Scientific and technological theme evolution stage prediction method and system based on multi-graph representation

Abstract

The invention discloses a science and technology theme evolution stage prediction method and system based on multi-graph representation. The method comprises the following steps: firstly, carrying out subject matching on science and technology subject keywords in science and technology literatures, screening out the science and technology subject keywords capable of representing a subject, then obtaining a word frequency sequence of the science and technology subject keywords, and extracting semantic association relationship characteristics among the science and technology subject keywords and reference relationship characteristics among the science and technology subject keywords; secondly, a science and technology theme keyword multi-graph network is constructed, and science and technology theme keywords representing themes are modeled; then, for features of different dimensions, different graph neural network models are adopted for feature extraction and modeling; and finally, predicting the evolution stage of the science and technology subject by using the prediction model, thereby solving the technical problem that the analysis granularity is too coarse in the research of the evolution of the science and technology subject in the prior art, and realizing the accurate prediction of the evolution process of the science and technology subject.

Description

Technical Field

[0001] The present invention relates to the technical field of scientific and technological theme evolution prediction, and in particular to a scientific and technological theme evolution stage prediction method and system based on multiple graph representations. Background Art

[0002] As science and technology themes evolve over time, they will show stage changes such as emergence, growth, integration, and decline. Accurately identifying and predicting the evolutionary stages of science and technology themes will help reveal the dynamic process of scientific knowledge production and dissemination, track the development trends of cutting-edge research hotspots and emerging technologies in the field, and thus more comprehensively judge the development trends of scientific and technological innovation in the field.

[0003] Tracking the evolution of scientific and technological topics is a long and slow process. However, in the current era of big data, with the increasing frequency of data interaction and the increasing complexity of scientific knowledge structures and relationships, it is difficult to fully grasp the context of topic evolution based on a single dimension. Summary of the invention

[0004] The present invention solves the technical problem of coarse analysis granularity in the prior art research on the evolution of scientific and technological themes by providing a method and system for predicting the evolution stages of scientific and technological themes based on multiple graph representations, and achieves accurate prediction of the evolution process of scientific and technological themes.

[0005] The present invention provides a method for predicting the evolution stage of scientific and technological topics based on multiple graph representations, comprising:

[0006] Obtain scientific and technological subject keywords in scientific and technological literature;

[0007] Obtaining a word frequency sequence of the scientific and technological topic keywords;

[0008] By formula Calculate the keywords of science and technology topics and technology-themed keywords The semantic distance between ij ;in, is the technology theme keyword in k years The word vector of is the technology theme keyword in k years The word vector of

[0009] Construct the adjacency matrix of the semantic features S of keywords within k years Where n is the number of all science and technology theme keywords in year k;

[0010] Obtain the reference relationship between the scientific and technological theme keywords, and obtain the reference frequency between the scientific and technological theme keywords Among them, count_num is the technology theme keyword in k years right The frequency of citations;

[0011] Construct the citation relationship adjacency matrix of each scientific and technological theme keyword

[0012] Construct a multigraph G = (V = {V w ,X},E={E s ,E c}); where V w represents the set of science and technology theme keyword nodes, X represents the word frequency sequence set of science and technology theme keyword nodes, In represents the word frequency sequence T corresponding to each technology theme keyword node, w represents the technology theme keyword node belonging to the multi-graph node set, E s Represents the adjacency matrix The semantic distance edge set between the keywords of each scientific and technological topic in E c Represents the adjacency matrix The set of citation relationship edges between the keywords of various scientific and technological topics;

[0013] The multi-graph G is input into the trained topic evolution stage prediction model, and the adjacency matrix is Perform graph convolution operations and obtain keywords for each technology topic Features after modeling Use GAT based on the self-attention mechanism to Modeling to obtain keywords for each technology topic Features after modeling The frequency sequence of the scientific and technological keywords is combined with the feature The features Perform splicing and fusion to obtain the fused features The characteristics Input to the MLP layer to predict the evolution stage of scientific and technological topics and obtain the final prediction results

[0014] Specifically, the acquisition of scientific and technological subject keywords in scientific and technological literature includes:

[0015] Combining the title and abstract corpus of the scientific and technological documents as the subject corpus of the scientific and technological documents;

[0016] The preset keywords are searched and matched in the subject corpus, and the keywords obtained by the search and matching are used as the scientific and technological subject keywords in the scientific and technological literature.

[0017] Specifically, obtaining the reference relationship between the scientific and technological subject keywords includes:

[0018] The citation relationship between the scientific and technological documents is obtained based on the detailed information of the scientific and technological documents, and the citation relationship between the keywords of the scientific and technological subjects is obtained through relationship mapping based on the citation relationship between the scientific and technological documents.

[0019] Specifically, after obtaining the word frequency sequence of the scientific and technological topic keywords, it also includes:

[0020] The word frequency sequences of the keywords of each scientific and technological topic are regarded as data points to perform shape distance calculation to construct an undirected weighted graph, thereby obtaining an adjacency matrix A;

[0021] Normalize the adjacency matrix A to obtain a similarity matrix W between graph vertices;

[0022] Add the elements of each column of the similarity matrix W and place them on the diagonal position to form a diagonal matrix to obtain a weighted degree matrix D;

[0023] Obtaining a Laplace matrix L according to the similarity matrix W and the weighted degree matrix D, and performing eigenvalue decomposition;

[0024] Taking the eigenvectors corresponding to the first λ smallest eigenvalues ​​of the Laplace matrix L to form a characteristic matrix H;

[0025] Clustering the feature matrix H to obtain cluster labels of corresponding word frequency sequences;

[0026] The frequency sequence of the science and technology topic keywords after clustering labels is input into the topic evolution stage prediction model to be trained and trained until convergence or the number of iterations is reached to obtain the trained topic evolution stage prediction model.

[0027] Specifically, the cross entropy loss is used to calculate the loss of the prediction of the evolution stage of scientific and technological topics. The cross entropy loss is

[0028] The present invention also provides a scientific and technological theme evolution stage prediction system based on multiple graph representation, comprising:

[0029] A science and technology subject keyword acquisition module is used to acquire science and technology subject keywords in scientific and technological literature;

[0030] A word frequency sequence acquisition module, used to acquire the word frequency sequence of the scientific and technological subject keywords;

[0031] The semantic distance calculation module is used to calculate the distance between Calculate the keywords of science and technology topics and technology-themed keywords The semantic distance betweenij ;in, is the technology theme keyword in k years The word vector of is the technology theme keyword in k years The word vector of

[0032] Semantic feature adjacency matrix construction module, used to construct the adjacency matrix of the semantic features S of keywords within k years Where n is the number of all science and technology theme keywords in year k;

[0033] The citation frequency acquisition module is used to obtain the citation relationship between the scientific and technological theme keywords and obtain the citation frequency between each of the scientific and technological theme keywords. Among them, count_num is the technology theme keyword in k years right The frequency of citations;

[0034] Citation relationship adjacency matrix construction module, used to construct the citation relationship adjacency matrix of each scientific and technological theme keyword

[0035] A multigraph construction module is used to construct a multigraph G = (V = {V w ,X},E={E s ,E c}); where V w represents the set of science and technology theme keyword nodes, X represents the word frequency sequence set of science and technology theme keyword nodes, In represents the word frequency sequence T corresponding to each technology theme keyword node, w represents the technology theme keyword node belonging to the multi-graph node set, E s Represents the adjacency matrix The semantic distance edge set between the keywords of each scientific and technological topic in E c Represents the adjacency matrix The set of citation relationship edges between the keywords of various scientific and technological topics;

[0036] The technology theme evolution stage prediction module is used to input the multi-graph G into the trained theme evolution stage prediction model, and use GCN to predict the adjacency matrix Perform graph convolution operations and obtain keywords for each technology topic Features after modeling Use GAT based on the self-attention mechanism to Modeling to obtain keywords for each technology topic Features after modeling The frequency sequence of the scientific and technological keywords is combined with the feature The features Perform splicing and fusion to obtain the fused features The characteristics Input to the MLP layer to predict the evolution stage of scientific and technological topics and obtain the final prediction results

[0037] Specifically, the science and technology theme keyword acquisition module includes:

[0038] A subject corpus building unit, used for combining the title and abstract corpus of the scientific and technological documents as the subject corpus of the scientific and technological documents;

[0039] The scientific and technological subject keyword query unit is used to query and match preset keywords in the subject corpus, and use the keywords obtained by the query and matching as the scientific and technological subject keywords in the scientific and technological literature.

[0040] Specifically, the citation frequency acquisition module includes:

[0041] A citation relationship acquisition unit is used to obtain the citation relationship between each scientific and technological document according to the detailed information of each scientific and technological document, and obtain the citation relationship between each scientific and technological subject keyword through relationship mapping according to the citation relationship between each scientific and technological document;

[0042] The citation frequency acquisition unit is used to obtain the citation frequency between each pair of scientific and technological theme keywords according to the citation relationship between the scientific and technological theme keywords. Among them, count_num is the technology theme keyword in k years right The frequency of citations.

[0043] Specifically, it also includes:

[0044] An adjacency matrix A construction module is used to construct an undirected weighted graph by treating the word frequency sequence of each scientific and technological theme keyword as a data point and performing shape distance calculation to obtain an adjacency matrix A;

[0045] A similarity matrix W obtaining module is used to normalize the adjacency matrix A to obtain a similarity matrix W between graph vertices;

[0046] A weighted degree matrix D obtaining module is used to add the elements of each column of the similarity matrix W and place them on the diagonal position to form a diagonal matrix to obtain the weighted degree matrix D;

[0047] An eigenvalue decomposition module, used to obtain a Laplace matrix L according to the similarity matrix W and the weighted degree matrix D, and perform eigenvalue decomposition;

[0048] A feature matrix H acquisition module is used to obtain the feature vectors corresponding to the first λ smallest eigenvalues ​​of the Laplace matrix L to form the feature matrix H;

[0049] A word frequency sequence labeling module is used to cluster the feature matrix H to obtain cluster labels of corresponding word frequency sequences;

[0050] The model training module is used to input the frequency sequence of the science and technology theme keywords after clustering labels into the theme evolution stage prediction model to be trained until convergence or the number of iterations is reached, so as to obtain the trained theme evolution stage prediction model.

[0051] Specifically, the cross entropy loss is used to calculate the loss of the prediction of the evolution stage of scientific and technological topics. The cross entropy loss is

[0052] One or more technical solutions provided in the present invention have at least the following technical effects or advantages:

[0053] First, the scientific and technological theme keywords in scientific and technological literature are subject-matched, and the scientific and technological theme keywords that can represent the theme are screened out. Then, the word frequency sequence of the scientific and technological theme keywords is obtained, and the semantic association relationship features between the scientific and technological theme keywords and the reference relationship features between the scientific and technological theme keywords are extracted. Secondly, a multi-graph network of scientific and technological theme keywords is constructed, and the scientific and technological theme keywords representing the theme are modeled. Subsequently, different graph neural network models are used for feature extraction and modeling for features of different dimensions. Finally, a prediction model is used to predict the evolution stage of scientific and technological themes, which solves the technical problem of coarse analysis granularity in the research on the evolution of scientific and technological themes in the prior art and realizes accurate prediction of the evolution process of scientific and technological themes. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A schematic diagram of a method for predicting the evolutionary stage of a scientific and technological theme based on multiple graph representations provided in an embodiment of the present invention;

[0055] Figure 2 A flowchart of a method for predicting the evolutionary stage of a scientific and technological theme based on multiple graph representations provided in an embodiment of the present invention;

[0056] Figure 3 A working principle diagram of a theme evolution stage prediction model in a technology theme evolution stage prediction method based on multiple graph representation provided in an embodiment of the present invention;

[0057] Figure 4 A module diagram of a scientific and technological theme evolution stage prediction system based on multiple graph representations provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The embodiments of the present invention solve the technical problem of coarse analysis granularity in the prior art research on the evolution of scientific and technological topics by providing a method and system for predicting the evolution stage of scientific and technological topics based on multiple graph representations, and achieve accurate prediction of the evolution process of scientific and technological topics.

[0059] The technical solution in the embodiment of the present invention is to solve the above technical problems, and the overall idea is as follows:

[0060] like Figure 1 As shown, the embodiment of the present invention is based on the prediction of the evolution stage of scientific and technological themes based on multiple graphs, which specifically includes the steps of scientific and technological theme extraction, scientific and technological theme stage discrimination, scientific and technological theme feature extraction, multiple graph modeling, and scientific and technological theme evolution stage prediction. First, extract the scientific and technological theme keywords that can represent the theme, and use the supervised learning model to judge the trend of the scientific and technological theme keywords after the theme extraction, and mark the trend label. Secondly, perform feature extraction of scientific and technological themes, specifically: calculate the word frequency of scientific and technological theme keywords to obtain the intensity characteristics of scientific and technological themes, and obtain the word embedding of scientific and technological theme keywords to extract the content characteristics of scientific and technological themes, and obtain the citation relationship of scientific and technological theme keywords based on the literature citation relationship mapping to extract the structural characteristics of scientific and technological themes. After obtaining the features of three dimensions, use the scientific and technological theme nodes and the features of three dimensions to perform multiple graph modeling of scientific and technological themes to realize the representation of multi-dimensional information. Finally, the extracted scientific and technological theme keywords representing the scientific and technological themes in the multiple graphs are predicted by the graph representation learning model and the regression model to obtain the trend of its next stage of evolution, thereby realizing the prediction of the evolution stage of scientific and technological themes.

[0061] Specifically, in the data preparation stage, it is necessary to collect subject-area literature from the scientific literature database as research data. First, data cleaning is achieved by removing abnormal data from the original scientific literature text data (abnormal data includes undefined and messy characters, empty data values, etc.). Secondly, it is necessary to extract scientific and technological themes. Specifically, for each document, the title and abstract corpus of the document are combined as the research subject corpus of the document, and then the scientific and technological subject keywords under the document are matched to check whether these keywords appear in the research subject corpus of the document, and the scientific and technological subject keywords that appear in the research subject corpus are selected to represent the scientific and technological themes in the documents in this field, and the scientific and technological subject keyword set M within each year k is obtained. k .

[0062] In order to make the subsequent semantic feature modeling more accurate, the obtained scientific and technological theme keywords are processed by camel case naming. For example, the scientific and technological theme keyword "Information System" is compressed and converted into "InformationSystem", and the scientific and technological theme keyword "Social Network Analysis" is converted into "SocialNetworkAnalysis".

[0063] In the topic strength feature extraction stage, the embodiment of the present invention extracts the frequency sequence of the science and technology topic keywords in multi-dimensional time as a main feature T of the science and technology topic. Specifically, the time window is set to 10, and the science and technology topic keywords within a specific year k are selected and collected. The word frequency sequence T over a ten-year time span is (t k-9 ,t k-8 ,t k-7 ,…,t k-1 ,t k ), using the word frequency sequence of ten consecutive years as the intensity feature of the science and technology theme, and obtaining rich temporal intensity information. By collecting word frequency information over a longer time span, it is helpful to comprehensively represent the intensity features of science and technology themes, making the identification and prediction of the theme evolution stage more accurate.

[0064] When extracting the subject content features, the specific process is as follows:

[0065] (1) Obtaining semantic word vectors for scientific topics

[0066] When obtaining word vectors, you first need to prepare a corpus. Taking the science and technology keywords of 2019 as an example, consistent with the time window of the theme intensity of 10, the embodiment of the present invention uses the titles and abstracts of all scientific and technological documents from 2010 to 2019 to form a corpus for 2019. Then perform specific word embedding operations to calculate the science and technology theme keywords. Semantic expression vector in the global context It is worth noting that the minimum word frequency MIN_WORD is set to 1 here in order to obtain the global semantic information of all scientific and technological topics in the corpus. The formula for obtaining word vectors is as follows:

[0067]

[0068] (2) Constructing the semantic content adjacency matrix

[0069] After obtaining the word embedding vector of each technology theme keyword, in order to construct the multigraph, it is necessary to obtain the semantic distance between technology theme keywords. Therefore, the cosine similarity between technology theme keyword word vectors is calculated to reflect the strength of the semantic relationship between them, and then the semantic distance adjacency matrix of technology theme keywords within unit year k is constructed.

[0070] The specific extraction process of topic citation features is as follows:

[0071] (1) Finding citation relationships between scientific and technological literature

[0072] First, we process the scientific literature data and extract the citation relationship between the literature in a given year k. For example, taking 2019 as an example, we find out which literature in the previous two years k-1 and k-2 is cited by the literature in the year k, so as to obtain the citation relationship between the literature in the three-year time window.

[0073] (2) Extend the citation relationship to scientific and technological topic keywords

[0074] After obtaining the citation relationship between documents within k years, this citation relationship is extended and mapped to the science and technology theme keywords. Specifically, for example, when document D1 cites document D2, all the science and technology theme keywords in document D1 also have a citation relationship with all the science and technology theme keywords in document D2. This mapping relationship belongs to a "keyword Cartesian product mapping". Finally, the citation frequency between all science and technology theme keywords is counted, that is, Among them, count_num is the technology theme keyword in year k right The frequency of citations.

[0075] (3) Constructing the reference structure adjacency matrix

[0076] After obtaining the citation relationship between the science and technology theme keywords, the adjacency matrix of the citation frequency between the science and technology theme keywords can be constructed.

[0077] The semantic content features S and citation structure features C of the above-mentioned science and technology theme keywords are actually based on the multi-graph structure between science and technology theme keywords, and two association relationship graphs are modeled in the content feature dimension and the structure feature dimension respectively.

[0078] Furthermore, a multi-graph G is constructed with science and technology theme keywords as nodes and semantic distance and citation frequency as edges.

[0079] After obtaining multiple graphs G, we use graph neural networks to learn graph features of different dimensions based on trend-labeled data sets. Specifically, we use GCN to learn the reference structure features of science and technology keywords, and use GAT to learn the semantic content features of science and technology keywords. We then concatenate the learned new feature representations with word frequency features for evolutionary stage prediction.

[0080] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0081] like Figure 2 As shown, the method for predicting the evolution stage of scientific and technological topics based on multiple graph representations provided by an embodiment of the present invention includes:

[0082] Step S110: Acquire scientific and technological subject keywords in scientific and technological documents;

[0083] This step is explained in detail to obtain the scientific and technological subject keywords in the scientific and technological literature, including:

[0084] Combine the title and abstract corpus of scientific and technological literature as the subject corpus of scientific and technological literature;

[0085] The preset keywords are queried and matched in the subject corpus, and the keywords obtained by the query matching are used as the scientific and technological subject keywords in the scientific and technological literature.

[0086] Step S120: Obtain the word frequency sequence of the science and technology theme keywords; using the word frequency sequence of the science and technology theme keywords as the theme strength directly measures the degree of attention paid to the science and technology theme, which is a key quantitative indicator reflecting the evolution stage of the science and technology theme.

[0087] Step S130: By formula Calculate the keywords of science and technology topics and technology-themed keywords The semantic distance between ij ;in, is the technology theme keyword in k years The word vector of is the technology theme keyword in k years The word vector of

[0088] Step S140: Construct an adjacency matrix of the semantic features S of keywords within k years Among them, n is the number of all science and technology theme keywords in year k; the semantic distance relationship between science and technology theme keywords is used as the content feature, which reflects the correlation relationship between text carriers in the changes of theme evolution stage.

[0089] Step S150: Obtain the reference relationship between each technology theme keyword, and obtain the reference frequency between each technology theme keyword Among them, count_num is the technology theme keyword in k years right The frequency of citations;

[0090] Specifically, obtain the reference relationship between keywords of various scientific and technological topics, including:

[0091] The citation relationship between scientific and technological documents is obtained based on the detailed information of each scientific and technological document, and the citation relationship between each scientific and technological subject keyword is obtained through relationship mapping based on the citation relationship between each scientific and technological document. The keyword citation frequency between scientific and technological subject keywords based on document citation is used as a structural feature to reflect the interactive relationship between scientific knowledge.

[0092] Take the science and technology theme keywords of 2019 as an example. After filtering and obtaining the science and technology theme keywords of 2019, query the science and technology theme keywords of 2019 Get the detailed information of the scientific and technological literature, obtain the DOI of the scientific and technological literature it cited, and then query whether the time of these cited scientific and technological literature ① is within the data collected in this field ② is within 2017-2019 (within 3 years). If ①② is met, the citation relationship between the keywords of each scientific and technological theme can be obtained through the logic of "Keyword—ArticleCitation—Keyword".

[0093] Step S160: Constructing the citation relationship adjacency matrix of each scientific and technological theme keyword

[0094] Step S170: Construct a multigraph G = (V = {V w ,X},E={E s ,E c}); where V w represents the set of science and technology theme keyword nodes, X represents the word frequency sequence set of science and technology theme keyword nodes, In represents the word frequency sequence T corresponding to each technology theme keyword node, w represents the technology theme keyword node belonging to the multi-graph node set, E s Represents the adjacency matrix The semantic distance edge set between the keywords of each scientific and technological topic in E c Represents the adjacency matrix The set of citation relationship edges between the keywords of various scientific and technological topics;

[0095] Step S180: Input the multi-graph G into the trained topic evolution stage prediction model, and use GCN to transform the adjacency matrix Perform graph convolution operations and obtain keywords for each technology topic Features after modeling The specific formula is Use GAT based on the self-attention mechanism to adjust the adjacency matrix Modeling to obtain keywords for each technology topic Features after modeling The specific formula is The frequency sequence and characteristics of scientific and technological keywords feature Perform splicing and fusion to obtain the fused features The specific formula is in, Indicates technology-themed keywords The word frequency sequence of Input to the MLP layer to predict the evolution stage of scientific and technological topics and obtain the final prediction results Right now like Figure 3 shown.

[0096] Specifically, the cross entropy loss is used to calculate the loss of the prediction of the evolution stage of scientific and technological topics. The cross entropy loss is

[0097] The following is a detailed description of the training process of the prediction model in the topic evolution stage. After obtaining the word frequency sequence of the scientific and technological topic keywords, it also includes:

[0098] The word frequency sequence of each scientific and technological theme keyword is regarded as a data point to calculate the shape distance to construct an undirected weighted graph and obtain the adjacency matrix A;

[0099] Specifically, the word frequency sequence data of each scientific and technological theme keyword is used as the vertex, and the dynamic time warping distance between each word frequency sequence is used as the edge weight to construct the adjacency matrix A.

[0100] Normalize the adjacency matrix A to obtain the similarity matrix W between graph vertices;

[0101] Add the elements of each column of the similarity matrix W and place them on the diagonal position to form a diagonal matrix to obtain the weighted degree matrix D;

[0102] The Laplace matrix L is obtained according to the similarity matrix W and the weighted degree matrix D, and the eigenvalue decomposition is performed;

[0103] Take the eigenvectors corresponding to the first λ smallest eigenvalues ​​of the Laplace matrix L to form the characteristic matrix H;

[0104] In this embodiment, the method for determining λ is as follows:

[0105] Cluster the Federer vectors of the Laplace matrix L, observe the changing relationship between the number of clusters k and the sum of squares of the clustering errors, and determine the value range of the number of clusters k by the elbow rule;

[0106] λ is set to three sets of values, namely k, k-1 and k-2. A smaller λ value is selected while ensuring that the selected features can distinguish the differences between clusters.

[0107] Cluster the feature matrix H to obtain the cluster labels of the corresponding word frequency sequences;

[0108] The frequency sequence of the science and technology topic keywords after clustering labels is input into the topic evolution stage prediction model to be trained and trained until convergence or the number of iterations is reached to obtain a trained topic evolution stage prediction model.

[0109] During the training process of the model, the embodiment of the present invention sets four hyperparameters: hidden layer, learning rate, epoch, and stopstep. The specific numerical settings are shown in Table 1.

[0110] Table 1 Model training hyperparameter settings

[0111] LearningRate Epoch EarlyStop HiddenDim 1e-2 10 5 10 1e-3 50 10 30 1e-4 100 20 50 — 200 — —

[0112] like Figure 4 As shown, the scientific and technological theme evolution stage prediction system based on multiple graph representation provided by the embodiment of the present invention includes:

[0113] A science and technology theme keyword acquisition module 100 is used to acquire science and technology theme keywords in scientific and technological documents;

[0114] Specifically, the science and technology theme keyword acquisition module 100 includes:

[0115] The subject corpus building unit is used to combine the title and abstract corpus of scientific and technological documents into the subject corpus of scientific and technological documents;

[0116] The scientific and technological subject keyword query unit is used to query and match preset keywords in the subject corpus, and use the keywords obtained by the query and matching as the scientific and technological subject keywords in the scientific and technological literature.

[0117] The word frequency sequence acquisition module 200 is used to obtain the word frequency sequence of the science and technology theme keywords; the word frequency sequence of the science and technology theme keywords is used as the theme strength to directly measure the degree of attention paid to the science and technology theme, which is a key quantitative indicator reflecting the evolution stage of the science and technology theme.

[0118] The semantic distance calculation module 300 is used to calculate the distance between the two components by the formula Calculate the keywords of science and technology topics and technology-themed keywords The semantic distance between ij ;in, is the technology theme keyword in k years The word vector of is the technology theme keyword in k years The word vector of

[0119] Semantic feature adjacency matrix construction module 400, used to construct the adjacency matrix of the semantic features S of the keywords in k years Among them, n is the number of all science and technology theme keywords in year k; the semantic distance relationship between science and technology theme keywords is used as the content feature, which reflects the correlation relationship between text carriers in the changes of theme evolution stage.

[0120] The citation frequency acquisition module 500 is used to obtain the citation relationship between each scientific and technological theme keyword, and obtain the citation frequency between each scientific and technological theme keyword. Among them, count_num is the technology theme keyword in k years right The frequency of citations;

[0121] Specifically, the citation frequency acquisition module 500 includes:

[0122] The citation relationship acquisition unit is used to obtain the citation relationship between each scientific and technological document according to the detailed information of each scientific and technological document, and obtain the citation relationship between each scientific and technological subject keyword through relationship mapping according to the citation relationship between each scientific and technological document. Taking the keyword citation frequency between scientific and technological subject keywords based on the document citation as a structural feature can reflect the interactive relationship between scientific knowledge.

[0123] The citation frequency acquisition unit is used to obtain the citation frequency between each pair of scientific and technological theme keywords according to the citation relationship between each scientific and technological theme keyword. Among them, count_num is the technology theme keyword in k years right The frequency of citations.

[0124] The reference relationship adjacency matrix construction module 600 is used to construct the reference relationship adjacency matrix of each scientific and technological theme keyword.

[0125] The multigraph construction module 700 is used to construct a multigraph G=(V={V w ,X},E={E s ,E c}); where V wrepresents the set of science and technology theme keyword nodes, X represents the word frequency sequence set of science and technology theme keyword nodes, In represents the word frequency sequence T corresponding to each technology theme keyword node, w represents the technology theme keyword node belonging to the multi-graph node set, E s Represents the adjacency matrix The semantic distance edge set between the keywords of each scientific and technological topic in E c Represents the adjacency matrix The set of citation relationship edges between the keywords of various scientific and technological topics;

[0126] The technology theme evolution stage prediction module 800 is used to input the multiple graphs G into the trained theme evolution stage prediction model and use GCN to predict the adjacency matrix Perform graph convolution operations and obtain keywords for each technology topic Features after modeling The specific formula is Use GAT based on the self-attention mechanism to adjust the adjacency matrix Modeling to obtain keywords for each technology topic Features after modeling The specific formula is The frequency sequence and characteristics of scientific and technological keywords feature Perform splicing and fusion to obtain the fused features The specific formula is in, Indicates technology-themed keywords The word frequency sequence of Input to the MLP layer to predict the evolution stage of scientific and technological topics and obtain the final prediction results Right now

[0127] Specifically, the cross entropy loss is used to calculate the loss of the prediction of the evolution stage of scientific and technological topics. The cross entropy loss is

[0128] The following is a detailed description of the training process of the prediction model in the topic evolution stage, including:

[0129] The adjacency matrix A building module is used to construct an undirected weighted graph by treating the word frequency sequence of each scientific and technological theme keyword as a data point and performing shape distance calculation to obtain the adjacency matrix A;

[0130] Specifically, the adjacency matrix A construction module is specifically used to construct the adjacency matrix A by taking the word frequency sequence data of each scientific and technological theme keyword as the vertex and the dynamic time warping distance between each word frequency sequence as the edge weight.

[0131] The similarity matrix W acquisition module is used to normalize the adjacency matrix A to obtain the similarity matrix W between graph vertices;

[0132] The weighted degree matrix D acquisition module is used to add the elements of each column of the similarity matrix W and place them on the diagonal position to form a diagonal matrix to obtain the weighted degree matrix D;

[0133] An eigenvalue decomposition module is used to obtain a Laplace matrix L according to a similarity matrix W and a weighted degree matrix D, and perform eigenvalue decomposition;

[0134] The characteristic matrix H acquisition module is used to obtain the characteristic vectors corresponding to the first λ smallest eigenvalues ​​of the Laplace matrix L to form the characteristic matrix H;

[0135] In this embodiment, the method for determining λ is as follows:

[0136] Cluster the Federer vectors of the Laplace matrix L, observe the changing relationship between the number of clusters k and the sum of squares of the clustering errors, and determine the value range of the number of clusters k by the elbow rule;

[0137] λ is set to three sets of values, namely k, k-1 and k-2. A smaller λ value is selected while ensuring that the selected features can distinguish the differences between clusters.

[0138] The word frequency sequence labeling module is used to cluster the feature matrix H and obtain the clustering labels of the corresponding word frequency sequences;

[0139] The model training module is used to input the frequency sequence of the science and technology theme keywords after clustering labels into the theme evolution stage prediction model to be trained until convergence or the number of iterations is reached, so as to obtain a trained theme evolution stage prediction model.

[0140] To sum up, in order to solve the technical problems of coarse analysis granularity and failure to fully utilize multi-dimensional evolutionary correlation information in the process of scientific and technological theme evolution in the prior art, an embodiment of the present invention proposes a new method for predicting the evolution stage of scientific and technological themes, which conducts a comprehensive analysis from multiple dimensions such as theme strength, theme content and theme structure to achieve a comprehensive and multi-level capture of theme evolution, and explores the multi-feature interaction in the process of scientific and technological theme evolution in subject areas, which will help to accurately track the theme evolution process, detect the stage of theme evolution, and predict future evolution stages.

[0141] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0143] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0144] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0145] The parts not described in detail in the embodiments of the present invention are all known technologies to those skilled in the art. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the evolutionary stage of scientific and technological topics based on multiple graph representations, characterized in that: include: Obtain scientific and technological subject keywords in scientific and technological literature; Obtaining a word frequency sequence of the scientific and technological topic keywords; By formula Calculate the keywords of science and technology topics and technology-themed keywords The semantic distance between ij ;in, is the technology theme keyword in k years The word vector of is the technology theme keyword in k years The word vector of Construct the adjacency matrix of the semantic features S of keywords within k years Where n is the number of all science and technology topic keywords in year k; Obtain the reference relationship between the scientific and technological theme keywords, and obtain the reference frequency between the scientific and technological theme keywords Among them, count_num is the technology theme keyword in k years right The frequency of citations; Construct the citation relationship adjacency matrix of each scientific and technological theme keyword Construct a multigraph G = (V = {V w ,X},E={E s ,E c }); where V w represents the set of science and technology theme keyword nodes, X represents the word frequency sequence set of science and technology theme keyword nodes, In represents the word frequency sequence T corresponding to each technology theme keyword node, w represents the technology theme keyword node belonging to the multi-graph node set, E s Represents the adjacency matrix The semantic distance edge set between the keywords of each scientific and technological topic in E c Represents the adjacency matrix The set of citation relationship edges between the keywords of various scientific and technological topics; The multi-graph G is input into the trained topic evolution stage prediction model, and the adjacency matrix is ​​quantized using GCN. Perform graph convolution operations and obtain keywords for various scientific topics Features after modeling Use GAT based on the self-attention mechanism to Modeling to obtain keywords for each technology topic Features after modeling The frequency sequence of the scientific and technological keywords is combined with the feature The features Perform splicing and fusion to obtain the fused features The characteristics Input to the MLP layer to predict the evolution stage of scientific and technological topics and obtain the final prediction results 2. The method for predicting the evolutionary stage of scientific and technological topics based on multiple graph representations according to claim 1, characterized in that: The obtaining of scientific and technological subject keywords in scientific and technological literature includes: Combining the title and abstract corpus of the scientific and technological documents as the subject corpus of the scientific and technological documents; The preset keywords are searched and matched in the subject corpus, and the keywords obtained by the search and matching are used as the scientific and technological subject keywords in the scientific and technological literature.

3. The method for predicting the evolutionary stage of scientific and technological topics based on multiple graph representations according to claim 1, characterized in that: The obtaining of the reference relationship between the scientific and technological subject keywords includes: The citation relationship between the scientific and technological documents is obtained based on the detailed information of the scientific and technological documents, and the citation relationship between the keywords of the scientific and technological subjects is obtained through relationship mapping based on the citation relationship between the scientific and technological documents.

4. The method for predicting the evolutionary stage of scientific and technological topics based on multiple graph representations according to claim 1, characterized in that: After obtaining the word frequency sequence of the scientific and technological subject keywords, the following steps are also included: The word frequency sequences of the keywords of each scientific and technological topic are regarded as data points to perform shape distance calculation to construct an undirected weighted graph, thereby obtaining an adjacency matrix A; Normalize the adjacency matrix A to obtain a similarity matrix W between graph vertices; Add the elements of each column of the similarity matrix W and place them on the diagonal position to form a diagonal matrix to obtain a weighted degree matrix D; Obtaining a Laplace matrix L according to the similarity matrix W and the weighted degree matrix D, and performing eigenvalue decomposition; Taking the eigenvectors corresponding to the first λ smallest eigenvalues ​​of the Laplace matrix L to form a characteristic matrix H; Clustering the feature matrix H to obtain cluster labels of corresponding word frequency sequences; The frequency sequence of the science and technology topic keywords after clustering labels is input into the topic evolution stage prediction model to be trained and trained until convergence or the number of iterations is reached to obtain the trained topic evolution stage prediction model.

5. The method for predicting the evolutionary stage of scientific and technological topics based on multiple graph representations according to any one of claims 1 to 4, characterized in that: The cross entropy loss is used to calculate the loss of the prediction of the evolution stage of scientific and technological topics. The cross entropy loss is 6. A scientific and technological theme evolution stage prediction system based on multiple graph representation, characterized in that: include: A science and technology subject keyword acquisition module is used to acquire science and technology subject keywords in scientific and technological literature; A word frequency sequence acquisition module, used to acquire the word frequency sequence of the scientific and technological subject keywords; The semantic distance calculation module is used to calculate the distance between Calculate the keywords of science and technology topics and technology-themed keywords The semantic distance between ij ;in, is the technology theme keyword in k years The word vector of is the technology theme keyword in k years The word vector of Semantic feature adjacency matrix construction module, used to construct the adjacency matrix of the semantic features S of keywords within k years Where n is the number of all science and technology topic keywords in year k; The citation frequency acquisition module is used to obtain the citation relationship between the scientific and technological theme keywords and obtain the citation frequency between each of the scientific and technological theme keywords. Among them, count_num is the technology theme keyword in k years right The frequency of citations; Citation relationship adjacency matrix construction module, used to construct the citation relationship adjacency matrix of each scientific and technological theme keyword A multigraph construction module is used to construct a multigraph G = (V = {V w ,X},E={E s ,E c }); where V w represents the set of science and technology theme keyword nodes, X represents the word frequency sequence set of science and technology theme keyword nodes, In represents the word frequency sequence T corresponding to each technology theme keyword node, w represents the technology theme keyword node belonging to the multi-graph node set, E s Represents the adjacency matrix The semantic distance edge set between the keywords of each scientific and technological topic in E c Represents the adjacency matrix The set of citation relationship edges between the keywords of various scientific and technological topics; The technology theme evolution stage prediction module is used to input the multi-graph G into the trained theme evolution stage prediction model, and use GCN to predict the adjacency matrix Perform graph convolution operations and obtain keywords for various scientific topics Features after modeling Use GAT based on the self-attention mechanism to Modeling to obtain keywords for each technology topic Features after modeling The frequency sequence of the scientific and technological keywords is combined with the feature The features Perform splicing and fusion to obtain the fused features The characteristics Input to the MLP layer to predict the evolution stage of scientific and technological topics and obtain the final prediction results 7. The system for predicting the evolutionary stages of scientific and technological topics based on multiple graph representations as claimed in claim 6, characterized in that: The science and technology theme keyword acquisition module includes: A subject corpus building unit, used for combining the title and abstract corpus of the scientific and technological documents as the subject corpus of the scientific and technological documents; The scientific and technological subject keyword query unit is used to query and match preset keywords in the subject corpus, and use the keywords obtained by the query and matching as the scientific and technological subject keywords in the scientific and technological literature.

8. The system for predicting the evolutionary stages of scientific and technological topics based on multiple graph representations as claimed in claim 6, characterized in that: The citation frequency acquisition module includes: A citation relationship acquisition unit is used to obtain the citation relationship between each scientific and technological document according to the detailed information of each scientific and technological document, and obtain the citation relationship between each scientific and technological subject keyword through relationship mapping according to the citation relationship between each scientific and technological document; The citation frequency acquisition unit is used to obtain the citation frequency between each pair of scientific and technological theme keywords according to the citation relationship between the scientific and technological theme keywords. Among them, count_num is the technology theme keyword in k years right frequency of citations.

9. The system for predicting the evolutionary stages of scientific and technological topics based on multiple graph representations as claimed in claim 6, characterized in that: Also includes: An adjacency matrix A construction module is used to construct an undirected weighted graph by treating the word frequency sequence of each scientific and technological theme keyword as a data point and performing shape distance calculation to obtain an adjacency matrix A; A similarity matrix W obtaining module is used to normalize the adjacency matrix A to obtain a similarity matrix W between graph vertices; A weighted degree matrix D obtaining module is used to add the elements of each column of the similarity matrix W and place them on the diagonal position to form a diagonal matrix to obtain the weighted degree matrix D; An eigenvalue decomposition module, used to obtain a Laplace matrix L according to the similarity matrix W and the weighted degree matrix D, and perform eigenvalue decomposition; A feature matrix H acquisition module is used to obtain the feature vectors corresponding to the first λ smallest eigenvalues ​​of the Laplace matrix L to form the feature matrix H; A word frequency sequence labeling module is used to cluster the feature matrix H to obtain cluster labels of corresponding word frequency sequences; The model training module is used to input the frequency sequence of the science and technology theme keywords after clustering labels into the theme evolution stage prediction model to be trained until convergence or the number of iterations is reached, so as to obtain the trained theme evolution stage prediction model.

10. The system for predicting the evolutionary stages of scientific and technological topics based on multiple graph representations according to any one of claims 6 to 9, characterized in that: The cross entropy loss is used to calculate the loss of the prediction of the evolution stage of scientific and technological topics. The cross entropy loss is

Images