A technology list generation method and system based on multi-source data and topic model

Through the multi-source data and theme model technology inventory generation method, using support vector machine and theme modeling algorithm, the problems of strong subjectivity, low accuracy and insufficient coverage in the existing technology inventory generation method are solved, and fast and accurate are achieved The technology list is generated.

CN114780617BActive Publication Date: 2024-11-22CHINESE ACAD OF ENG STRATEGIC CONSULTING CENT +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210483086.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-05-05
Publication Date
2024-11-22
Estimated Expiration
2042-05-05

AI Technical Summary

Technical Problem

The existing technology list generation methods have problems such as strong subjectivity, high cost, difficulty in guaranteeing accuracy, and low coverage of technology lists, especially based on expert analysis and literature measurement methods.

Method used

Using a technology inventory generation method based on multi-source data and theme models, by obtaining scientific and technological literature, using support vector machine models to divide technical categories, combining the theme modeling algorithms LDA and Louvain algorithms, first-level and second-level technology inventory are generated to improve the technology List accuracy and coverage.

Benefits of technology

It realizes the rapid and effective generation of technical lists, reduces the identification cost, improves the objectivity and accuracy of technical lists, enhances the technical coverage, and avoids the problems of subjective bias and insufficient in-depth mining.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114780617B_ABST
    Figure CN114780617B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for generating a technology list based on multi-source data and a topic model, comprising: obtaining a plurality of scientific and technological documents corresponding to a certain technical field, performing data processing on the scientific and technological documents, and storing the processed plurality of scientific and technological documents in a scientific and technological document database, inputting all scientific and technological documents in the scientific and technological document database into a trained support vector machine model SVM for technology category division, and storing the technology category division result in the scientific and technological document database, merging the scientific and technological documents belonging to the same technical category in the scientific and technological document database into one document, forming a document library with all documents, successively performing stop word removal and low-frequency word removal processing on the document library to obtain an updated document library, and using a topic modeling algorithm LDA to perform topic clustering on the updated document library to obtain a word distribution matrix corresponding to all topics. The present invention can solve the technical problems of strong subjectivity and high cost in the method based on expert judgment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of data mining, and more specifically, relates to a method and system for generating a technology list based on multi-source data and a topic model. Background Art

[0002] Technology foresight is a process of systematically exploring the long-term future of science, technology, economy and society. Its purpose is to select common new technologies and strategic research areas that may produce the greatest economic, environmental and social benefits, and form a list of key technologies. Technology foresight is of great significance in promoting the promotion and application of technology, accelerating the industrialization of technology and technology transfer. Developed countries such as the United States, Japan and Germany have conducted extensive research and application on technology foresight.

[0003] Existing methods for generating technology lists include expert judgment methods and bibliometric methods. The expert judgment-based method is centered on domain experts, and is usually organized by authoritative organizations to gather experts in related fields through multiple rounds of discussions to form a final technology list by gathering the knowledge and wisdom of domain experts. These methods include peer review, Delphi method, scenario analysis, etc. The bibliometric method usually takes scientific and technological literature such as papers and patents as research objects, and explores the hidden patterns of technology by analyzing the objective information in scientific and technological literature, in order to infer the future development trend of technology and form a technology list. These methods mainly include citation analysis and keyword analysis.

[0004] However, both of the above two methods for generating technology lists have some technical defects that cannot be ignored: First, the effectiveness of the method based on expert judgment depends entirely on the breadth and depth of the experts' knowledge in the relevant field, and it has strong subjective opinions. Although this method has high accuracy for specific technical fields, the identification cost is relatively high. Second, the method based on bibliometrics is better at identifying the overall development of a certain technical field, and because it is driven by objective literature data, it helps to solve the problem of subjective bias to a certain extent. However, this method is often based on simple statistical analysis, and the depth of mining is not enough. It is difficult to form a complete and specific technical description, and the accuracy of the generated technology list is difficult to guarantee. Third, the method based on bibliometrics is mostly based on one-dimensional data, and has the disadvantage of low coverage of the technology list. Summary of the invention

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a method and system for generating a technology list based on multi-source data and topic models, with the aim of quickly and effectively completing the generation of a technology list in a field through primary technology identification and secondary technology identification, thereby solving the technical problems of strong subjectivity and high cost in methods based on expert judgment, the technical problem of difficulty in ensuring accuracy in methods based on bibliometrics, and the technical problem of low coverage of technology lists based on single-dimensional literature data.

[0006] To achieve the above object, according to one aspect of the present invention, a method for generating a technology list based on multi-source data and a topic model is provided, comprising the following steps:

[0007] (1) Acquire multiple scientific and technological documents corresponding to a certain technical field, perform data processing on the scientific and technological documents, and store the processed multiple scientific and technological documents in a scientific and technological document database.

[0008] (2) All scientific and technological documents in the scientific and technological document database obtained in step (1) are input into the trained support vector machine model SVM for technical category classification, and the technical category classification results are stored in the scientific and technological document database.

[0009] (3) merging the scientific and technological documents belonging to the same technical category in the scientific and technological document database processed by step (2) into one document, and forming a document library with all the documents;

[0010] (4) The document library obtained in step (3) is subjected to stop word removal and low-frequency word removal processing to obtain an updated document library.

[0011] (5) Use the topic modeling algorithm LDA to perform topic clustering on the document library updated in step (4) to obtain a word distribution matrix corresponding to all topics in the document library, and generate a first-level technology list based on the word distribution matrix.

[0012] (6) Construct a word co-occurrence network for each topic in the first-level technology list obtained in step (5), cluster the word co-occurrence network corresponding to the topic using the Louvain algorithm to obtain multiple clusters corresponding to the topic, and analyze the subject word content in all clusters corresponding to the topic to obtain the second-level technology list corresponding to the topic.

[0013] (7) Associating the first-level technology list obtained in step (5) with the second-level technology list obtained in step (6) to generate a complete technology list.

[0014] Preferably, step (1) comprises the following sub-steps:

[0015] (1-1) Formulate a search formula for papers and patents related to a certain technical field, and search them in the World of Science Citation Database and the Derwent Patent Database respectively to obtain the scientific and technological literature corresponding to the technical field;

[0016] (1-2) For each scientific document corresponding to the technical field obtained in step (1-1), title extraction and abstract extraction are performed on the technical document respectively, and the title and abstract of each obtained scientific document are successively converted to lowercase and lemmatized using the natural language processing toolkit NLTK, and the processed scientific documents are stored in a scientific document database;

[0017] Preferably, the SVM model is trained in the following way:

[0018] (2-1) For each scientific document corresponding to a certain technical field processed in step (1), obtain the title and abstract of the technical document, and merge the obtained title and abstract to obtain a description text of the scientific document;

[0019] (2-2) using the descriptive text corresponding to each scientific document in the scientific document database obtained in step (2-1) to input the Doc2vec model to obtain the text vector corresponding to the scientific document;

[0020] (2-3) establishing a technology category library, which includes multiple technology categories and multiple technical documents corresponding to each technology category;

[0021] (2-4) constructing a data set using the text vectors corresponding to the scientific and technological documents in the scientific and technological document database obtained in step (2-2) and the technical category library obtained in step (2-3), and dividing the data set into a training set and a test set in a ratio of 8:2;

[0022] (2-5) Inputting the training set obtained in step (2-4) into the SVM model to obtain an updated SVM model;

[0023] Preferably, the weight parameters of the Doc2vec model include text vectors and word vectors. During training, the text vectors are connected with the context word vectors to jointly predict the next word, and the model weight parameters are updated by minimizing the loss of word prediction.

[0024] During the training process of the Doc2vec model, the word vector dimension of the Doc2vec model is set to 200, the number of training rounds is 100, and the sampling window width is 8;

[0025] The final result of the Doc2vec model is the mapping relationship between scientific literature and text vectors.

[0026] Preferably, step (2-5) specifically comprises iteratively setting the hyperparameters of the SVM model, and iteratively verifying the iteratively trained SVM model using the test set in the data set obtained in step (2-4) until the classification accuracy reaches the optimal level, thereby obtaining a trained SVM model.

[0027] Preferably, the stop words in step (4) are removed by removing the stop words in the document library, and the specific details can be found in the stop word list provided in NLTK;

[0028] Low-frequency words are words that appear less than 10 times in a document.

[0029] Preferably, each row of the word distribution matrix represents a different topic, and each column represents a different word. The element in the i-th row and j-th column of the matrix represents the probability that the j-th word belongs to the i-th topic, where i∈[1, the total number of topics in the document library], j∈[1, the total number of words in the document library].

[0030] In step (5), the first-level technology list is generated according to the word distribution matrix. Specifically, all elements in the first row of the word distribution matrix are first obtained, elements greater than a preset threshold are selected from all elements, the words corresponding to these elements are obtained, the technical fields to which these words belong are determined according to these words, and a mapping relationship between the technical field and the subject corresponding to the row is established; then, the above process is repeated for the remaining rows in the word distribution matrix, thereby obtaining a mapping relationship between multiple technical fields and their corresponding subjects, and all mapping relationships constitute the first-level technology list.

[0031] Preferably, in step (6), the subject word contents in all clusters corresponding to the subject are analyzed to obtain the secondary technology list corresponding to the subject. Specifically, the process is as follows: first, the elements selected from the first row of the word distribution matrix in step (5) are obtained, the words corresponding to these elements are obtained, and a word co-occurrence network is constructed based on the obtained words. If two words appear in the same scientific document at the same time, the weights corresponding to the two words are increased by 1; then, the word co-occurrence network is input into the Louvain algorithm to obtain a plurality of divided clusters, and clusters whose number of words is greater than a preset threshold are selected to determine the technical field to which they belong, and a mapping relationship between the technical field and the corresponding cluster is established; finally, the above process is repeated for the remaining rows of the word distribution matrix obtained in step (5), so as to obtain a mapping relationship between a plurality of technical fields and the subject to which they belong, and all mapping relationships constitute the secondary technology list.

[0032] According to another aspect of the present invention, a technology list generation system based on multi-source data and a topic model is provided, comprising:

[0033] The first module is used to obtain a plurality of scientific and technological documents corresponding to a certain technical field, perform data processing on the scientific and technological documents, and store the processed scientific and technological documents in a scientific and technological document database.

[0034] The second module is used to input all the scientific and technological documents in the scientific and technological document database obtained by the first module into the trained support vector machine model SVM to divide the technology categories, and store the technology category division results in the scientific and technological document database.

[0035] The third module is used to merge the scientific and technological documents belonging to the same technical category in the scientific and technological document database processed by the second module into one document, and all the documents form a document library;

[0036] The fourth module is used to remove stop words and low-frequency words from the document library obtained by the third module to obtain an updated document library;

[0037] The fifth module is used to use the topic modeling algorithm LDA to perform topic clustering on the document library updated by the fourth module to obtain the word distribution matrix corresponding to all topics in the document library, and generate a first-level technology list based on the word distribution matrix.

[0038] The sixth module is used to construct a word co-occurrence network for each topic in the first-level technology list obtained in the fifth module, cluster the word co-occurrence network corresponding to the topic using the Louvain algorithm to obtain multiple clusters corresponding to the topic, and analyze the subject word content in all clusters corresponding to the topic to obtain the second-level technology list corresponding to the topic.

[0039] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0040] (1) Since the present invention adopts step (1), it uses objective scientific and technological literature as data drive, which ensures the objectivity of the generation of the technology list, and thus can solve the technical problems of strong subjectivity and high cost existing in the existing qualitative methods;

[0041] (2) Since the present invention adopts step (1), it combines paper and patent data, expands the data dimension, makes the technology identification result more comprehensive, and solves the technical problem of low coverage of the technology list.

[0042] (3) Since the present invention adopts step (5), the topic modeling algorithm used by the present invention takes into account the semantics of the document and can automatically mine potential topics in the document. Compared with simple statistical measurement methods, the method has deeper semantic information. At the same time, the method can solve the technical problem of insufficient mining depth in the bibliometric method to a certain extent.

[0043] (4) The present invention adopts steps (2) and (5) to embed expert prior knowledge into the topic modeling algorithm, thereby improving the accuracy of the topic modeling algorithm and solving the technical problem of insufficient accuracy of the technology list generation method based on bibliometrics. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of a method for generating a technology list based on multi-source data and a topic model of the present invention;

[0045] Figure 2 It is the architecture diagram of the Doc2vec model of the present invention;

[0046] Figure 3 is a schematic diagram of the word distribution matrix generated in step (5) of the method of the present invention;

[0047] Figure 4 It is a schematic diagram of multiple clusters generated in step (6) of the method of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0049] The basic idea of ​​the present invention is to combine multi-source data and use a topic modeling algorithm embedded with expert knowledge to reduce the cost of generating a technology list while improving the accuracy of generating a technology list.

[0050] like Figure 1 As shown, the present invention provides a method for generating a technology list based on multi-source data and a topic model, comprising the following steps:

[0051] (1) Acquire multiple scientific and technological documents corresponding to a certain technical field, perform data processing on the scientific and technological documents, and store the processed multiple scientific and technological documents in a scientific and technological document database.

[0052] The advantage of this step is that it combines paper and patent data, expands the data dimension, and makes the technology list generation results more comprehensive.

[0053] This step includes the following sub-steps:

[0054] (1-1) Formulate a search formula for papers and patents related to a certain technical field, and search the Web of Science (WOS) and Derwent Innovations Index (DII) respectively to obtain the scientific and technological literature corresponding to the technical field;

[0055] Specifically, the search formula is developed with the participation of experts in the technical field.

[0056] (1-2) For each scientific document corresponding to the technical field obtained in step (1-1), title extraction and abstract extraction are performed on the technical document respectively, and the title and abstract of each obtained scientific document are successively converted to lowercase and lemmatized using the Natural Language Toolkit (NLTK), and the processed scientific document is stored in a scientific document database;

[0057] Specifically, the word form restoration process in this step can convert a word into its prototype word.

[0058] (2) Inputting all the scientific and technological documents in the scientific and technological document database obtained in step (1) into the trained support vector machine model (SVM) for technical category classification, and storing the technical category classification results in the scientific and technological document database.

[0059] The advantage of this step is that it combines expert knowledge and classifies scientific literature into technical categories based on machine learning algorithms, embeds expert knowledge into the subsequent LDA topic model, and improves the accuracy of the LDA topic model.

[0060] The SVM model in the present invention is trained in the following way:

[0061] (2-1) For each scientific document corresponding to a certain technical field processed in step (1), obtain the title and abstract of the technical document, and merge the obtained title and abstract to obtain a description text of the scientific document;

[0062] (2-2) using the descriptive text corresponding to each scientific document in the scientific document database obtained in step (2-1) to input the Doc2vec model to obtain the text vector corresponding to the scientific document;

[0063] The Doc2vec model in this step is an unsupervised text vector representation model. The model architecture is as follows: Figure 2As shown in the figure, the model weight parameters include text vectors and word vectors. During training, it connects the text vector with the context word vector to jointly predict the next word, and updates the model weight parameters by minimizing the loss of word prediction. Specifically, when training each text, the text vector and the sampled context word vector form a training example. After multiple sampling trainings with the sliding window, the text vector's understanding of the main idea expressed in the text becomes more and more accurate. When the training is completed, each text will obtain a fixed-length vector representation that takes into account both semantics and word order.

[0064] During the training process of the Doc2vec model in this step, the word vector dimension of the Doc2vec model is set to 200, the number of training rounds is 100, and the sampling window width is 8;

[0065] The final result of the Doc2vec model is the mapping relationship between scientific literature and text vectors.

[0066] The advantage of this step is that the Doc2vec model can automatically obtain the text vectors of scientific and technological documents while fully considering the text semantics, avoiding various complex text processing and indicator definition processes.

[0067] (2-3) establishing a technology category library, which includes multiple technology categories and multiple technical documents corresponding to each technology category;

[0068] Specifically, the technology category library is established by experts in the technical field mentioned in step (1) based on their accumulated expert knowledge. When establishing the library, efforts should be made to ensure that the number of papers and patents corresponding to each technology category is basically consistent.

[0069] (2-4) constructing a data set using the text vectors corresponding to the scientific and technological documents in the scientific and technological document database obtained in step (2-2) and the technical category library obtained in step (2-3), and dividing the data set into a training set and a test set in a ratio of 8:2;

[0070] (2-5) Inputting the training set obtained in step (2-4) into the SVM model to obtain an updated SVM model;

[0071] The advantage of this step is that, using machine learning algorithms, only a small amount of expert annotation cost is required to complete the automatic classification of a large amount of scientific and technological literature data.

[0072] Specifically, the hyperparameters of the SVM model are iteratively set, and the iteratively trained SVM model is iteratively verified using the test set in the data set obtained in step (2-4) until the classification accuracy reaches the optimal level, thereby obtaining a trained SVM model.

[0073] (3) merging the scientific and technological documents belonging to the same technical category in the scientific and technological document database processed by step (2) into one document, and forming a document library with all the documents;

[0074] Specifically, in this step, the number of documents in the document library is equal to the number of technical categories obtained by division.

[0075] (4) performing stop word removal and low-frequency word removal processing on the document library obtained in step (3) to obtain an updated document library;

[0076] Specifically, stop word removal is to remove the stop words in the document library (for details, please refer to the stop word list provided in NLTK). Low-frequency words refer to words that appear less than 10 times in the document.

[0077] (5) Use a topic modeling algorithm (Latent Dirichlet Allocation, LDA for short) to perform topic clustering on the document library updated in step (4) to obtain a word distribution matrix corresponding to all topics in the document library, and generate a first-level technology list based on the word distribution matrix.

[0078] The advantage of this step is that the use of topic modeling algorithms can automatically mine potential topics in the document library and have deeper semantic information.

[0079] After the LDA algorithm is used in this step, the word distribution matrix is ​​obtained as follows: Figure 3 As shown in , each row of the word distribution matrix represents a different topic, and each column represents a different word. The element in the i-th row and j-th column of the matrix represents the probability that the j-th word belongs to the i-th topic, where i∈[1, the total number of topics in the document library], j∈[1, the total number of words in the document library].

[0080] In this step, the first-level technology list is generated according to the word distribution matrix. Specifically, all elements in the first row of the word distribution matrix are first obtained, elements greater than a preset threshold are selected from all elements, words corresponding to these elements are obtained, the technical fields to which these words belong are determined according to these words, and a mapping relationship between the technical field and the subject corresponding to the row is established; then, the above process is repeated for the remaining rows in the word distribution matrix to obtain a mapping relationship between multiple technical fields and their corresponding subjects, and all mapping relationships constitute the first-level technology list.

[0081] Specifically, the preset threshold in this step is the average value of all elements in the row.

[0082] (6) Construct a word co-occurrence network for each topic in the first-level technology list obtained in step (5), and use the Louvain algorithm to cluster the word co-occurrence network corresponding to the topic to obtain multiple clusters corresponding to the topic (such as Figure 4 As shown in the figure, the subject word contents in all clusters corresponding to the subject are analyzed to obtain the secondary technology list corresponding to the subject.

[0083] In this step, the word co-occurrence network reveals the connection between the subject words. The Louvain algorithm can quickly and effectively mine the association relationship in the network. Using the Louvain algorithm to cluster the word co-occurrence network is conducive to summarizing the secondary technologies under the topic.

[0084] In this step, the subject word contents in all clusters corresponding to the subject are analyzed to obtain the secondary technology list corresponding to the subject. Specifically, the process is as follows: first, the elements selected in the first row of the word distribution matrix in step (5) are obtained, and the words corresponding to these elements are obtained. A word co-occurrence network is constructed based on the obtained words. If two words appear in the same scientific document at the same time, the weights corresponding to the two words are increased by 1; then, the word co-occurrence network is input into the Louvain algorithm to obtain multiple divided clusters, and clusters with a word number greater than a preset threshold are selected to determine the technical field to which they belong, and a mapping relationship between the technical field and the corresponding cluster is established; finally, the above process is repeated for the remaining rows of the word distribution matrix obtained in step (5), so as to obtain a mapping relationship between multiple technical fields and the subject to which they belong, and all mapping relationships constitute the secondary technology list.

[0085] Specifically, the preset threshold in this step is 50.

[0086] (7) Associating the first-level technology list obtained in step (5) with the second-level technology list obtained in step (6) to generate a complete technology list.

[0087] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for generating a technology list based on multi-source data and topic model, characterized in that: The following steps are involved: (1) obtaining a plurality of scientific and technological documents corresponding to a certain technical field, performing data processing on the scientific and technological documents, and storing the processed scientific and technological documents in a scientific and technological document database; (2) inputting all the scientific and technological documents in the scientific and technological document database obtained in step (1) into the trained support vector machine model SVM to perform technology category classification, and storing the technology category classification results in the scientific and technological document database; (3) merging the scientific and technological documents belonging to the same technical category in the scientific and technological document database processed in step (2) into one document, and forming a document library with all the documents; (4) performing stop word removal and low-frequency word removal processing on the document library obtained in step (3) to obtain an updated document library; (5) Using the topic modeling algorithm LDA to perform topic clustering on the document library updated in step (4) to obtain a word distribution matrix corresponding to all topics in the document library, and generating a first-level technology list based on the word distribution matrix; (6) Construct a word co-occurrence network for each topic in the first-level technology list obtained in step (5), cluster the word co-occurrence network corresponding to the topic using the Louvain algorithm to obtain multiple clusters corresponding to the topic, and analyze the subject word content in all clusters corresponding to the topic to obtain the second-level technology list corresponding to the topic; (7) Associating the first-level technology list obtained in step (5) with the second-level technology list obtained in step (6) to generate a complete technology list.

2. The method for generating a technology list based on multi-source data and a topic model according to claim 1, characterized in that: Step (1) includes the following sub-steps: (1-1) Develop a search formula for papers and patents related to a certain technical field, and search them in the World of Science Citation Database and the Derwent International Patent Database respectively to obtain the scientific and technological literature corresponding to the technical field; (1-2) For each scientific document corresponding to the technical field obtained in step (1-1), the title and abstract of the technical document are extracted respectively, and the title and abstract of each obtained scientific document are converted to lowercase and lemmatized using the natural language processing toolkit NLTK, and the processed scientific documents are stored in the scientific document database.

3. The method for generating a technology list based on multi-source data and a topic model according to claim 1 or 2, characterized in that: The SVM model is trained in the following way: (2-1) For each scientific document corresponding to a certain technical field processed in step (1), obtain the title and abstract of the technical document, and merge the obtained title and abstract to obtain a description text of the scientific document; (2-2) Using the descriptive text corresponding to each scientific document in the scientific document database obtained in step (2-1) to input the Doc2vec model to obtain the text vector corresponding to the scientific document; (2-3) establishing a technology category library, which includes multiple technology categories and multiple technical documents corresponding to each technology category; (2-4) Using the text vectors corresponding to the scientific literature in the scientific literature database obtained in step (2-2) and the technical category library obtained in step (2-3) to construct a data set, the data set is divided into a training set and a test set in a ratio of 8:2; (2-5) Input the training set obtained in step (2-4) into the SVM model to obtain an updated SVM model.

4. The method for generating a technology list based on multi-source data and a topic model according to any one of claims 1 to 3, characterized in that: The weight parameters of the Doc2vec model include text vectors and word vectors. During training, it connects the text vectors with the context word vectors to jointly predict the next word and updates the model weight parameters by minimizing the loss of word prediction. During the training process of the Doc2vec model, the word vector dimension of the Doc2vec model is set to 200, the number of training rounds is 100, and the sampling window width is 8; The final result of the Doc2vec model is the mapping relationship between scientific literature and text vectors.

5. The method for generating a technology list based on multi-source data and a topic model according to any one of claims 1 to 3, characterized in that: Step (2-5) specifically involves iteratively setting the SVM model hyperparameters, and iteratively verifying the iteratively trained SVM model using the test set in the data set obtained in step (2-4) until the classification accuracy reaches the optimal level, thereby obtaining a trained SVM model.

6. The method for generating a technology list based on multi-source data and a topic model according to claim 1, characterized in that: The stop word removal in step (4) is to remove the stop words in the document library. For details, please refer to the stop word list provided in NLTK; Low-frequency words are words that appear less than 10 times in a document.

7. The method for generating a technology list based on multi-source data and a topic model according to claim 1, characterized in that: Each row of the word distribution matrix represents a different topic, and each column represents a different word. The element in the i-th row and j-th column of the matrix represents the probability that the j-th word belongs to the i-th topic, where i∈[1, the total number of topics in the document library], j∈[1, the total number of words in the document library]; In step (5), the first-level technology list is generated according to the word distribution matrix. Specifically, all elements in the first row of the word distribution matrix are obtained, elements greater than a preset threshold are selected from all elements, the words corresponding to these elements are obtained, the technical fields to which these words belong are determined according to these words, and a mapping relationship between the technical fields and the corresponding topics in the row is established; Then, the above process is repeated for the remaining rows in the word distribution matrix to obtain the mapping relationship between multiple technical fields and their corresponding topics. All mapping relationships constitute the first-level technology list.

8. The method for generating a technology list based on multi-source data and a topic model according to claim 1, characterized in that: In step (6), the subject word contents in all clusters corresponding to the topic are analyzed to obtain the secondary technology list corresponding to the topic. Specifically, the process is as follows: first, the elements selected from the first row of the word distribution matrix in step (5) are obtained, the words corresponding to these elements are obtained, and a word co-occurrence network is constructed based on the obtained words. If two words appear in the same scientific document at the same time, the weights corresponding to the two words are increased by 1; Then, the word co-occurrence network is input into the Louvain algorithm to obtain multiple divided clusters. The clusters with the number of words greater than the preset threshold are selected to determine the technical field to which they belong, and a mapping relationship between the technical field and the corresponding cluster is established. Finally, the above process is repeated for the remaining rows of the word distribution matrix obtained in step (5), thereby obtaining the mapping relationship between multiple technical fields and the topics to which they belong. All mapping relationships constitute the secondary technology list.

9. A technology list generation system based on multi-source data and topic model, characterized in that: include: The first module is used to obtain a plurality of scientific and technological documents corresponding to a certain technical field, perform data processing on the scientific and technological documents, and store the processed scientific and technological documents in a scientific and technological document database; The second module is used to input all the scientific and technological documents in the scientific and technological document database obtained in the first module into the trained support vector machine model SVM to perform technology category classification, and store the technology category classification results in the scientific and technological document database; The third module is used to merge the scientific and technological documents belonging to the same technical category in the scientific and technological document database processed by the second module into one document, and all the documents form a document library; The fourth module is used to remove stop words and low-frequency words from the document library obtained by the third module to obtain an updated document library; The fifth module is used to perform topic clustering on the document library updated by the fourth module using the topic modeling algorithm LDA to obtain a word distribution matrix corresponding to all topics in the document library, and generate a first-level technology list based on the word distribution matrix; The sixth module is used to construct a word co-occurrence network for each topic in the first-level technology list obtained in the fifth module, cluster the word co-occurrence network corresponding to the topic using the Louvain algorithm to obtain multiple clusters corresponding to the topic, and analyze the subject word content in all clusters corresponding to the topic to obtain the second-level technology list corresponding to the topic.