nodes. In a specific example, the edge weights of the technical subgraph encode the strength of the association between node pairs. By designing the edge weight attention adjustment factor , and use the Hadamard product to Inject the original multi-head attention mechanism to implement edge weight-guided attention calculation based on correlation strength. The modified attention weight calculation method is as follows: in represents the attention adjustment factor of node i relative to node j, is a hidden dimension, represents the edge weight, is a learnable parameter, represents the modified node representation, is the query Q, key K and value V in the attention mechanism. Attention adjustment factor By amplifying the attention scores of strongly associated nodes and reducing the attention scores of weakly associated nodes, the source node can give priority attention to nodes with strong technical associations.

[0072] Furthermore, the node attribute information of the technology subgraph includes numerical information, category information, or text information. This node attribute information can complement the topological structure information of the technology subgraph to improve the quality of technology representation. As the research object, the technology subgraph contains nodes corresponding to specific CPC subclasses. Each subclass has an officially provided standardized technical definition, namely the classification number node definition information. For example, Doc2vec is used to obtain the textual information representation of CPC and use it as node attribute information to provide a foundation for subsequent graph and text feature fusion.

[0073] This embodiment realizes the fusion of node attribute information and topological structure information through the graph-text feature fusion module GTF. Specifically, for each technical subgraph, the topological structure information and node attribute information are pooled to obtain the structural features. and node attribute characteristics , and then stack the two and process them through the self-attention mechanism to obtain the fusion structure feature and node attribute characteristics Graph features , the pseudo code of the feature fusion process is as follows: in, is the set of nodes of the technology subgraph, and Represents nodes in the technology subgraph respectively Topological structure information and node attribute information. Structural features Z struc , node attribute characteristics Z text It is the graph-level representation after pooling. represents the stacked feature matrix, It is a graph-level collaborative representation that achieves semantic-structural fusion through the self-attention mechanism.

[0074] Furthermore, the technology representation extraction model is trained by using positive samples consisting of technology subgraphs corresponding to existing patents and negative samples consisting of randomly generated technology subgraphs.

[0075] It should be noted that technology opportunity discovery is essentially the exploration of feasible technology fusion solutions. This embodiment regards the technology subgraph as a technology fusion solution. The concept of whether the technology fusion solution is feasible naturally divides the technology subgraph into two categories: positive and negative. Positive samples indicate that the fusion solution is feasible, which is equivalent to a technology opportunity, while negative samples indicate the opposite.

[0076] Understandably, patents are actually feasible technical solutions, and extracting their CPC combinations can construct positive samples. The core significance of technology opportunity discovery is to use historical data to predict future technological opportunities. Therefore, patent data needs to be divided into training sets, validation sets, and test sets in chronological order. The CPC combination of each patent can be reconstructed into a technology subgraph.

[0077] The validation set only contains technical subgraphs that did not appear in the training set, and the test set only contains technical subgraphs that did not appear in either the training or validation set. It's worth noting that the technical subgraphs are constructed based on a co-occurrence network. To eliminate future information interference, the training set data can be used to construct a classification code co-occurrence network and combined with node embedding to support the subsequent construction of technical subgraphs. The training set contains patents from a variety of fields, representing a comprehensive field patent dataset. When using the technical representation extraction model to extract representations from patents in the target field, the model trained in the comprehensive field patent training set can be reused.

[0078] Regarding the composition of positive and negative samples in the training set, specifically, the technical subgraphs constructed through existing patents are all represented as positive samples. In order for the classification model to be trained normally, a negative sample generation strategy needs to be designed. Therefore, the technical subgraphs constructed by CPC combinations that are not covered by patents are recorded as negative samples. To obtain such CPC combinations, for example, a random walk algorithm is used to obtain several node sequences on the classification number co-occurrence network, and the nodes contained in the sequence are extracted to construct negative samples. In order to improve the diversity of negative samples and the robustness of the model, hundreds of times more negative samples are generated than positive samples, and negative samples equal to the positive samples are randomly selected for training in each round of training. The visualization process of data partitioning and negative sample generation is shown in the figure. Figure 4 shown.

[0079] For example, a random walk algorithm is used to generate negative samples of technology subgraphs, consisting of a comprehensive range of patents from 2013 to 2021. Specifically, a random walk is performed on a classification code co-occurrence network constructed from patent data from 2013 to 2021 to generate node sequences of a specified length equal to the positive samples. These node sequences cannot be identical to the node sequences of the positive samples. These generated negative sample node sequences are then converted into negative sample technology subgraphs. The training set composed of positive and negative samples is used for model training. Multiple technology subgraphs from 2022 and 2023 are divided into validation and test sets, respectively. Each technology subgraph sample is independent and non-repetitive.

[0080] It is understandable that through the improvement of the traditional Transformer model and the design of the image-text fusion module, the technical representation extraction model obtained after sample training has a strong perception ability of the graph structure of the technical sub-graph, which helps to better capture graph information and improve the extraction effect of technical representation information.

[0081] Step S50, for each candidate technology characterization information, determining the comprehensive similarity between the candidate technology characterization information and the reference sample pool; Step S60: Determine a technology opportunity recommendation result based on a comprehensive similarity comparison between all candidate technology representation information.

[0082] Specifically, after extracting the technical representations of the candidate technology subgraphs and the reference technology subgraphs, the candidate sample pool and the reference sample pool are compared based on the technical representation information to obtain the comprehensive similarity between each candidate technology subgraph and the reference sample pool in order to recommend technology opportunities.

[0083] In a feasible implementation manner, step S50 includes steps C10 to C20: Step C10 : for each candidate technology characterization information, respectively confirm the cosine similarity between the candidate technology characterization information and each reference characterization information in the reference sample pool.

[0084] In step C20 , the maximum cosine similarity is used as the comprehensive similarity between the candidate technology representation information and the reference sample pool.

[0085] Step S60 includes steps D10 to D20: Step D10: determining the comprehensive similarity between all candidate technology characterization information in the candidate sample pool and the reference sample pool.

[0086] In step D20 , based on all comprehensive similarities, a preset number of candidate technology subgraphs are selected from high to low to be determined as technology opportunity recommendation results.

[0087] Specifically, refer to Figure 5, Figure 5 Recommend a process flow diagram for the technology opportunity.

[0088] For ease of understanding, illustratively, all technology subgraphs are input into the technology representation extraction model, and the dense vector before the last multi-layer perceptron of each input technology representation extraction model is retained as the technology representation of the current sample, and the corresponding reference technology representation information and candidate technology representation information are obtained. Subsequently, for each candidate technology subgraph, the cosine similarity of the candidate technology representation information of the candidate sample subgraph and each reference technology representation information in the reference technology pool of the target field are calculated (that is, 1 to n calculation), and the maximum cosine similarity is selected as the comprehensive similarity between the candidate technology subgraph and the reference sample pool. After obtaining the comprehensive similarity between each candidate technology subgraph and the reference sample pool, the comprehensive similarity is sorted, and the top 10 candidate technology subgraphs most similar to the reference technology pool are selected as recommended technology opportunities.

[0089] It can be understood that this embodiment explores the essential relationship between the innovation fusion model of multiple classification codes and technological opportunities, and proposes a unique graph modeling method for multi-factor technological opportunities, which expands the ideas for subsequent research on multi-dimensional technological opportunity mining. In addition, EGA is designed to strengthen the use of relational data, and GTF is designed to achieve the integration of text and graph structure information, providing reference ideas for subsequent TOD research from the perspective of model improvement. The technological opportunities predicted according to the method proposed in this embodiment have certain reference significance for R&D activities, and the designed technology opportunity mining framework is implemented in an automated form, which can flexibly convert the target field according to needs, saving costs and improving efficiency for R&D parties.

[0090] It should be noted that the above embodiments are only used to understand the present application and do not constitute a limitation on the technical opportunity recommendation method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0091] This application also provides a technology opportunity recommendation device, please refer to Figure 6 , technical opportunities recommended devices include: The acquisition module 10 is used to obtain the target field patent dataset and the classification code co-occurrence network; wherein the classification code co-occurrence network is constructed by the classification code co-occurrence relationship corresponding to each patent in the comprehensive field patent dataset.

[0092] The reference sample pool construction module 20 is used to determine the first connected subgraph containing the target field patent classification number for each target field patent in the target field patent data set based on the classification number co-occurrence network, and construct a reference sample pool; the reference sample pool includes a reference technology subgraph obtained by reconstructing the first connected subgraph through a spanning tree.

[0093] The candidate sample pool construction module 30 is used to specify a classification number from the target field patent data set as a candidate classification number, determine a second connected subgraph containing the candidate classification number based on the classification number co-occurrence network, and construct a candidate sample pool; the candidate sample pool includes a candidate technology subgraph obtained by reconstructing the second connected subgraph through a spanning tree.

[0094] The feature extraction module 40 is configured to perform feature extraction processing on the reference technology subgraph and the candidate technology subgraph respectively, to obtain reference technology characterization information corresponding to the reference technology subgraph and candidate technology characterization information corresponding to the candidate technology subgraph.

[0095] The similarity calculation module 50 is configured to determine, for each candidate technology characterization information, a comprehensive similarity between the candidate technology characterization information and the reference sample pool.

[0096] The recommendation module 60 is configured to determine a technology opportunity recommendation result based on a comprehensive similarity comparison between all candidate technology representation information.

[0097] The technology opportunity recommendation device provided in this application utilizes the technology opportunity recommendation method described in the aforementioned embodiments, resolving the technical issues of the related art in terms of insufficient comprehensiveness and accuracy in identifying technology opportunities. Compared to the related art, the technology opportunity recommendation device provided in this application achieves the same beneficial effects as the technology opportunity recommendation method described in the aforementioned embodiments. Other technical features of the technology opportunity recommendation device are the same as those disclosed in the aforementioned embodiments and are not further elaborated here.

[0098] The present application provides a technology opportunity recommendation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the technology opportunity recommendation method in the above-mentioned embodiment.

[0099] Reference below Figure 7 , which shows a schematic diagram of the structure of a technology opportunity recommendation device suitable for implementing an embodiment of the present application. The technology opportunity recommendation device in the embodiment of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (such as in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5The technology opportunity recommendation device shown is only an example and should not limit the functions and scope of use of the embodiments of the present application.

[0100] like Figure 7 As shown, the technology opportunity recommendation device may include a processing device 1001 (e.g., a central processing unit, graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in a read-only memory (ROM) 1002 or programs loaded from a storage device 1003 into a random access memory (RAM) 1004. RAM 1004 also stores various programs and data required for device operation. Processing device 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems may be connected to I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage device 1003 including, for example, a magnetic tape or hard disk; and communication devices 1009. The communication device 1009 can allow the technology opportunity recommendation device to communicate with other devices wirelessly or wired to exchange data. Although the figure shows a technology opportunity recommendation device with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented or have alternatively.

[0101] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0102] The technology opportunity recommendation device provided in this application utilizes the technology opportunity recommendation method described in the aforementioned embodiment, resolving the technical issues of the related art in terms of insufficient comprehensiveness and accuracy in identifying technology opportunities. Compared to the related art, the technology opportunity recommendation device provided in this application achieves the same beneficial effects as the technology opportunity recommendation method described in the aforementioned embodiment. Other technical features of this technology opportunity recommendation device are the same as those disclosed in the aforementioned embodiment and are not further elaborated here.

[0103] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0104] The above are only specific embodiments of the present application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0105] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer program) stored thereon, wherein the computer-readable program instructions are used to execute the technology opportunity recommendation method in the above-mentioned embodiment.

[0106] The computer-readable storage medium provided herein may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including, but not limited to, wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0107] The computer-readable storage medium may be included in the technology opportunity recommendation device; or may exist independently without being incorporated into the technology opportunity recommendation device.

[0108] The computer-readable storage medium carries one or more programs. When executed by a technology opportunity recommendation device, the one or more programs cause the technology opportunity recommendation device to: obtain password requirement information of a target user; select a standard technology opportunity recommendation solution consistent with the password requirement information from a preset cryptography library. The standard technology opportunity recommendation solution includes a standard cryptographic algorithm, a standard cryptographic protocol, and a standard cryptographic scheme; and output the standard technology opportunity recommendation solution.

[0109] Computer program code for performing the operations of the present application may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Python, Java, Smalltalk, Python, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0110] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0111] The modules described in the embodiments of the present application may be implemented in software or hardware, wherein the name of a module does not necessarily limit the unit itself.

[0112] The computer-readable storage medium provided in this application stores computer-readable program instructions (i.e., a computer program) for executing the aforementioned technology opportunity recommendation method. This computer-readable storage medium can address the technical issues of the related art in terms of insufficient comprehensiveness and accuracy in identifying technology opportunities. Compared to the related art, the beneficial effects of the computer-readable storage medium provided in this application are similar to those of the technology opportunity recommendation method provided in the aforementioned embodiments and are not further elaborated here.

[0113] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned technology opportunity recommendation method when executed by a processor.

[0114] The computer program product provided in this application can address the technical issues of the related art in terms of insufficient comprehensiveness and accuracy in identifying technology opportunity recommendations. Compared to the related art, the beneficial effects of the computer program product provided in this application are the same as those of the technology opportunity recommendation method provided in the above-mentioned embodiments, and are not further elaborated here.

[0115] The above are only some embodiments of the present application and are not intended to limit the patent scope of the present application. All equivalent structural transformations made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect application in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A technology opportunity recommendation method, characterized in that: The method includes: Obtaining a target field patent dataset and a classification code co-occurrence network; wherein the classification code co-occurrence network is constructed by the classification code co-occurrence relationship corresponding to each patent in the comprehensive field patent dataset; For each target field patent in the target field patent dataset, based on the classification code co-occurrence network, a first connected subgraph containing the target field patent classification code is determined to construct a reference sample pool; the reference sample pool includes a reference technology subgraph obtained by performing spanning tree reconstruction on the first connected subgraph; Designate a classification code as a candidate classification code from the target field patent dataset, determine a second connected subgraph containing the candidate classification code based on the classification code co-occurrence network, and construct a candidate sample pool; the candidate sample pool includes a candidate technology subgraph obtained by performing spanning tree reconstruction on the second connected subgraph; Performing feature extraction processing on the reference technology subgraph and the candidate technology subgraph respectively to obtain reference technology characterization information corresponding to the reference technology subgraph and candidate technology characterization information corresponding to the candidate technology subgraph; For each candidate technology characterization information, determining a comprehensive similarity between the candidate technology characterization information and a reference sample pool; Based on the comprehensive similarity comparison between all the candidate technology representation information, the technology opportunity recommendation results are determined.

2. The method according to claim 1, wherein The step of determining, for each target field patent in the target field patent dataset, based on the classification code co-occurrence network, a first connected subgraph containing the target field patent classification code, and constructing a reference sample pool comprises: Abstracting the target patent classification number contained in the target field patent into a first network node; the first network node includes node feature information composed of node embeddings of the corresponding classification number in the classification number co-occurrence network; Adding an edge between any two network node pairs formed by the network nodes to obtain the first connected subgraph; Based on the node feature information, edge weight information of all edges of the first connected subgraph is calculated to determine the association strength between the first network nodes in the first connected subgraph; wherein the association strength is negatively correlated with the edge weight information; Based on the association strength, reconstructing the first connected subgraph by a minimum spanning tree algorithm to obtain a reference technology subgraph; A reference sample pool is constructed based on the reference technology subgraphs corresponding to all patents in the target field.

3. The method according to claim 2, wherein The steps of designating a classification number from the target field patent dataset as a candidate classification number, determining a second connected subgraph containing the candidate classification number based on the classification number co-occurrence network, and constructing a candidate sample pool include: Obtaining a sample sequence of a specified length containing the candidate classification number as a second network node sequence by using a depth-first search algorithm in the classification number co-occurrence network; Adding an edge between any pair of network nodes in the second network node sequence to obtain the second connected subgraph; Calculating edge weight information of all edges of the second connected subgraph based on the node feature information of the second network nodes, and determining the association strength between the second network nodes in the second connected subgraph; Based on the association strength, reconstructing the second connected subgraph by a minimum spanning tree algorithm to obtain a candidate technology subgraph; Based on the candidate technology subgraphs corresponding to all the patents in the target field, a candidate selection sample pool is constructed.

4. The method according to claim 3, wherein The step of performing feature extraction processing on the reference technology subgraph and the candidate technology subgraph respectively to obtain reference technology characterization information corresponding to the reference technology subgraph and candidate technology characterization information corresponding to the candidate technology subgraph comprises: The node feature information and node position information of the reference technology subgraph and the candidate technology subgraph are respectively input into a technology representation extraction model to obtain reference technology representation information corresponding to the reference technology subgraph and candidate technology representation information corresponding to the candidate technology subgraph. Among them, the technical representation extraction model is obtained by training positive samples composed of technical sub-graphs corresponding to existing patents and negative samples composed of randomly generated technical sub-graphs; the original feature extraction model includes parallel graph neural network units and Transformer units and a graphic feature fusion unit serially connected to the graph neural network units and Transformer units.

5. The method according to claim 1, wherein The step of determining, for each piece of candidate technology characterization information, the comprehensive similarity between the candidate technology characterization information and the reference sample pool includes: For each piece of candidate technology representation information, respectively determining the cosine similarity between the candidate technology representation information and each piece of reference representation information in the reference sample pool; The maximum cosine similarity is used as the comprehensive similarity between the candidate technology representation information and the reference sample pool.

6. The method according to claim 5, wherein The step of determining a technology opportunity recommendation result based on the comprehensive similarity comparison between all candidate technology representation information includes: Determining the comprehensive similarity between all candidate technology characterization information in the candidate sample pool and the reference sample pool; Based on all the comprehensive similarities, a preset number of candidate technology subgraphs are selected from high to low to determine as the technology opportunity recommendation results.

7. A technology opportunity recommendation device, characterized in that: The device comprises: An acquisition module is used to obtain a target field patent dataset and a classification code co-occurrence network; wherein the classification code co-occurrence network is constructed by the classification code co-occurrence relationship corresponding to each patent in the comprehensive field patent dataset; A reference sample pool construction module is used to extract, for each target field patent in the target field patent dataset, a first connected subgraph containing the target field patent classification number from the classification number co-occurrence network to construct a reference sample pool; the reference sample pool includes a reference technology subgraph obtained by performing spanning tree reconstruction on the first connected subgraph; A candidate sample pool construction module is used to designate a classification code from the target field patent dataset as a candidate classification code, extract a second connected subgraph containing the candidate classification code from the classification code co-occurrence network, and construct a candidate sample pool; the candidate sample pool includes a candidate technology subgraph obtained by performing spanning tree reconstruction on the second connected subgraph; a feature extraction module, configured to perform feature extraction processing on the reference technology subgraph and the candidate technology subgraph respectively, to obtain reference technology characterization information corresponding to the reference technology subgraph and candidate technology characterization information corresponding to the candidate technology subgraph; A similarity calculation module is used to determine, for each candidate technology characterization information, a comprehensive similarity between the candidate technology characterization information and a reference sample pool; The recommendation module is used to determine the technology opportunity recommendation results based on the comprehensive similarity comparison between all the candidate technology representation information.

8. A technology opportunity recommendation device, characterized in that: The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the technology opportunity recommendation method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the technology opportunity recommendation method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the technology opportunity recommendation method according to any one of claims 1 to 6 are implemented.

Images