Technical maturity calculation system and method based on graph convolutional neural network

Abstract

The invention discloses a technology maturity calculation system based on a knowledge graph and a graph convolutional neural network, and the system comprises a scientific and technical literature classification module which is used for training the graph neural network according to a preset knowledge graph, and obtaining a scientific and technical literature classification model, a scientific and technical literature classification model is used for conducting technical field classification on the selected public scientific and technical literature set, public scientific and technical literature subsets corresponding to different technical fields are obtained, the preset knowledge graph is obtained according to public scientific and technical literatures, and the public scientific and technical literatures comprise public patents and public papers; and the technology maturity calculation module is used for calculating the technology maturity of the technical field to which the corresponding public science and technology literature subsets belong according to the reference relationship and similarity between the public science and technology literatures in each public science and technology literature subset. According to the method, on the basis of the knowledge graph, potential rules and future trends of technology development are deeply mined by using the graph convolutional neural network, and the accuracy and foresight of innovative technology prediction can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of patent analysis, and in particular to a technical maturity calculation system and method based on a knowledge graph and a graph convolutional neural network. Background Art

[0002] In today's rapidly developing technological environment, the prediction and evaluation of emerging technologies have become increasingly important. Accurate prediction of technological trends can help enterprises and research institutions plan ahead, optimize resource allocation, accelerate the innovation process, and ultimately achieve commercial success and scientific breakthroughs. However, with the continuous expansion and deepening of the technological field, the interdependence and complexity among technologies are also increasing, posing great challenges to traditional technological prediction methods.

[0003] Currently, although there are various technological prediction methods, such as methods based on trend analysis, expert consultation, machine learning, etc., these methods often fail to fully capture the complex correlation relationships and dynamic evolution processes among technologies. For example, methods based on trend analysis may ignore the non-linear changes in the technological development path; expert consultation methods may be affected by personal experience and cognitive biases; while traditional machine learning methods often have difficulty revealing the deep-seated correlations and patterns among data when dealing with large-scale and high-dimensional technological data. Summary of the Invention

[0004] The purpose of the present invention is to address the deficiencies of the prior art and provide a technical maturity calculation system based on a knowledge graph and a graph convolutional neural network, including:

[0005] A scientific and technological literature classification module, which is used to train a graph neural network according to a preset knowledge graph to obtain a scientific and technological literature classification model, and use the scientific and technological literature classification model to classify the selected set of publicly available scientific and technological literature into different technical fields to obtain subsets of publicly available scientific and technological literature corresponding to different technical fields. The preset knowledge graph is obtained from publicly available scientific and technological literature, and the publicly available scientific and technological literature includes publicly available patents and publicly available papers;

[0006] A technical maturity calculation module, which is used to calculate the technical maturity of the technical field to which the corresponding subset of publicly available scientific and technological literature belongs according to the citation relationships and similarities among the publicly available scientific and technological literature in each subset of publicly available scientific and technological literature.

[0007] The preset knowledge graph is obtained from publicly available scientific and technological literature, and the publicly available scientific and technological literature includes publicly available patents and publicly available papers.

[0008] Furthermore, in the scientific and technological literature classification module, the specific method for training a graph neural network according to a preset knowledge graph to obtain a scientific and technological literature classification model is as follows:

[0009] The graph neural network is a multi-subgraph convolutional network model;

[0010] Divide the preset knowledge graph into multiple knowledge subgraphs. The publicly available scientific and technical documents in each knowledge subgraph belong to the same technical field. Use a natural language processing model to convert the text of each publicly available scientific and technical document in the knowledge subgraph into a publicly available scientific and technical document word embedding vector. Then use a knowledge graph embedding model based on translational distance to convert the attributes of the publicly available scientific and technical documents in the knowledge subgraph into publicly available scientific and technical document attribute embedding vectors, and splice the publicly available scientific and technical document attribute embedding vectors and the publicly available scientific and technical document word embedding vectors into node feature vectors as the input node initial representation of the training dataset of the multi-subgraph convolutional network model, that is, map the publicly available scientific and technical document attribute embedding vectors into the publicly available scientific and technical document word embedding vector space, and splice the mapped publicly available scientific and technical document attribute embedding vectors and the publicly available scientific and technical document word embedding vectors to obtain node feature vectors;

[0011] In the multi-subgraph convolutional network model, for each input node, sample several neighbor nodes from its neighbor nodes in a random or fixed manner to form a feature subgraph. Each feature subgraph corresponds to a technical field. Use the aggregated neighbor node representation to update the representation of the central node through a convolution operation. The formula is as follows:

[0012] e ij =LeakyReLU(a T [Wu i ||Wu j )

[0013]

[0014] Among them, u i and u j are the feature representations of the end node i and node j of the current input layer. W is the weight matrix. || is the splicing operation between vectors. e ij is the original attention score between node i and node j. a ij is the attention score between node i and node j. a T is the transpose of the weight matrix W;

[0015] For a single node u i in the multi-subgraph convolutional network model, use the cross-entropy loss as the loss function l i to train the multi-subgraph convolutional network model to obtain a scientific and technical document classification model:

[0016]

[0017] Among them, y i is node u iThe technical field to which the knowledge subgraph in the preset knowledge graph belongs is the technical field to which the feature subgraph obtained by clustering the nodes u by the multi-subgraph convolutional network model i belongs. After training, a scientific and technological literature classification model is obtained

[0018] Furthermore, in the technical maturity calculation module, the specific method for calculating the technical maturity of the technical field to which the corresponding public scientific and technological literature subset belongs according to the citation relationship and similarity between the public scientific and technological literatures in each public scientific and technological literature subset is as follows

[0019] For a certain public scientific and technological literature subset, define a time interval T t to represent the time period t, where t is the time interval serial number and is sorted in chronological order. Define the time interval pair (T c , T d ) to represent the time period between time interval T c and time interval T d , where the time interval T c is earlier than the time interval T d . In the time period specified by the time interval pair (T c , T d ) of this public scientific and technological literature subset, the scientific and technological literatures published in the time interval c are formed into the first public scientific and technological literature set Pat c , and the scientific and technological literatures published in the time interval d are formed into the second scientific and technological literature set Pat d . If a certain public scientific and technological literature in the first public scientific and technological literature set Patc and a certain public scientific and technological literature in the second public scientific and technological literature set Patd have a citation relationship, then the pair of two public scientific and technological literatures with a citation relationship is put into the third public scientific and technological literature set Cit(c, d);

[0020] According to the third public scientific and technological literature set Cit(c, d), use the technical maturity to evaluate the technology life cycle. The formula for the technical maturity is as follows

[0021]

[0022] In the formula, the technical maturity I j represents the technical maturity of the technical field to which the public scientific and technological literature subset obtained by the scientific and technological literature classification model belongs in the jth time interval, and θ m represents the weight of the mth evaluation index, and X m represents the index value of the mth evaluation technology life cycle. The index value X mIncluding: the mean value MeaN of the cosine similarity between publicly available scientific and technological documents with citation relationships, the median value MedN of the cosine similarity between publicly available scientific and technological documents with citation relationships, the number of citations and the number of times cited NCiteP, and the number of citations NCitiP as indicators for evaluating the technology life cycle; among them, the number of citations and the number of times cited NCiteP is used to represent the number of publicly available scientific and technological documents with citation relationships for a certain publicly available scientific and technological document within the time interval T c within the time interval T d for the number of publicly available scientific and technological documents with citation relationships, and the number of citations NCitiP is used to represent the number of publicly available scientific and technological documents with citation relationships for a certain publicly available scientific and technological document before the publication date of this publicly available scientific and technological document within the time interval d;

[0023] Set the technology maturity threshold, regard the technology fields with technology maturity less than the technology maturity threshold as innovative technology fields, calculate the technology maturity of the subset of publicly available scientific and technological documents, and predict whether the technology fields to which they belong are innovative technology fields.

[0024] Furthermore, the publicly available scientific and technological documents are publicly available patents, and the calculation method of the cosine similarity of the publicly available patents is as follows:

[0025] Convert the IPC international patent classification numbers of each patent into IPC vectors, which are expressed as follows:

[0026] IPC x =[a 1 ,a 2 ,…,a n

[0027] where, a n represents the nth category in the IPC international patent classification number, and the value of a is 0 or 1. When a is 1, it means that patent x belongs to the nth category. When a is 0, it means that patent x does not belong to the nth category, and the level of category a n is one of section, subclass, main group, subgroup;

[0028] Input the IPC vector into the BERT model (Bidirectional Encoder Representations from Transformers) for multi-label classification tasks to obtain the publicly available patent feature vector;

[0029] Calculate the cosine similarity score PITS(a,b) between two publicly available patents, and the formula is as follows:

[0030]

[0031] where, PITVa and PITVb represent the publicly available patent a feature vector and the publicly available patent b feature vector respectively, and ‖PITVa‖·‖PITVb‖ represents the two-norm between publicly available patent a and publicly available patent b;​

[0032] An innovative technology prediction method based on a knowledge graph and a graph convolutional neural network, comprising the following steps:

[0033] Train a graph neural network according to a preset knowledge graph to obtain a scientific and technological literature classification model, and use the scientific and technological literature classification model to classify the selected set of publicly available scientific and technological literature into different technical fields to obtain subsets of publicly available scientific and technological literature corresponding to different technical fields. The preset knowledge graph is obtained from publicly available scientific and technological literature, and the publicly available scientific and technological literature includes publicly available patents and publicly available papers;

[0034] Calculate the technology maturity of the technical field to which the corresponding subset of publicly available scientific and technological literature belongs according to the citation relationship and similarity between the publicly available scientific and technological literature in each subset of publicly available scientific and technological literature.

[0035] The preset knowledge graph is obtained from publicly available scientific and technological literature, and the publicly available scientific and technological literature includes publicly available patents and publicly available papers.

[0036] Further, the specific method for training the graph neural network according to the preset knowledge graph to obtain a scientific and technological literature classification model is as follows:

[0037] The graph neural network is a multi-subgraph convolutional network model;

[0038] Divide the preset knowledge graph into multiple knowledge subgraphs. The publicly available scientific and technological literature in each knowledge subgraph belongs to the same technical field. Use a natural language processing model to convert the text of the publicly available scientific and technological literature in the knowledge subgraph into a publicly available scientific and technological literature word embedding vector, and then use a knowledge graph embedding model based on translational distance to convert the attributes of the publicly available scientific and technological literature in the knowledge subgraph into a publicly available scientific and technological literature attribute embedding vector, and splice the publicly available scientific and technological literature attribute embedding vector and the publicly available scientific and technological literature word embedding vector into a node feature vector as the initial representation of the input node of the training data set of the multi-subgraph convolutional network model, that is, map the publicly available scientific and technological literature attribute embedding vector into the publicly available scientific and technological literature word embedding vector space, and splice the mapped publicly available scientific and technological literature attribute embedding vector and the publicly available scientific and technological literature word embedding vector to obtain a node feature vector;

[0039] In the multi-subgraph convolutional network model, for each input node, sample a number of neighbor nodes from its neighbor nodes in a random or fixed manner to form a feature subgraph. Each feature subgraph corresponds to a technical field, and use the aggregated neighbor node representation to update the representation of the central node through a convolution operation. The formula is as follows:

[0040] e ij =LeakyReLU(a T [Wui ||Wu j )

[0041]

[0042] Among them, u i and u j are the feature representations of the end nodes i and j of the current input layer, W is the weight matrix, || is the concatenation operation between vectors, e ij is the original attention score between nodes i and j, a ij is the attention score between nodes i and j, a T is the transpose of the weight matrix W;

[0043] For a single node u in the multi-subgraph convolutional network model i , the cross-entropy loss is used as the loss function l i to train the multi-subgraph convolutional network model to obtain a scientific and technical literature classification model:

[0044]

[0045] Among them, y i is the technical field to which the knowledge subgraph of node u i belongs in the preset knowledge graph, is the technical field to which the feature subgraph clustered by the multi-subgraph convolutional network model for node u i belongs. After training, a scientific and technical literature classification model is obtained.

[0046] Furthermore, the specific method for calculating the technical maturity of the corresponding publicly available scientific and technical literature subset according to the citation relationship and similarity between publicly available scientific and technical literatures in each publicly available scientific and technical literature subset is as follows:

[0047] For a certain publicly available scientific and technical literature subset, a time interval T t is defined to represent the time period t, where t is the time interval serial number and is sorted in chronological order. A time interval pair (T c , T d ) is defined to represent the time period between time interval T c and time interval T d , where the time when time interval T c is earlier than the time when time interval T d . In the time period specified by the time interval pair (T c , T d ) in this publicly available scientific and technical literature subset, the scientific and technical literatures published in time interval c are used to form the first publicly available scientific and technical literature set Pat c, the scientific and technological documents published within the time interval d form the second set of scientific and technological documents Pat d , if there is a citation relationship between a certain scientific and technological document in the first set of publicly disclosed scientific and technological documents Patc and a certain scientific and technological document in the second set of publicly disclosed scientific and technological documents Patd, then the pair of the two publicly disclosed scientific and technological documents with the citation relationship is put into the third set of publicly disclosed scientific and technological documents Cit(c,d);

[0048] According to the third set of scientific and technological documents Cit(c,d), the technology maturity evaluation technology life cycle is used, and the formula for technology maturity is as follows:

[0049]

[0050] In the formula, the maturity I of the technology j represents the technology maturity of the technical field to which the subset of publicly disclosed scientific and technological documents obtained by the scientific and technological document classification model belongs in the j-th time interval, and θ m represents the weight of the m-th evaluation index, and X m represents the index value of the m-th evaluation technology life cycle. The index value X of the evaluation technology life cycle m includes: the average value MeaN of the cosine similarity between publicly disclosed scientific and technological documents with citation relationships, the median MedN of the cosine similarity between publicly disclosed scientific and technological documents with citation relationships, the number of citations and the number of cited documents NCiteP, and the number of citations NCitiP as the evaluation indexes of the technology life cycle; among them, the number of citations and the number of cited documents NCiteP are used to represent the number of publicly disclosed scientific and technological documents with citation relationships of a certain publicly disclosed scientific and technological document within the time interval T c and within the time interval T d for a certain publicly disclosed scientific and technological document, and the number of citations NCitiP is used to represent the number of publicly disclosed scientific and technological documents with citation relationships of a certain publicly disclosed scientific and technological document before the publication date of this publicly disclosed scientific and technological document within the time interval d;

[0051] Set the technology maturity threshold, take the technical field with a technology maturity less than the technology maturity threshold as the innovative technical field, calculate the technology maturity of the subset of publicly disclosed scientific and technological documents, and predict whether the technical field to which it belongs is an innovative technical field.

[0052] A computer-readable medium, on which a computer program / instructions are stored, and the computer program / instructions execute the above-mentioned innovative technology prediction method based on the knowledge graph and the graph convolutional neural network when running.

[0053] A computer program product includes a computer program / instructions, and when the computer program / instructions are executed by a processor, the innovative technology prediction method based on a knowledge graph and a graph convolutional neural network described above is implemented.

[0054] The beneficial effects of the present invention are as follows:

[0055] 1. A knowledge graph can effectively organize and represent complex relationships within a technical field, including dependencies between technologies, substitution relationships, and their connections with external factors such as market demands and social events. As a powerful graph data analysis tool, a graph convolutional neural network can learn deep features and patterns of technological development from the knowledge graph, and thus achieve accurate prediction of technological trends. Based on the knowledge graph, by using a graph convolutional neural network to deeply explore the potential laws and future trends of technological development, the accuracy and foresight of innovative technology prediction can be effectively improved.

[0056] 2. By dividing a preset knowledge graph into multiple knowledge subgraphs, each knowledge subgraph focuses on the publicly available scientific and technological literature in the same technical field. This enables the graph convolutional neural network to more deeply understand and capture the knowledge and relationships within a specific technical field. This subdivision of fields helps the graph convolutional neural network to more accurately learn the characteristics and patterns of each technical field during the training process.

[0057] 3. The multi-subgraph convolutional network model is a type of graph convolutional neural network. The multi-subgraph convolutional network model samples a number of neighbor nodes from neighbor nodes in a random or fixed manner to form a feature subgraph. This flexibility enables the multi-subgraph convolutional network model to adapt to knowledge graphs of different scales and densities. The multi-subgraph convolutional network model can capture the complex relationships between nodes and improve the prediction accuracy by aggregating the representations of neighbor nodes and updating the representation of the central node.

[0058] 4. Since the scientific and technological literature classification model is constructed based on a knowledge graph and deep learning technologies, it has good generalization ability and can adapt to the patent classification requirements of different fields and scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a system block diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] In order to make the technical problems, technical solutions, and beneficial effects to be solved by the present application clearer and more understandable, the present application 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 only used to explain the present application and are not used to limit the present application.

[0061] Embodiment 1

[0062] Reference Figure 1, a technology maturity calculation system based on a knowledge graph and a graph convolutional neural network, comprising:

[0063] A scientific and technological literature classification module, configured to train a graph neural network according to a preset knowledge graph to obtain a scientific and technological literature classification model, and use the scientific and technological literature classification model to classify a selected set of publicly available scientific and technological literature into technical fields, so as to obtain subsets of publicly available scientific and technological literature corresponding to different technical fields. The preset knowledge graph is obtained according to publicly available scientific and technological literature, and the publicly available scientific and technological literature includes publicly available patents and publicly available papers;

[0064] A technology maturity calculation module, configured to calculate the technology maturity of the technical field to which the corresponding subset of publicly available scientific and technological literature belongs according to the citation relationship and similarity among the publicly available scientific and technological literature in each subset of publicly available scientific and technological literature.

[0065] Among them, the preset knowledge graph reflects the attributes of publicly available patents and papers, their mutual relationships, etc.

[0066] The knowledge graph can effectively organize and represent the complex relationships within a technical field, including the dependencies and substitution relationships between technologies and their connections with external factors such as market demands and social events. As a powerful graph data analysis tool, the graph convolutional neural network can learn the deep features and patterns of technological development from the knowledge graph, and then achieve accurate prediction of technological trends. Based on the knowledge graph, using the graph convolutional neural network to deeply explore the potential laws and future trends of technological development can effectively improve the accuracy and forward-looking of technological prediction. Since the model is constructed based on the knowledge graph and deep learning technology, it has good generalization ability and can adapt to the patent classification needs of different fields and scenarios.

[0067] (1) In the scientific and technological literature classification module, the specific method for training the graph neural network according to the preset knowledge graph to obtain the scientific and technological literature classification model is:

[0068] The graph neural network is a multi-subgraph convolutional network model;

[0069] Divide the preset knowledge graph into multiple knowledge sub - graphs. The publicly available scientific and technical documents in each knowledge sub - graph belong to the same technical field. Use a natural language processing model to convert the text of each publicly available scientific and technical document in the knowledge sub - graph into a publicly available scientific and technical document word embedding vector. Then, use a knowledge graph embedding model based on translational distance to convert the attributes of the publicly available scientific and technical documents in the knowledge sub - graph into publicly available scientific and technical document attribute embedding vectors, and splice the publicly available scientific and technical document attribute embedding vectors and the publicly available scientific and technical document word embedding vectors into node feature vectors as the initial representation of the input nodes of the training data set of the multi - sub - graph convolutional network model. That is, map the publicly available scientific and technical document attribute embedding vectors into the publicly available scientific and technical document word embedding vector space, and splice the mapped publicly available scientific and technical document attribute embedding vectors and the publicly available scientific and technical document word embedding vectors to obtain node feature vectors;

[0070] In the multi - sub - graph convolutional network model, for each input node, sample a number of neighbor nodes from its neighbor nodes in a random or fixed manner to form a feature sub - graph. Each feature sub - graph corresponds to a technical field. Use the aggregated neighbor node representation to update the representation of the central node through a convolution operation. The formula is as follows:

[0071] e ij = LeakyReLU(a T [Wu i ||Wu j )

[0072]

[0073] where u i and u j are the feature representations of the end node i and node j of the current input layer, W is the weight matrix, || is the splicing operation between vectors, e ij is the original attention score between node i and node j, a ij is the attention score between node i and node j, and a T is the transpose of the weight matrix W;

[0074] For a single node u i in the multi - sub - graph convolutional network model, use the cross - entropy loss as the loss function l i to train the multi - sub - graph convolutional network model to obtain a scientific and technical document classification model:

[0075]

[0076] where y i is the technical field to which the knowledge sub - graph of node u i belongs in the preset knowledge graph, is the technical field to which the feature subgraph obtained by clustering node u by the multi-subgraph convolutional network model belongs. After training, a scientific and technological literature classification model is obtained. i By dividing the preset knowledge graph into multiple knowledge subgraphs, each knowledge subgraph focuses on the publicly available scientific and technological literature in the same technical field. This enables the graph convolutional neural network to more deeply understand and capture the knowledge and relationships within a specific technical field. This subdivision of fields helps the graph convolutional neural network to more accurately learn the characteristics and patterns of each technical field during the training process. The multi-subgraph convolutional network model is a type of graph convolutional neural network. The multi-subgraph convolutional network model samples a number of neighbor nodes from the neighbor nodes in a random or fixed manner to form a feature subgraph. This flexibility enables the multi-subgraph convolutional network model to adapt to knowledge graphs of different scales and densities. The multi-subgraph convolutional network model aggregates the representations of neighbor nodes and updates the representation of the central node. The model can capture the complex relationships between nodes and improve the accuracy of predictions. By concatenating the publicly available scientific and technological literature attribute embedding vector and the publicly available scientific and technological literature word embedding vector into a node feature vector, this concatenation enables the model to comprehensively consider multiple information sources and improve the accuracy and robustness of predictions. Using the cross-entropy loss function to train the multi-subgraph convolutional network model can enable the model to continuously optimize its prediction ability during the training process until a relatively high classification accuracy is achieved.

[0077]

[0078] (2) In the technical maturity calculation module, the specific method for calculating the technical maturity of the technical field to which the corresponding publicly available scientific and technological literature subset belongs according to the citation relationship and similarity between the publicly available scientific and technological literatures in each publicly available scientific and technological literature subset is as follows:

[0079] For a certain publicly available scientific and technological literature subset, define the time interval T t to represent the time period t, where t is the time interval serial number and is sorted in chronological order. Define the time interval pair (T c , T d ) to represent the time period from time interval T c to time interval T d , where the time interval T c is earlier than the time interval T d . In the time period specified by the time interval pair (T c , T d ) in this publicly available scientific and technological literature subset, the scientific and technological literatures published in time interval c form the first publicly available scientific and technological literature set Pat c , and the scientific and technological literatures published in time interval d form the second scientific and technological literature set Pat d ​, if there is a citation relationship between a certain publicly disclosed scientific and technological document in the first set of publicly disclosed scientific and technological documents Patc and a certain publicly disclosed scientific and technological document in the second set of publicly disclosed scientific and technological documents Patd, then the pair of the two publicly disclosed scientific and technological documents with the citation relationship is placed in the third set of publicly disclosed scientific and technological documents Cit(c, d);

[0080] According to the third set of scientific and technological documents Cit(c, d), use the technology maturity evaluation technology life cycle, and the formula for technology maturity is as follows:

[0081]

[0082] In the formula, the maturity I of the technology j represents the technology maturity of the technical field to which the subset of publicly disclosed scientific and technological documents obtained by the scientific and technological document classification model belongs in the jth time interval, and θ m represents the weight of the mth evaluation index, and X m represents the index value of the mth evaluation technology life cycle. The index value X of the evaluation technology life cycle m includes: the average value MeaN of the cosine similarity between publicly disclosed scientific and technological documents with citation relationships, the median MedN of the cosine similarity between publicly disclosed scientific and technological documents with citation relationships, the number of citations and the number of cited documents NCiteP, and the number of citations NCitiP as evaluation indexes of the technology life cycle; among them, the number of citations and the number of cited documents NCiteP is used to represent the number of publicly disclosed scientific and technological documents with citation relationships of a certain publicly disclosed scientific and technological document within the time interval T c and within the time interval T d for a certain publicly disclosed scientific and technological document, and the number of citations NCitiP is used to represent the number of publicly disclosed scientific and technological documents with citation relationships of a certain publicly disclosed scientific and technological document before the publication date of this publicly disclosed scientific and technological document within the time interval d;

[0083] Set the technology maturity threshold, regard the technical field with a technology maturity less than the technology maturity threshold as an innovative technical field, calculate the technology maturity of the subset of publicly disclosed scientific and technological documents, and predict whether the technical field to which it belongs is an innovative technical field.

[0084] The publicly disclosed scientific and technological document is a publicly disclosed patent, and the calculation method of the cosine similarity of the publicly disclosed patent is as follows:

[0085] Convert the IPC international patent classification number of each patent into an IPC vector, which is expressed as follows:

[0086] IPC x =[a 1 , a 2 , …, a n

[0087] Among them, a n represents the nth category in the IPC international patent classification number. The value of a is 0 or 1. When a is 1, it means that patent x belongs to the nth category. When a is 0, it means that patent x does not belong to the nth category. The level of category a n is one of section, subclass, main group, subgroup;

[0088] Input the IPC vector into the BERT model for multi-label classification tasks to obtain the public patent feature vector;

[0089] Calculate the cosine similarity score PITS(a, b) between two public patents. The formula is as follows:

[0090]

[0091] Among them, PITVa and PITVb represent the feature vectors of public patents a and b respectively, and ‖PITVa‖·‖PITVb‖ represents the two-norm between public patents a and b.

[0092] Evaluate the technology maturity by comprehensively considering multiple evaluation indicators, including the average and median of the cosine similarity between publicly available scientific and technical documents with citation relationships, the number of citations and the number of times cited, and use quantitative indicators to evaluate the technology maturity, avoiding the interference of subjective judgment. These indicators reflect the innovation, influence and development trend of the technology from different angles, so as to evaluate the maturity of the technology field more comprehensively and synthetically.

[0093] Set the time interval pair (T c , T d ) to dynamically examine the change of technology field maturity. As time goes by, new patents emerge continuously, and the citation relationships of technologies are also changing continuously. By regularly updating the patent set and calculating the technology maturity, the development dynamics and trends of the technology field can be reflected in real time, providing timely and accurate information support for decision-makers.

[0094] When calculating the cosine similarity, convert the IPC international patent classification number into an IPC vector, so that the classification information of the patent can be represented efficiently and compactly. This representation method not only retains the classification characteristics of the patent, but also facilitates the subsequent calculation and processing of the cosine similarity. Specifically, the IPC vector usually contains multiple categories, forming high-dimensional data, and the cosine similarity has good robustness when dealing with high-dimensional data, which can reduce the influence of noise and redundant information on the similarity calculation.

[0095] Example 2

[0096] An innovative technology prediction method based on a knowledge graph and a graph convolutional neural network, comprising the following steps

[0097] Train a graph neural network according to a preset knowledge graph to obtain a scientific and technological literature classification model, and use the scientific and technological literature classification model to classify the selected set of publicly available scientific and technological literature into different technical fields to obtain subsets of publicly available scientific and technological literature corresponding to different technical fields. The preset knowledge graph is obtained from publicly available scientific and technological literature, and the publicly available scientific and technological literature includes publicly available patents and publicly available papers;

[0098] Calculate the technological maturity of the technical field to which each subset of publicly available scientific and technological literature belongs according to the citation relationship and similarity among the publicly available scientific and technological literature in each subset of publicly available scientific and technological literature.

[0099] The knowledge graph can effectively organize and represent the complex relationships within a technical field, including the dependencies between technologies, substitution relationships, and their connections with external factors such as market demand and social events. As a powerful graph data analysis tool, the graph convolutional neural network can learn the deep features and patterns of technological development from the knowledge graph, and then achieve accurate prediction of technological trends. Based on the knowledge graph, using the graph convolutional neural network to deeply explore the potential laws and future trends of technological development can effectively improve the accuracy and foresight of technology prediction. Since the model is constructed based on the knowledge graph and deep learning technology, it has good generalization ability and can adapt to the patent classification needs of different fields and scenarios.

[0100] (1) The specific method for training the graph neural network according to the preset knowledge graph to obtain a scientific and technological literature classification model is as follows:

[0101] The graph neural network is a multi-subgraph convolutional network model;

[0102] Divide the preset knowledge graph into multiple knowledge subgraphs. The publicly available scientific and technological literature in each knowledge subgraph belongs to the same technical field. Use a natural language processing model to convert the text of each publicly available scientific and technological literature in the knowledge subgraph into a publicly available scientific and technological literature word embedding vector, and then use a knowledge graph embedding model based on translational distance to convert the publicly available scientific and technological literature attributes in the knowledge subgraph into publicly available scientific and technological literature attribute embedding vectors, and splice the publicly available scientific and technological literature attribute embedding vectors and the publicly available scientific and technological literature word embedding vectors into a node feature vector as the input node initial representation of the training data set of the multi-subgraph convolutional network model, that is, map the publicly available scientific and technological literature attribute embedding vectors into the publicly available scientific and technological literature word embedding vector space, and splice the mapped publicly available scientific and technological literature attribute embedding vectors and the publicly available scientific and technological literature word embedding vectors to obtain a node feature vector;

[0103] In a multi-subgraph convolutional network model, for each input node, a number of neighbor nodes are sampled from its neighbor nodes in a random or fixed manner to form a feature subgraph. Each feature subgraph corresponds to a technical field. The representation of the central node is updated through a convolution operation using the aggregated neighbor nodes. The formula is as follows:

[0104] e ij = LeakyReLU(a T [Wu i ||Wu j )

[0105]

[0106] Where u i and u j are the feature representations of the end nodes i and j of the current input layer, W is the weight matrix, || is the concatenation operation between vectors, e ij is the original attention score between nodes i and j, a ij is the attention score between nodes i and j, and a T is the transpose of the weight matrix W;

[0107] For a single node u i in the multi-subgraph convolutional network model, the cross-entropy loss is used as the loss function l i to train the multi-subgraph convolutional network model to obtain a scientific and technical literature classification model:

[0108]

[0109] Where y i is the technical field to which the knowledge subgraph of node u i belongs in the preset knowledge graph, is the technical field to which the feature subgraph clustered by the multi-subgraph convolutional network model for node u i belongs. After training, a scientific and technical literature classification model is obtained.

[0110] By dividing the preset knowledge graph into multiple knowledge subgraphs, each knowledge subgraph focusing on the published scientific and technological literature in the same technical field, this enables the graph convolutional neural network to more deeply understand and capture the knowledge and relationships within a specific technical field. This subdivision of fields helps the graph convolutional neural network to more accurately learn the characteristics and patterns of each technical field during the training process. The multi-subgraph convolutional network model is a type of graph convolutional neural network. The multi-subgraph convolutional network model samples a number of neighbor nodes from the neighbor nodes in a random or fixed manner to form a feature subgraph. This flexibility enables the multi-subgraph convolutional network model to adapt to knowledge graphs of different scales and densities. The multi-subgraph convolutional network model can capture the complex relationships between nodes and improve the prediction accuracy by aggregating the representations of neighbor nodes and updating the representation of the central node. By concatenating the published scientific and technological literature attribute embedding vectors and the published scientific and technological literature word embedding vectors into a node feature vector, this concatenation mechanism enables the model to comprehensively consider multiple information sources and improve the accuracy and robustness of the prediction. Training the multi-subgraph convolutional network model using the cross-entropy loss function can enable the model to continuously optimize its prediction ability during the training process until a relatively high classification accuracy is achieved.

[0111] (2) The specific method for calculating the technological maturity of the technical field to which the corresponding published scientific and technological literature subset belongs according to the citation relationship and similarity between the published scientific and technological literatures in each published scientific and technological literature subset is as follows:

[0112] For a certain published scientific and technological literature subset, define the time interval T t to represent the time period t, where t is the time interval serial number and is sorted in chronological order. Define the time interval pair (T c , T d ) to represent the time period from time interval T c to time interval T d , where the time interval T c is earlier than the time interval T d . In the time period specified by the time interval pair (T c , T d ) for this published scientific and technological literature subset, form the first published scientific and technological literature set Pat c with the scientific and technological literatures published in time interval c, and form the second scientific and technological literature set Pat d with the scientific and technological literatures published in time interval d. If a certain published scientific and technological literature in the first published scientific and technological literature set Patc and a certain published scientific and technological literature in the second published scientific and technological literature set Patd have a citation relationship, then put the pair of the two published scientific and technological literatures with the citation relationship into the third published scientific and technological literature set Cit(c, d);

[0113] According to the third scientific and technological literature collection Cit(c, d), the technology maturity evaluation technology life cycle is used, and the formula for technology maturity is as follows:

[0114]

[0115] In the formula, the technology maturity I j represents the technology maturity of the technical field to which the subset of public scientific and technological literature obtained by the scientific and technological literature classification model belongs in the j-th time interval, and θ m represents the weight of the m-th evaluation index, and X m represents the index value of the m-th evaluation technology life cycle. The index value X of the evaluation technology life cycle m includes: the average value MeaN of the cosine similarity between public scientific and technological literatures with citation relationships, the median MedN of the cosine similarity between public scientific and technological literatures with citation relationships, the number of citations NCiteP, and the number of citations NCitiP as evaluation indexes of the technology life cycle; among them, the number of citations NCiteP is used to represent the number of public scientific and technological literatures with citation relationships of a certain public scientific and technological literature within the time interval T c and the number of citations NCitiP is used to represent the number of public scientific and technological literatures with citation relationships of a certain public scientific and technological literature before the publication date of this public scientific and technological literature within the time interval d; d

[0116] Set the technology maturity threshold, regard the technical field with a technology maturity less than the technology maturity threshold as an innovative technology field, calculate the technology maturity of the subset of public scientific and technological literature, and predict whether the technical field to which it belongs is an innovative technology field.

[0117] The public scientific and technological literature is a published patent, and the calculation method of the cosine similarity of the published patent is as follows:

[0118] Convert the IPC international patent classification number of each patent into an IPC vector, which is expressed as follows:

[0119] IPC x = [a 1 , a 2 , …, a n

[0120] Among them, a n represents the n-th category in the IPC international patent classification number. The value of a is 0 or 1. When a is 1, it means that patent x belongs to the n-th category. When a is 0, it means that patent x does not belong to the n-th category. The category a nThe level is one of section, large class, subclass, main group, and subgroup;

[0121] Input the IPC vector into the BERT model for multi-label classification tasks to obtain the public patent feature vector;

[0122] Calculate the cosine similarity score PITS(a, b) between two public patents. The formula is as follows:

[0123]

[0124] Where PITVa and PITVb represent the feature vectors of public patent a and public patent b respectively, and ‖PITVa‖·‖PITVb‖ represents the two-norm between public patent a and public patent b.

[0125] Evaluate the technology maturity by comprehensively considering multiple evaluation indicators, including the average and median of the cosine similarity between publicly available scientific and technical literature with citation relationships, the number of citations, the number of times cited, and the number of citations. Use quantitative indicators to evaluate the technology maturity, avoiding the interference of subjective judgment. These indicators reflect the innovation, influence, and development trend of the technology from different perspectives, so as to be able to evaluate the maturity of the technology field more comprehensively and integrally.

[0126] Set a time interval pair (T c , T d ) to dynamically examine the change of the technology field maturity. As time goes by, new patents emerge continuously, and the citation relationships of technologies are also changing continuously. By regularly updating the patent set and calculating the technology maturity, the development dynamics and trends of the technology field can be reflected in real time, providing timely and accurate information support for decision-makers.

[0127] When calculating the cosine similarity, convert the IPC international patent classification number into an IPC vector so that the classification information of the patent can be represented efficiently and compactly. This representation method not only retains the classification features of the patent but also facilitates the subsequent calculation and processing of the cosine similarity. Specifically, the IPC vector usually contains multiple categories, forming high-dimensional data, and the cosine similarity has good robustness when dealing with high-dimensional data, which can reduce the influence of noise and redundant information on the similarity calculation.

[0128] Example 3

[0129] A computer-readable medium, on which a computer program is stored, and the computer program executes the innovation technology prediction method based on the knowledge graph and graph convolutional neural network in Example 2 when running.

[0130] Example 4

[0131] A computer program product includes a computer program / instructions, and when the computer program / instructions are executed by a processor, an innovative technology prediction method based on a knowledge graph and a graph convolutional neural network in Embodiment 2 is implemented.

[0132] Contents not described in detail in this specification belong to the prior art well-known to those skilled in the art. Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0134] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0135] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, so that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the scope of its protection. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that after reading the present invention, various changes, modifications or equivalent substitutions can still be made to the specific implementation manners of the invention. However, these changes, modifications or equivalent substitutions are all within the scope of protection of the pending claims of the invention.

Claims

1. A technology maturity calculation system based on knowledge graph and graph convolutional neural network, characterized in that: include: A scientific and technological literature classification module is used to train the graph neural network according to a preset knowledge graph to obtain a scientific and technological literature classification model, and use the scientific and technological literature classification model to classify the selected public scientific and technological literature collection in technical fields to obtain subsets of public scientific and technological literature corresponding to different technical fields; The technology maturity calculation module is used to calculate the technology maturity of the technical field to which the corresponding public scientific and technological document subset belongs based on the citation relationship and similarity between the public scientific and technological documents in each public scientific and technological document subset.

2. The technology maturity calculation system based on knowledge graph and graph convolutional neural network according to claim 1, characterized in that: In the scientific and technological literature classification module, the specific method of training the graph neural network according to the preset knowledge graph to obtain the scientific and technological literature classification model is as follows: The graph neural network is a multi-subgraph convolutional network model; Divide the preset knowledge graph into multiple knowledge subgraphs, the public scientific and technological documents in each knowledge subgraph belong to the same technical field, use the natural language processing model to convert the text of each public scientific and technological document in the knowledge subgraph into a public scientific and technological document word embedding vector, then use the knowledge graph embedding model based on translation distance to convert the public scientific and technological document attributes in the knowledge subgraph into a public scientific and technological document attribute embedding vector, and concatenate the public scientific and technological document attribute embedding vector and the public scientific and technological document word embedding vector into a node feature vector as the input node initial representation of the training data set of the multi-subgraph convolutional network model; In the multi-subgraph convolutional network model, for each input node, several neighbor nodes are randomly or fixedly sampled from its neighbor nodes to form a feature subgraph. Each feature subgraph corresponds to a technical field. The representation of the central node is updated through the convolution operation using the aggregated neighbor node representation. The formula is as follows: and ij =LeakyReLU(a T [Wu i ||Wu j ]) Among them, u i and u j is the feature representation of the end node i and node j of the current input layer, W is the weight matrix, || is the concatenation operation between vectors, e ij is the raw attention score between node i and node j, a ij is the attention score between node i and node j, a T is the transpose of the weight matrix W; For a single node u in the multi-subgraph convolutional network model i , using cross entropy loss as the loss function l i The multi-subgraph convolutional network model is trained to obtain the scientific and technological literature classification model: Among them, y i is node u i The technical field to which the knowledge subgraph in the preset knowledge graph belongs, is the multi-subgraph convolutional network model for node u i The technical field to which the characteristic subgraphs obtained by clustering belong is trained to obtain a scientific and technological literature classification model.

3. The technology maturity calculation system based on knowledge graph and graph convolutional neural network according to claim 1, characterized in that: In the technology maturity calculation module, the specific method for calculating the technology maturity of the technical field to which the corresponding public scientific and technological document subset belongs according to the citation relationship and similarity between the public scientific and technological documents in each public scientific and technological document subset is: For a subset of public scientific and technological literature, define the time interval T t Represents time period t, t is the time interval serial number and is sorted in chronological order. Define the time interval pair (T c ,T d ) represents the time interval T c To time interval T d The time period between c The time is earlier than the time interval T d , in which the subset of public scientific and technical literature is in the time interval (T c ,T d ), the scientific and technological documents published in the time interval c constitute the first public scientific and technological document set Pat c , the scientific and technological documents published in the time interval d constitute the second scientific and technological document set Pat d , if a public scientific and technological document in the first public scientific and technological document set Patc and a public scientific and technological document in the second public scientific and technological document set Patd have a citation relationship, then the two public scientific and technological document pairs with the citation relationship are put into the third public scientific and technological document set Cit(c,d); According to the third scientific and technological literature set Cit(c,d), the formula for technology maturity is as follows: In the formula, the maturity of technology I j represents the technological maturity of the technical field of the subset of public scientific and technological documents obtained by the scientific and technological document classification model in the jth time interval, θ m represents the weight of the mth evaluation index, X m represents the index value of the mth evaluation technology life cycle, the index value of the evaluation technology life cycle X m Including: the average value MeaN of cosine similarity between public scientific and technological documents with citation relationship, the median MedN of cosine similarity between public scientific and technological documents with citation relationship, the number of citations NCiteP, the number of citations NCitiP as indicators for evaluating technology life cycle; among them, the number of citations NCiteP is used to represent the number of citations in the time interval T c A certain public scientific and technological document within a time interval T d The number of public scientific and technological documents with citation relationships in the time interval d is NCitiP, and the number of citations NCitiP is used to indicate the number of public scientific and technological documents with citation relationships between a public scientific and technological document and the date before the publication of the public scientific and technological document within the time interval d.

4. The technology maturity calculation system based on knowledge graph and graph convolutional neural network according to claim 3 is characterized by: The public scientific and technological documents are public patents, and the calculation method of the cosine similarity of the public patents is as follows: The IPC international patent classification number of each patent is converted into an IPC vector, which is expressed as follows: IPC x =[a1,a2,…,a n ] Among them, a n Indicates the nth category in the IPC International Patent Classification Number. The value of a is 0 or 1. When a is 1, it means that patent x belongs to the nth category. When a is 0, it means that patent x does not belong to the nth category. Category a n The level is one of department, major category, minor category, large group, and small group; The IPC vector is input into the BERT model for multi-label classification tasks to obtain the public patent feature vector; The cosine similarity score PITS(a,b) between two published patents is calculated as follows: Among them, PITVa and PITVb represent the feature vector of public patent a and public patent b respectively, and ‖PITVa‖·‖PITVb‖ represents the binary norm between public patent a and public patent b.

5. An innovative technology prediction method based on knowledge graph and graph convolutional neural network, characterized in that: The following steps are involved: The graph neural network is trained according to the preset knowledge graph to obtain a scientific and technological literature classification model, and the scientific and technological literature classification model is used to classify the selected public scientific and technological literature collection into technical fields to obtain subsets of public scientific and technological literature corresponding to different technical fields; The technology maturity of the technical field to which the corresponding public scientific and technological document subset belongs is calculated based on the citation relationship and similarity between the public scientific and technological documents in each public scientific and technological document subset.

6. The technology prediction method based on knowledge graph and graph convolutional neural network according to claim 5, characterized in that: The specific method of training the graph neural network according to the preset knowledge graph to obtain the scientific and technological literature classification model is: The graph neural network is a multi-subgraph convolutional network model; Divide the preset knowledge graph into multiple knowledge subgraphs, the public scientific and technological documents in each knowledge subgraph belong to the same technical field, use the natural language processing model to convert the text of each public scientific and technological document in the knowledge subgraph into a public scientific and technological document word embedding vector, then use the knowledge graph embedding model based on translation distance to convert the public scientific and technological document attributes in the knowledge subgraph into a public scientific and technological document attribute embedding vector, and concatenate the public scientific and technological document attribute embedding vector and the public scientific and technological document word embedding vector into a node feature vector as the input node initial representation of the training data set of the multi-subgraph convolutional network model; In the multi-subgraph convolutional network model, for each input node, several neighbor nodes are randomly or fixedly sampled from its neighbor nodes to form a feature subgraph. Each feature subgraph corresponds to a technical field. The representation of the central node is updated through the convolution operation using the aggregated neighbor node representation. The formula is as follows: and ij =LeakReLU(a T [Wu i ||Wu j ]) Among them, u i and u j is the feature representation of the end node i and node j of the current input layer, W is the weight matrix, || is the concatenation operation between vectors, e ij is the raw attention score between node i and node j, a ij is the attention score between node i and node j, a T is the transpose of the weight matrix W; For a single node u in the multi-subgraph convolutional network model i , using cross entropy loss as the loss function l i The multi-subgraph convolutional network model is trained to obtain the scientific and technological literature classification model: Among them, y i is node u i The technical field to which the knowledge subgraph in the preset knowledge graph belongs, is the multi-subgraph convolutional network model for node u i The technical field to which the characteristic subgraphs obtained by clustering belong is trained to obtain a scientific and technological literature classification model.

7. The technology prediction method based on knowledge graph and graph convolutional neural network according to claim 5, characterized in that: The specific method for calculating the technology maturity of the technical field to which the corresponding public scientific and technological document subset belongs according to the citation relationship and similarity between the public scientific and technological documents in each public scientific and technological document subset is: For a subset of public scientific and technological literature, define the time interval T t Represents time period t, t is the time interval serial number and is sorted in chronological order. Define the time interval pair (T c ,T d ) represents the time interval T c To time interval T d The time period between c The time is earlier than the time interval T d , in which the subset of public scientific and technical literature is in the time interval (T c ,T d ), the scientific and technological documents published in the time interval c constitute the first public scientific and technological document set Pat c , the scientific and technological documents published in the time interval d constitute the second scientific and technological document set Pat d , if a public scientific and technological document in the first public scientific and technological document set Patc and a public scientific and technological document in the second public scientific and technological document set Patd have a citation relationship, then the two public scientific and technological document pairs with the citation relationship are put into the third public scientific and technological document set Cit(c,d); According to the third scientific and technological literature set Cit(c,d), the formula for technology maturity is as follows: In the formula, the maturity of technology I j represents the technological maturity of the technical field of the subset of public scientific and technological documents obtained by the scientific and technological document classification model in the jth time interval, θ m represents the weight of the mth evaluation index, X m represents the index value of the mth evaluation technology life cycle, the index value of the evaluation technology life cycle X m Including: the average value MeaN of cosine similarity between public scientific and technological documents with citation relationship, the median MedN of cosine similarity between public scientific and technological documents with citation relationship, the number of citations NCiteP, the number of citations NCitiP as indicators for evaluating technology life cycle; among them, the number of citations NCiteP is used to represent the number of citations in the time interval T c A certain public scientific and technological document within a time interval T d The number of public scientific and technological documents with citation relationships in the time interval d is NCitiP, and the number of citations NCitiP is used to indicate the number of public scientific and technological documents with citation relationships between a public scientific and technological document and the date before the publication of the public scientific and technological document within the time interval d.

8. The technology prediction method based on knowledge graph and graph convolutional neural network according to claim 7, characterized in that: The public scientific and technological documents are public patents, and the calculation method of the cosine similarity of the public patents is as follows: The IPC international patent classification number of each patent is converted into an IPC vector, which is expressed as follows: IPC x =[a1,a2,…,a n ] Among them, a n Indicates the nth category in the IPC International Patent Classification Number. The value of a is 0 or 1. When a is 1, it means that patent x belongs to the nth category. When a is 0, it means that patent x does not belong to the nth category. Category a n The level is one of department, major category, minor category, large group, and small group; The IPC vector is input into the BERT model for multi-label classification tasks to obtain the public patent feature vector; The cosine similarity score PITS(a,b) between two published patents is calculated as follows: Among them, PITVa and PITVb represent the feature vector of public patent a and public patent b respectively, and ‖PITVa‖·‖PITVb‖ represents the binary norm between public patent a and public patent b.

9. A computer-readable medium having a computer program / instruction stored thereon, wherein the computer program / instruction, when run, executes the technology prediction method based on knowledge graph and graph convolutional neural network as described in any one of claims 5 to 8.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by the processor, the technology prediction method based on the knowledge graph and graph convolutional neural network described in any one of claims 5 to 8 is implemented.

Images