A patent references a network analysis method, system, device, and storage medium.

Abstract

This invention discloses a method, system, apparatus, and storage medium for patent citation network analysis. The method includes: constructing a patent citation network in graph form based on a patent dataset; dividing the patent citation network into a predetermined number of subgraphs, and establishing two coupled systems for each subgraph; obtaining node features of each node through the first system; determining a first degree of influence based on the node features and citation relationships; updating the second degree of influence of the second system based on the first degree of influence; obtaining a steady-state network structure by alternately optimizing the first and second systems through subgraphs; and obtaining a target network structure based on the steady-state network structure through evolutionary learning. This invention, based on the idea of ​​evolutionary dynamics, trains the patent citation network graph model through citation relationships. In network applications, it can more effectively extract information from patents and, to a certain extent, explain the citation relationships between patents. It can be widely applied in the field of network analysis and processing technology.

Description

Technical Field

[0001] This invention relates to the field of network analysis and processing technology, and in particular to a patented network analysis method, system, device, and storage medium. Background Technology

[0002] Patents are one of the most important sources of intellectual property, and analyzing them is a highly effective method for examining technological trends. Patents contain a wealth of information; for example, the text of a patent reveals its technological value, and the citations it contains reflect its influence. Over the past few decades, with a large number of patents being applied for and granted, patent analysis—the mining of information from existing patents—has become a hot topic. The results of patent analysis provide useful information for patent quality analysis, technology trend prediction, and patent recommendation systems.

[0003] With the emergence of technologies such as Machine Learning (ML), Natural Language Processing (NLP), and Graph Neural Networks (GNNs), scholars have conducted in-depth explorations of patent texts, citations, and other information based on these advanced technologies. Considering that the value of a patent comes from the patent itself, in addition to the structured information immediately available after publication, the textual content of the patent, such as the title, abstract, and claims, also contains a wealth of information. A classic patent text analysis method is to extract key phrases or technical phrases from the patent to summarize and represent it. To further analyze the semantic relationships between words or sentences, many methods employ traditional semantic analysis methods from the field of natural language processing, such as Latent Dirichlet Allocation (LDA) and Latent Semantic Analysis (LSA). Alternatively, deep learning-based methods can be used to analyze patent texts, such as using the textual content of patents to predict promising patents, and employing convolutional neural networks (CNNs) and bidirectional long short-term memory (Bi-LSTM) networks to analyze patent abstracts and claims. While positive citations are also frequently used as an analytical metric for patents, they often have significant limitations because these works rely on statistical analysis of citation counts of collected published patents. Therefore, how to efficiently analyze patent citation networks through citation relationships is a pressing problem that needs to be solved. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a patent citation network analysis method, system, apparatus and storage medium, which can efficiently analyze the patent citation network based on the citation relationship of the patent.

[0005] On one hand, embodiments of the present invention provide a patent citation network analysis method, including:

[0006] Based on the patent dataset, a patent citation network is constructed in the form of a graph. The patent citation network includes several nodes and edges connecting the nodes. The nodes represent each patent in the patent dataset, and the edges represent the citation relationships between the patents.

[0007] The patent citation network is divided into a predetermined number of subgraphs, and two coupled systems are established for each subgraph; the system includes a first system and a second system.

[0008] The node characteristics of each node are obtained through the first system;

[0009] Based on node characteristics and referencing relationships, determine the degree of first influence;

[0010] The second degree of impact of the second system is updated based on the first degree of impact.

[0011] The steady-state network structure is obtained by alternately optimizing the first and second systems through subgraphs.

[0012] Based on the steady-state network structure, the target network structure is obtained through evolutionary learning.

[0013] Optionally, two coupled systems are established for each subgraph, including:

[0014] Based on the first and second degree of influence, two coupled systems are established for each subgraph;

[0015] The system's expression is:

[0016] S1:

[0017] S2:

[0018] Where S1 represents the first system and S2 represents the second system. D' represents a subgraph 0 Y represents the random initial state representing the second degree of influence. 0 Y represents the initial state of the first level of influence, D' represents the first level of influence after learning through the first system, and D' represents the second level of influence after learning through the second system.

[0019] Optionally, the node characteristics of each node are obtained through the first system, including:

[0020] The first system obtains the topological structure information of each node in the subgraph, as well as the title text information of the corresponding patent for each node.

[0021] Based on the title text information, word vector representations are obtained;

[0022] The node features are obtained by merging word vector representations and topological structure information and performing L2 norm processing.

[0023] Optionally, based on node characteristics and referencing relationships, the degree of first influence can be determined, including:

[0024] Based on the node characteristics of each node, determine the text similarity between each patent;

[0025] The influence column vector is determined based on text similarity and citation relationships; the influence column vector is calculated using an activation function; the citation relationships are determined based on state vectors from multiple aspects.

[0026] The degree of first influence is determined based on the influence column vector and the influence row vector.

[0027] Optionally, the second degree of influence of the second system is updated according to the first degree of influence, including:

[0028] Based on the degree of first influence, determine the projection matrix for parameter updating of the first system;

[0029] Update the second influence level of the second system based on the projection matrix.

[0030] Optionally, a steady-state network structure is obtained by alternately optimizing the first and second systems through subgraphs, including:

[0031] Based on a preset number of iterations, the first and second systems are learned by alternating subgraph optimization until the preset number of iterations is reached, resulting in a steady-state network structure.

[0032] The expression for alternating optimization learning is:

[0033] S1:

[0034] S2:

[0035] Where S1 represents the first system and S2 represents the second system. Let Y represent a subgraph, q represent the iteration round of alternating optimization, and Y represent the subgraph. q Y represents the degree of first influence in the q-th iteration. q+1 Y represents q The initial level of influence after the first system learning, D q D represents the degree of the second influence at the q-th iteration. q+1 D representsq The second degree of influence after learning through the second system.

[0036] Optionally, based on the steady-state network structure, the target network structure is obtained through evolutionary learning, including:

[0037] Based on the steady-state network structure, the target network structure is obtained by optimizing the topology and semantic interaction of the patent text through a pre-constructed loss function.

[0038] On the other hand, embodiments of the present invention provide a patent citation network analysis system, including:

[0039] The first module is used to construct a patent citation network in the form of a graph based on the patent dataset. The patent citation network includes several nodes and edges connecting each node. The nodes represent each patent in the patent dataset, and the edges represent the citation relationships between the patents.

[0040] The second module is used to divide the patent citation network into a predetermined number of subgraphs and establish two coupled systems for each subgraph; the system includes a first system and a second system.

[0041] The third module is used to obtain the node characteristics of each node through the first system;

[0042] The fourth module is used to determine the degree of first influence based on node characteristics and reference relationships;

[0043] The fifth module is used to update the second degree of influence of the second system based on the first degree of influence.

[0044] The sixth module is used to obtain the steady-state network structure by alternately optimizing the first and second systems through subgraphs.

[0045] The seventh module is used to obtain the target network structure based on the steady-state network structure through evolutionary learning.

[0046] On the other hand, embodiments of the present invention provide a patent reference network analysis device, including a processor and a memory;

[0047] Memory is used to store programs;

[0048] The processor executes the program as described above.

[0049] On the other hand, embodiments of the present invention provide a computer-readable storage medium storing a program that is executed by a processor to implement the method described above.

[0050] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned method.

[0051] This invention first constructs a patent citation network in graph form based on a patent dataset. The patent citation network includes several nodes and edges connecting each node. Nodes represent individual patents in the dataset, and edges represent citation relationships between patents. The patent citation network is divided into a predetermined number of subgraphs, and two coupled systems are established for each subgraph. These systems include a first system and a second system. The node features of each node are obtained through the first system. Based on the node features and citation relationships, a first degree of influence is determined. The second degree of influence of the second system is updated based on the first degree of influence. A steady-state network structure is obtained by alternately optimizing the first and second systems through the subgraphs. Based on the steady-state network structure, the target network structure is obtained through evolutionary learning. This invention divides the patent citation network into several subgraphs and then establishes two coupled systems in each subgraph. Alternating optimization learning based on citation relationships is achieved, ultimately realizing the evolutionary learning of the patent citation network. Based on the idea of ​​evolutionary dynamics, the patent citation network graph model is trained through citation relationships. In network applications, this can more effectively extract information from patents and, to a certain extent, explain the citation relationships between patents. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 A schematic flowchart of a patent citation network analysis method provided in an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of the number of citations of patent data provided in the embodiments of the present invention;

[0055] Figure 3 A schematic diagram of the number of citations of patent data provided in the embodiments of the present invention;

[0056] Figure 4 A schematic diagram of the evolutionary learning process provided in an embodiment of the present invention;

[0057] Figure 5This is a schematic diagram of the overall process of a patent citation network analysis method provided in an embodiment of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0059] On the one hand, such as Figure 1 As shown, an embodiment of the present invention provides a patent citation network analysis method, including:

[0060] S100. Based on the patent dataset, construct a patent citation network in the form of a graph;

[0061] The patent citation network consists of several nodes and edges connecting the nodes; nodes represent patents in the patent dataset, and edges represent the citation relationships between patents.

[0062] Specifically, in some specific embodiments, for existing patent datasets, patents are treated as nodes, and the citation relationships between patents are treated as edges to establish a patent citation network.

[0063] S200. Divide the patent citation network into a predetermined number of subgraphs, and establish two coupled systems for each subgraph;

[0064] The system includes a first system and a second system. It should be noted that in some embodiments, two coupled systems are established for each subgraph, including: establishing two coupled systems for each subgraph based on a first degree of influence and a second degree of influence; wherein the system expression is:

[0065] S1:

[0066] S2:

[0067] Where S1 represents the first system and S2 represents the second system. D' represents a subgraph 0 Y represents the random initial state representing the second degree of influence. 0 Y represents the initial state of the first level of influence, D' represents the first level of influence after learning through the first system, and D' represents the second level of influence after learning through the second system.

[0068] Specifically, in some embodiments, a patent citation network is established. Patents contain rich information, and patent citation patterns reveal the technological development process. To better extract the valuable information contained in patents, this invention views the patent citation network as a graph G, where nodes represent patents and directed arcs represent citations between patents. Graph G is divided into k1 subgraphs, and for each subgraph... Two coupled systems, S1 and S2, are established. The final target network structure is then learned k2 times by alternatingly learning S1 and S2 from each subgraph. The system expressions for each subgraph are as follows:

[0069] S1:

[0070] S2:

[0071] Where S1 represents the first system and S2 represents the second system. D' represents a subgraph 0 Y represents the random initial state representing the second degree of influence. 0 Y represents the initial state of the first level of influence, D' represents the first level of influence after learning through the first system, and D' represents the second level of influence after learning through the second system.

[0072] S300. Obtain the node characteristics of each node through the first system;

[0073] It should be noted that in some embodiments, the process includes: obtaining the topological structure information of each node in the subgraph and the title text information of the corresponding patent for each node through a first system; obtaining word vector representations based on the title text information; and performing a merging operation and L2 norm processing on the word vector representations and topological structure information to obtain node features.

[0074] Specifically, in some specific embodiments, for each patent node i in the subgraph, its node features n are learned. i ∈R L ×1 In addition to topological information, the patent title text was also considered, and the learned node representation method can more comprehensively represent the effective information in the patent.

[0075] n i =l2(t i ⊕e i )

[0076] Where l2 is the L2 norm, which is defined as the square root of the sum of the squares of all elements of a vector, t i To train the word vector representation of the title of patent i using the GloVe algorithm, the embedding vector e i For the topology information of patent i, ⊕ represents the merge operation.

[0077] S400. Based on node characteristics and referencing relationships, determine the degree of first influence;

[0078] The first degree of influence includes the degree of influence from multiple different aspects. It should be noted that in some embodiments, this includes: determining the text similarity between patents based on the node features of each node; determining the influence column vector based on text similarity and citation relationships; calculating the influence column vector through an activation function; determining the citation relationship based on state vectors from multiple aspects; and determining the first degree of influence based on the influence column vector and influence row vector.

[0079] Specifically, in some specific embodiments, the impact of different aspects on patent citations is analyzed based on node characteristics and citation relationships, as described below. Further analysis is performed on the node characteristics extracted in the previous step. If there is a citation relationship between node i and node j, and node j is the cited node, c is calculated respectively. ij and n ij This represents the citation relationship and text similarity between patent i and patent j, and calculates the degree of influence in different aspects. Assume there is an edge between node i and node j, with node j being the target node. Typically, the citation between patents is determined by the influence of the target patent j.

[0080] c ij =T d d j

[0081] n ij =n i ⊙n j

[0082] Among them, c ij Indicates the citation relationship between patent i and patent j, n ij Indicates the textual similarity between patent i and patent j; T d ∈R I×I To represent the weight matrix, dj∈[0,1] I represents a state vector with I aspects; ⊙ represents the Hadamard product.

[0083] F ij =Σc ij +Σn ij

[0084] F ij Used to indicate the overall impact of patent citations and textual content.

[0085] D ij =W cT c ij +W nT n ij +b

[0086] D ij An I-dimensional column vector is used to measure the impact of different aspects, and the values ​​in the row vector k = {1, ..., I} represent the impact of a specific aspect of the target patent. b∈R L×1 Parameters used to measure the degree of influence from different aspects.

[0087] To illustrate the different aspects of the impact, we aim to use the aspects of the target patent that are most important to the source patent to demonstrate the reasons for the citations between them. To this end, α is calculated using the gumbel-softMax activation function. ij (k):

[0088]

[0089] Where, α ij (k) represents the column vector α ij The k-th element in π. k (k = 1, ..., I) represents the class probability. k It comes from the Gumbel(0,1) distribution.

[0090] Y ij =max(α) ij D ij ,0)

[0091] Y ij This indicates the degree of influence of patent i and patent j in a certain aspect; max() returns the maximum value of the given parameter.

[0092] S500: Update the second degree of influence of the second system according to the first degree of influence;

[0093] It should be noted that in some embodiments, the process includes: determining the projection matrix for parameter updating of the first system based on the first degree of influence; and updating the second degree of influence of the second system according to the projection matrix.

[0094] Specifically, in some specific embodiments, the network state is updated based on the varying degrees of influence determined in the previous step, and the multifaceted influence of each node also changes accordingly:

[0095] D q (n+1)=PD q (n)

[0096] P = βE + υ(X + Z)

[0097] Where n represents the propagation time step, P represents the projection matrix, which is obtained by updating the relevant parameters learned by system S1; β = 0.05 / N is the hyperparameter, N is the number of nodes, υ is the parameter, and E∈R N×N×kZ is a matrix of all 1s, and X is a constant matrix with elements 1 / N used to handle dangling nodes in the graph. X is used to handle nodes in the graph that are connected to it by edges.

[0098] X(:,:,k)=x(k) i→j k=1,…,I

[0099]

[0100] x(k)∈R N×N This indicates that there is a reference between patent i and patent j, and that patent j has a direct influence on patent i in aspect k.

[0101] S600: The first and second systems are learned through alternating subgraph optimization to obtain a steady-state network structure;

[0102] It should be noted that in some embodiments, the method includes: based on a preset number of iterations, alternately optimizing and learning the first and second systems through subgraphs until the preset number of iterations is reached to obtain a steady-state network structure; wherein, the expression for the alternate optimization and learning is:

[0103]

[0104]

[0105] Where S1 represents the first system and S2 represents the second system. Let Y represent a subgraph, q represent the iteration round of alternating optimization, and Y represent the subgraph. q Y represents the degree of first influence in the q-th iteration. q+1 Y represents q The initial level of influence after the first system learning, D q D represents the degree of the second influence at the q-th iteration. q+1 D represents q The second degree of influence after learning through the second system.

[0106] Specifically, in some specific embodiments, S1 and S2 are learned alternately, as follows: by jointly learning the two coupled systems S1 and S2, a subgraph is learned. A stable network representation:

[0107] S1:

[0108] S1:

[0109] Where S1 represents the first system and S2 represents the second system. Let Y represent a subgraph, q represent the iteration round of alternating optimization, and Y represent the subgraph. q Y represents the degree of first influence in the q-th iteration.q+1 Y represents q The initial level of influence after the first system learning, D q D represents the degree of the second influence at the q-th iteration. q+1 D represents q The second degree of influence after learning through the second system.

[0110] S700: Based on a steady-state network structure, the target network structure is obtained through evolutionary learning;

[0111] It should be noted that in some embodiments, the target network structure is obtained by optimizing the topology and semantic interaction of the patent text through a pre-constructed loss function based on the steady-state network structure.

[0112] Specifically, in some specific embodiments, the evolutionary learning model is as follows: the final network structure evolves from some previous structures with fewer edges, that is, the target network structure is obtained by evolutionary learning from the stable states of multiple sub-networks trained above. Two losses are defined. e and loss t These are used to optimize the topology of the patent citation network and the semantic interaction of the patent text, respectively.

[0113] loss e =max{0,λ e -(F i,j -F i,k )}

[0114] loss t =max{0,λ t -(D i,j |α i,j -D i,k |α i,j )}

[0115] Where, λ e and λ t These are all parameters of the loss function, which must satisfy the following condition: given a source node i and any two other nodes j and k, if there is no edge between node k and node i, and there is an edge between node j and node i, then F i,j >F i,k And D i,j |α i,j >D i,k |α i,j .

[0116] After obtaining the patent citation network of the final target network structure, the patent data selected included a portion of Class G patents (patent publication numbers starting with CN) published by the State Intellectual Property Office, containing 52,156 patents and corresponding 48,250 citation relationships. The link prediction method was used to verify the effectiveness of the algorithm in network analysis. 90% of the edges were randomly selected from the original network for training, and the link prediction performance on the remaining network was tested based on the ranking results. For example... Figure 2 and Figure 3 The figure shows the distribution of the number of citations and the number of cited patent data used in the embodiments of the present invention.

[0117] Table 1 shows the Top-1 accuracy of link prediction on the patent citation network for the algorithm implemented in this embodiment of the invention and other algorithms on the patent dataset.

[0118] Table 1

[0119] algorithm Node2Vec GraphSAGE SEAL TGAT Algorithm of this invention Top-1 0.3235 0.4756 0.6213 0.6805 0.8123

[0120] Table 2 compares the mean accuracy (mAP), recall, and normalized depreciation cumulative gain (NDCG) of the algorithm implemented in this embodiment of the invention in the Top-k time range.

[0121] Table 2

[0122] Top-1 Top-3 Top-5 Top 10 mAP 0.8133 0.8658 0.8735 0.8763 Recall 0.7573 0.9377 0.9661 0.9840 NDCG 0.8133 0.8862 0.8983 0.9044

[0123] In some specific embodiments, the overall evolutionary learning process of the patent citation network is as follows: Figure 4 As shown, the implementation process of this invention is as follows: Figure 5 As shown, it includes:

[0124] 1. Input patent data, initialize the maximum number of iterations k1 and k2, and initialize the current number of iterations p = 1 and q = 1;

[0125] 2. Construct a patent citation network G, using patents as nodes and citations between patents as edges;

[0126] 3. k1 subgraphs of G Establish two coupled systems, S1 and S2, respectively;

[0127] 4. Using system S1, based on the citation relationships between patents... ij Text similarity n ij The degree of influence Y of different aspects of the patent citation in the (q+1)th iteration is obtained through learning. q+1 ;

[0128] 5. Using system S2, based on Y in S1 q+1 The update yields the different influence aspects D between nodes at the (q+1)th time. q+1;

[0129] 6. In the subgraph The alternating optimization learning of S1 and S2 is performed k2 times to obtain... A stable network state;

[0130] 7. Calculate the loss for the two topologies. e and semantic interaction loss t By using the idea of ​​evolutionary learning, the final target network structure is obtained by evolutionary learning based on the stable states of k1 subnetworks.

[0131] 8. Output the target network structure representation.

[0132]

[0133]

[0134] In summary, compared with the prior art, the present invention has at least the following beneficial effects:

[0135] First, by unifying patent text information and patent citation information within a single framework for training, and using evolutionary dynamics to train the patent citation network graph model, information can be extracted from patents more effectively. Second, the patent citation network analysis method in this invention, based on evolutionary dynamics, divides a patent network into different sub-networks, allowing the model to evolve and learn from these network structures to arrive at the final network model. By combining patent title information and graph topology information, the reliability of patent analysis is improved. Third, the citation analysis problem in the patent citation network is defined as the link prediction problem in graph mining. By considering the multifaceted influence of patent citations, the citation relationships between patents can be predicted and explained to a certain extent. Fourth, this invention has broad applicability and can be applied to most Chinese patent data, enhancing the feasibility of patent information mining.

[0136] On the other hand, embodiments of the present invention provide a patent citation network analysis system, comprising: a first module for constructing a patent citation network in graph form based on a patent dataset; the patent citation network includes several nodes and edges connecting each node; nodes represent each patent in the patent dataset, and edges represent the citation relationships between each patent; a second module for dividing the patent citation network into a predetermined number of subgraphs, and establishing two coupled systems for each subgraph; the systems include a first system and a second system; a third module for obtaining node features of each node through the first system; a fourth module for determining a first degree of influence based on node features and citation relationships; a fifth module for updating a second degree of influence of the second system based on the first degree of influence; a sixth module for obtaining a steady-state network structure by alternately optimizing the first system and the second system through subgraphs; and a seventh module for obtaining a target network structure based on the steady-state network structure through evolutionary learning.

[0137] The content of the method embodiments of the present invention is applicable to the system embodiments. The specific functions implemented in the system embodiments are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above methods.

[0138] Another aspect of the present invention provides a patent reference network analysis device, including a processor and a memory;

[0139] Memory is used to store programs;

[0140] The processor executes the program as described above.

[0141] The content of the method embodiments of the present invention is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above methods.

[0142] Another aspect of this invention provides a computer-readable storage medium storing a program that is executed by a processor to implement the method described above.

[0143] The content of the method embodiments of the present invention is applicable to the computer-readable storage medium embodiments. The specific functions implemented by the computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above methods.

[0144] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned method.

[0145] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0146] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0147] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0148] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution means, apparatus, or device (such as a computer-based device, a processor-including device, or other means that can fetch and execute instructions from, or in conjunction with, an instruction execution means, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution means, apparatus, or device.

[0149] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0150] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution device. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0151] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0152] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0153] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A patent citation network analysis method, characterized in that, include: Based on a patent dataset, a patent citation network is constructed in the form of a graph; the patent citation network includes several nodes and edges connecting each node; each node represents a patent in the patent dataset, and each edge represents the citation relationship between the patents. The patent citation network is divided into a predetermined number of subgraphs, and two coupled systems are established for each subgraph; the systems include a first system and a second system. The node characteristics of each node are obtained through the first system; Based on the node characteristics and the reference relationships, the first degree of influence is determined; The second impact level of the second system is updated based on the first impact level. A steady-state network structure is obtained by alternately optimizing the first system and the second system through the subgraph; Based on the aforementioned steady-state network structure, the target network structure is obtained through evolutionary learning; The establishment of two coupled systems for each subgraph includes: Based on the first degree of influence and the second degree of influence, two coupled systems are established for each subgraph; The system is expressed as follows: Where S1 represents the first system and S2 represents the second system. Represents a subgraph. Y represents the random initial state representing the second degree of influence. 0 Y represents the initial state of the first level of influence, and Y represents the first level of influence after learning by the first system. This indicates the degree of second influence after learning through the second system; The step of obtaining a steady-state network structure by alternately optimizing the first system and the second system through the subgraph includes: Based on a preset number of iterations, the first system and the second system are alternately optimized and learned through the subgraph until the preset number of iterations is reached, thereby obtaining a steady-state network structure. The expression for the alternating optimization learning is: Where S1 represents the first system and S2 represents the second system. Let Y represent a subgraph, q represent the iteration round of alternating optimization, and Y represent the subgraph. q Y represents the degree of first influence in the q-th iteration. q+1 Y represents q The initial level of influence after the first system learning, D q D represents the degree of the second influence at the q-th iteration. q+1 D represents q The second degree of influence after learning through the second system.

2. The patent citation network analysis method according to claim 1, characterized in that, The step of obtaining the node characteristics of each node through the first system includes: The first system obtains the topological structure information of each node in the subgraph, as well as the title text information of the patent corresponding to each node. Based on the title text information, word vector representations are obtained; The word vector representation and the topological structure information are merged and processed using L2 norm to obtain node features.

3. The patent citation network analysis method according to claim 1, characterized in that, The step of determining the first degree of influence based on the node characteristics and the reference relationship includes: Based on the node features of each node, determine the text similarity between each patent; Based on the text similarity and the reference relationship, the influencing column vector is determined; and the influencing row vector is calculated using an activation function; the reference relationship is determined based on multiple aspects of the state vector. The first degree of influence is determined based on the influence column vector and the influence row vector.

4. The patent citation network analysis method according to claim 1, characterized in that, The step of updating the second impact level of the second system based on the first impact level includes: Based on the first degree of influence, determine the projection matrix for parameter updating of the first system; The second influence level of the second system is updated based on the projection matrix.

5. The patent citation network analysis method according to claim 1, characterized in that, The process of obtaining the target network structure based on the steady-state network structure through evolutionary learning includes: Based on the aforementioned steady-state network structure, the topology and semantic interaction of the patent text are optimized using a pre-constructed loss function to obtain the target network structure.

6. A patent citation network analysis system, characterized in that, include: The first module is used to construct a patent citation network in the form of a graph based on a patent dataset; the patent citation network includes several nodes and edges connecting each node; the nodes represent each patent in the patent dataset, and the edges represent the citation relationships between the patents; The second module is used to divide the patent citation network into a predetermined number of subgraphs, and to establish two coupled systems for each subgraph; the systems include a first system and a second system. The third module is used to obtain the node characteristics of each node through the first system; The fourth module is used to determine the first degree of influence based on the node characteristics and the reference relationship; The fifth module is used to update the second degree of influence of the second system based on the first degree of influence. The sixth module is used to obtain a steady-state network structure by alternately optimizing the first system and the second system through the subgraph; The seventh module is used to obtain the target network structure based on the steady-state network structure through evolutionary learning; The establishment of two coupled systems for each subgraph includes: Based on the first degree of influence and the second degree of influence, two coupled systems are established for each subgraph; The system is expressed as follows: Where S1 represents the first system and S2 represents the second system. Represents a subgraph. Y represents the random initial state representing the second degree of influence. 0 Y represents the initial state of the first level of influence, and Y represents the first level of influence after learning by the first system. This indicates the degree of second influence after learning through the second system; The step of obtaining a steady-state network structure by alternately optimizing the first system and the second system through the subgraph includes: Based on a preset number of iterations, the first system and the second system are alternately optimized and learned through the subgraph until the preset number of iterations is reached, thereby obtaining a steady-state network structure. The expression for the alternating optimization learning is: Where S1 represents the first system and S2 represents the second system. Let Y represent a subgraph, q represent the iteration round of alternating optimization, and Y represent the subgraph. q Y represents the degree of first influence in the q-th iteration. q+1 Y represents q The initial level of influence after the first system learning, D q D represents the degree of the second influence at the q-th iteration. q+1 D represents q The second degree of influence after learning through the second system.

7. A patent reference network analysis device, comprising a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a program that is executed by a processor to implement the method as described in any one of claims 1 to 5.

Images