Subject theme identification method and system fusing network structure features

Abstract

The invention discloses a subject theme recognition method and system fusing network structure features, and the method comprises the steps: extracting themes in a plurality of text data in a class of subjects, and generating a theme data set; selecting a plurality of high-influence themes from the theme data set as an experiment set, and randomly selecting a plurality of themes as a contrast set; taking the topics as nodes, connecting the associated topics, and constructing a topic network; respectively calculating multi-dimensional features of the experiment set topic network and the contrast set topic network; obtaining a comprehensive score of each topic based on the multi-dimensional features to determine an influence sequence of the topics; dividing theme data in the experiment set and the contrast set into a training set and a test set for training a machine learning model respectively to learn features of high-influence themes, and comparing a result output by the machine learning model with a result obtained by ranking theme influences in the theme network to obtain performance of the machine learning model. The subject recognition accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the field of machine learning technology, and in particular to a subject recognition method and system integrating network structure features. Background Art

[0002] Identification of emerging discipline topics is an effective way to timely discover the development trends of scientific and technological innovation and monitor the development trends of disciplines. At present, the identification methods of emerging discipline topics have formed a relatively stable research paradigm, forming a system of emerging discipline topic identification methods with scientometric analysis and knowledge network analysis as the main approaches. Scientometric analysis methods mainly analyze the metadata of discipline topic carriers and integrate multi-dimensional indicator characteristics to calculate their emerging indexes. Knowledge network analysis methods mainly rely on citation networks with scientific and technological literature as nodes and co-word network clustering with keywords as nodes to discover emerging document clusters and emerging keyword communities.

[0003] However, scientific quantitative analysis methods need to deal with the problem of multi-dimensional indicator integration, that is, how to determine the weight of each indicator, but there is currently a lack of effective methods to characterize the interaction between multiple indicators. Knowledge network analysis methods often use clustering to generate network communities to determine disciplinary themes, and then identify emerging themes from the dimensions of novelty and influence. However, this method ignores the position and structure of disciplinary themes in the knowledge network, which often reflect their own characteristics. Summary of the invention

[0004] The present invention provides a subject topic identification method and system integrating network structure characteristics, which can integrate the position and structural characteristics of subject topics in a subject network, thereby improving the accuracy of subject identification.

[0005] According to one aspect of the present invention, a subject topic identification method integrating network structure features is provided, comprising:

[0006] Acquire a number of text data of a subject, extract themes from the text data, and generate a theme data set;

[0007] Selecting several highly influential topics from the topic data set as an experimental set, and randomly selecting several topics from the topic data set as a control set, wherein the start and end time of publication of the text data corresponding to the topics in the control set and the topics in the experimental set are the same;

[0008] Taking the topics as nodes, connecting related topics, and respectively constructing the topic networks of the experimental set and the control set within a preset time period;

[0009] Calculating multi-dimensional features of the experimental set subject network and the control set subject network respectively;

[0010] Obtaining a comprehensive score for each topic based on the multi-dimensional features to determine the influence ranking of topics in the topic network of the experimental set and the influence ranking of topics in the control set;

[0011] The topic data in the experimental set and the control set are divided into a training set and a test set, which are used to train machine learning models respectively to learn the characteristics of high-influence topics, and the results output by the machine learning model are compared with the results obtained by ranking the topic influence in the topic network to obtain the performance of the machine learning model.

[0012] Optionally, the selecting of several highly influential topics from the topic data set as an experimental set, randomly selecting several topics from the topic data set as a control set, and the publication start and end time of the text data corresponding to the topics in the control set and the topics in the experimental set are the same, including:

[0013] Selecting several highly influential topics from the topic dataset as experimental sets;

[0014] A random matching algorithm is used to generate topics with the same start and end time of the text data publication as the topics in the experimental set as a control set.

[0015] Optionally, the acquiring of a number of text data of a subject and extracting themes from the text data includes:

[0016] A plurality of text data of a subject are obtained, and the knowledge entities in the text data are subject-coded using an edit distance algorithm to obtain a plurality of coded subjects.

[0017] Optionally, taking the topics as nodes, connecting related topics, and constructing the topic networks of the experimental set and the control set within a preset time period, respectively, including:

[0018] Taking topics as nodes and the mean of the probabilities of any two topics appearing in the same text data as the length of the edge between the two topics, two related topics are connected to construct a topic network of the topic data in the experimental set and the control set within a preset time period.

[0019] Optionally, the multi-dimensional features include at least the number of topic-related papers published, the influence of authors focusing on the topic, the frequency of citations of the topic, the influence of journals carrying the topic, network centrality, network centrality change rate, network PageRank, network PageRank change rate and the degree of network structure mutation.

[0020] Optionally, the degree of mutation of the network structure is calculated by the WL subtree kernel algorithm to calculate the similarity between the subtrees in the topic network structure diagram, thereby measuring the degree of mutation of the topic. The calculation formula is as follows:

[0021] S i (tn,t)=k WL (G i,t-n ,G i,t )

[0022] Where t represents time, n represents the time span; S i (tn,t) represents the mutation degree of the topic network from time tn to time t; k WL (G i,t-n ,G i,t ) is the WL subtree kernel algorithm used to calculate the topic network G i,t-n and G i,t 's structural similarity.

[0023] Optionally, obtaining a comprehensive score for each topic based on the multi-dimensional features to determine the influence ranking of topics in the topic network of the experimental set and the influence ranking of topics in the control set includes:

[0024] At least TOPSIS method, grey correlation method, entropy weight method, and rank sum ratio method are used to process the first multi-dimensional feature and the second multi-dimensional feature, thereby obtaining a comprehensive score for each topic;

[0025] Based on the comprehensive score, the topic samples were ranked by influence, and the proportion of high-impact topics was obtained based on the Top-n ranking.

[0026] Optionally, dividing the subject data in the experimental set and the control set into a training set and a test set for respectively training machine learning models to learn the characteristics of high-impact topics, and comparing the results output by the machine learning model with the results obtained by ranking the subject influence in the subject network to obtain the performance of the machine learning model, includes:

[0027] The topic samples in the experimental set are marked as 1, and the topic samples in the control set are marked as 0, and a five-fold cross-validation strategy is used to train multiple machine learning models respectively to learn the characteristics of high-impact topics;

[0028] A variety of classification model performance evaluation indicators are used to evaluate the classification results of multiple machine learning models, and the evaluation results are compared with the results obtained by ranking the topic influence in the topic network to obtain the performance of each machine learning model, and the optimal machine learning model is selected as the subject topic recognition model.

[0029] Optionally, the machine learning model includes at least logistic regression, support vector machine, decision tree, K-nearest neighbor, random forest and gradient boosting, multi-layer perceptron and TabNet model.

[0030] According to another aspect of the present invention, a subject identification system integrating network structure features is provided, comprising:

[0031] A subject extraction module is used to obtain a number of text data of a subject, extract the subject in the text data, and generate a subject data set;

[0032] A topic selection module is used to select several highly influential topics from the topic data set as an experimental set, and randomly select several topics from the topic data set as a control set, wherein the start and end time of the publication of the text data corresponding to the topics in the control set and the topics in the experimental set are the same;

[0033] A network construction module, used to connect related topics with topics as nodes, and respectively construct a topic network of the experimental set and the control set within a preset time period;

[0034] A feature calculation module, used to calculate the multi-dimensional features of the experimental set subject network and the control set subject network respectively;

[0035] An influence calculation module, used to obtain a comprehensive score for each topic based on the multi-dimensional features, so as to determine the influence ranking of the topics in the topic network of the experimental set and the influence ranking of the topics in the control set;

[0036] A training module is used to divide the topic data in the experimental set and the control set into a training set and a test set, and to train a machine learning model to learn the characteristics of high-impact topics, and to compare the results output by the machine learning model with the results obtained by ranking the topic influence in the topic network to obtain the accuracy of the machine learning model.

[0037] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0038] at least one processor; and

[0039] a memory communicatively connected to the at least one processor; wherein,

[0040] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the subject identification method integrating network structure features described in any embodiment of the present invention.

[0041] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the subject topic identification method integrating network structure features described in any embodiment of the present invention when executed.

[0042] The technical solution of the embodiment of the present invention constructs a topic network according to the correlation between topics to ensure that the connection strength between topics in the network is based on the actual co-occurrence of documents; calculates the multi-dimensional characteristics of the topic network, and adopts multiple methods to obtain the comprehensive score of each topic based on the multi-dimensional characteristics to determine the influence ranking of the topic; and trains a machine learning model to learn the characteristics of high-impact topics, compares the results output by the machine learning model with the results obtained by the topic influence ranking in the topic network to obtain the accuracy of the machine learning model, and then selects the best machine learning model to identify subject topics. The present invention can integrate the position and structural characteristics of subject topics in the topic network, thereby improving the accuracy of subject identification.

[0043] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0045] Figure 1 is a flow chart of a subject topic identification method integrating network structure features provided according to the first embodiment of the present invention;

[0046] Figure 2 is a flow chart of a subject topic identification method integrating network structure features provided in accordance with the second embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of a subject topic identification method integrating network structure features provided in Embodiment 2 of the present invention;

[0048] Figure 4 A schematic diagram of a curve of the proportion of high-influence topics obtained by ranking topics by influence using four ranking methods in Embodiment 2 of the present invention;

[0049] Figure 5 A system architecture diagram of a subject topic identification system integrating network structure features provided in Embodiment 3 of the present invention;

[0050] Figure 6It is a structural diagram of an electronic device for implementing a subject topic identification method integrating network structure features according to a fourth embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0052] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0053] Embodiment 1

[0054] Figure 1 A flowchart of a subject topic identification method integrating network structure features is provided for the first embodiment of the present invention. Figure 1 As shown, the method includes:

[0055] S101. Obtain some text data of a subject, extract themes from the text data, and generate a theme data set.

[0056] A class of disciplines may include fields such as medicine, semiconductors, communications, and chemical materials; text data may include literature data or patent data. In this embodiment, several text data in a class of disciplines may be randomly selected as experimental samples, and several subject topics may be extracted from the text data. The subject name may be a currently popular subject topic or a general topic, such as tumor immunity, targeted therapy, immunotherapy, etc. The extracted several topics are used as the subject data set of the experiment.

[0057] S101. Select several highly influential topics from the topic data set as an experimental set, and randomly select several topics from the topic data set as a control set, wherein the publication start and end time of the text data corresponding to the topics in the control set and the topics in the experimental set are the same.

[0058] According to the current popular topics, several popular topics can be selected from the topic data set as high-impact topics to form an experimental set; and several topics can be randomly selected from the topic data set as a control set. It should be noted that the publication start and end time of the text data of all topics in the control set should be the same as the publication start and end time of the text data of the topics in the experimental set, ensuring that the experimental set and the control set are the same in time, so that the multi-dimensional features of the experimental set obtained subsequently are comparable with the multi-dimensional features of the control set, thereby verifying the accuracy and effectiveness of the topic identification algorithm of this scheme.

[0059] S102, taking topics as nodes, connecting related topics, and respectively constructing a topic network of the experimental set and the control set within a preset time period.

[0060] For the experimental set and the control set, the subject networks of the experimental set and the control set can be constructed respectively with the subject data. In this embodiment, the subjects in the experimental set can be used as nodes, and the subjects with association relationships can be connected in pairs based on the association relationships between the subjects, thereby forming the subject network of the experimental set. Similarly, the control set also adopts the same construction method as the subject network of the experimental set.

[0061] Since the degree of association between two topics with an association relationship in the experimental set or the control set is different, the length of the connecting line between the two topics can be determined according to the degree of association, so that the lengths of the connecting lines between topics in the obtained topic network are different.

[0062] It should be noted that the preset time period can be one year or two years or other time settings. In this embodiment, the subject network of the experimental set or the control set can be obtained according to the preset time period. For example, 10 years can be used as the preset time period to obtain the subject network of the subject structure corresponding to the text data from 2000 to 2010 in the experimental set and the control set, and obtain the subject network of the subject structure corresponding to the text data from 2010 to 2020 in the experimental set and the control set, so as to be used for the subsequent comparison of the multi-dimensional features of the subject network of the experimental set and the control set in the same time period.

[0063] S103, respectively calculating the multi-dimensional features of the experimental set topic network and the control set topic network.

[0064] The multi-dimensional features may include external attribute features and topic network features of the topic. The external features may include features related to the influence of the text data corresponding to the topic; the topic network features may include the position and structural features of the topic in the topic network; the combination of the external features of the topic and the multi-dimensional features composed of the topic network features can reflect the influence of the topic. In this embodiment, the influence of the topic can be quantified, for example, the mean and standard deviation of the multi-dimensional features can be calculated respectively, and the influence of the topic can be reflected by the mean and standard deviation.

[0065] S104. Obtain a comprehensive score for each topic based on the multi-dimensional features to determine the influence ranking of the topics in the topic network of the experimental set and the influence ranking of the topics in the control set.

[0066] In this embodiment, a preset method can be used to calculate the quantitative results of the multi-dimensional features of each subject in the experimental set and the control set respectively, so as to sort the subjects and obtain the influence sorting of the subjects in the experimental set and the control set. For example, a sorting method such as TOPSIS method, grey correlation method, entropy weight method, rank sum ratio method, etc. can be used.

[0067] S105. Divide the topic data in the experimental set and the control set into a training set and a test set, respectively for training machine learning models to learn the characteristics of high-influence topics, and compare the results output by the machine learning model with the results obtained by ranking the topic influence in the topic network to obtain the performance of the machine learning model.

[0068] The experimental set data and the control set data in step S101 can be used as data sets to train the machine learning model, thereby ensuring that the sorting result in step S104 is comparable with the result output by the machine learning model.

[0069] Among them, the method of dividing the subject data in the experimental set and the control set into a training set and a test set can be to mix the experimental set and the control set as a sample set, and make different annotations on the subject data in the experimental set and the subject data in the control set, randomly extract a certain proportion of the subject data from the sample set as a training set, and randomly extract another proportion of the subject data from the sample set as a test set.

[0070] By comparing the topic results output by the machine learning model with the topic ranking results obtained by ranking the topic influence in the topic network, the accuracy of the identification results of high-impact topics can be determined.

[0071] The technical solution of the embodiment of the present invention constructs a topic network according to the correlation between topics to ensure that the connection strength between topics in the network is based on the actual co-occurrence of documents; calculates the multi-dimensional characteristics of the topic network, and adopts multiple methods to obtain the comprehensive score of each topic based on the multi-dimensional characteristics to determine the influence ranking of the topic; and trains a machine learning model to learn the characteristics of high-impact topics, compares the results output by the machine learning model with the results obtained by the topic influence ranking in the topic network to obtain the accuracy of the machine learning model, and then selects the best machine learning model to identify subject topics. The present invention can integrate the position and structural characteristics of subject topics in the topic network, thereby improving the accuracy of subject identification.

[0072] Embodiment 2

[0073] Figure 2 A subject topic identification method integrating network structure features is provided in the second embodiment of the present invention. Figure 2 Included:

[0074] S201. Acquire a number of text data of a subject, and use an edit distance algorithm to subject-code knowledge entities in the text data to obtain a number of coded subjects.

[0075] In this embodiment, biomedical disciplines can be used as experimental scenarios, and the metadata of the entire library of PubMed (medical database) can be parsed. Mainly based on the high-impact documents of Nature and Science, as well as the Nobel Prize-winning achievements in the field of biomedicine, the text data of 165 highly influential journal articles are finally obtained. The domain knowledge descriptors in the MeSH (Medical Subject Headings) knowledge base can be used to calculate the similarity between knowledge entities in the text data through the edit distance algorithm, and then the 14,741,948 knowledge entities in the high-impact documents are subject-coded. The metadata of the entire library of PubMed documents is parsed to obtain high-impact topics. The purpose of this analysis is to identify topics that have not yet been included in MeSH, and to support subsequent thematic network analysis, and finally obtain effective thematic coding and network structure data. Among them, the edit distance algorithm is a method for measuring the similarity of two strings, defining the minimum number of single-character editing operations required to convert from one string to another, which includes inserting, deleting or replacing a character.

[0076] Mesh Medical Subject Headings (MeSH) is an authoritative subject heading list and the most commonly used standard medical subject heading list in the biomedical field.

[0077] S202, selecting several highly influential topics from the topic data set as an experimental set; using a random matching algorithm to generate topics with the same publication start and end time as the text data corresponding to the topics in the experimental set as a control set.

[0078] In this embodiment, 165 high-quality biomedical research results were screened, and 231 high-quality research topics were finally selected as the experimental set. In order to verify the identification method of emerging discipline topics, the start and end time of the text data corresponding to the topic was kept the same, and another control set was generated by a random matching method. The random matching process is to combine a control group with the same time range from different literature libraries according to the start time of the experimental set and the same relevance standard, thereby ensuring the comparability of the experimental group and the control group in terms of time and subject characteristics, and then verifying the accuracy and effectiveness of the identification method.

[0079] S203, taking topics as nodes and the mean of the probabilities of any two topics appearing in the same text data as the length of the edge connecting the two topics, two related topics are connected, thereby constructing a topic network of the topic data in the experimental set and the control set within a preset time period.

[0080] In this embodiment, the coded topics can be used as nodes, and the co-occurrence relationship between two related topics in the same document can be used as edges to construct a topic co-occurrence network. The specific connection line construction method includes normalizing the co-occurrence edge weights: first calculate the probability that the labeled topics T1 and T2 appear in the same text data, respectively recorded as P(T1|T2) and P(T2|T1). These two probabilities represent the ratio of the labeled topics T1 and T2 appearing in the document at the same time. Then, the weight of the edge is represented by calculating the mean of the two probabilities, which is reflected in the topic network as the length of the two topic edges. The structure of the subject topic network finally obtained can be as follows: Figure 3 The network structure shown in the “Subject Topic Network” in .

[0081] This method ensures that the connection strength between nodes (topics) in the network is based on the actual document co-occurrence. Since each topic network corresponds to a different time period, the dynamic topic network displayed in time series can analyze the changes in the topic network in different time periods, thereby quantifying the relationship between topics and their evolution trend in the network.

[0082] S204, respectively calculating the multi-dimensional features of the experimental set topic network and the control set topic network.

[0083] In this embodiment, the multi-dimensional features include at least the number of papers related to the topic, the influence of authors focusing on the topic, the frequency of citations of the topic, the influence of journals carrying the topic, network centrality, network centrality change rate, network PageRank, network PageRank change rate and the degree of network structure mutation.

[0084] Among them, the number of topic-related papers published, the influence of authors focusing on the topic, the frequency of citations of the topic, and the influence of journals carrying the topic are external attribute characteristics of the topic, which are used to reflect the influence of the literature on the topic; network centrality, network centrality change rate, network PageRank, network PageRank change rate and network structure mutation degree are topic network characteristics, which are used to reflect the position relationship of the topic in the topic network.

[0085] Specifically, the number of papers related to the topic reflects the research activity of the topic in academic literature and represents the extensiveness of the topic in the literature. Secondly, pay attention to the influence of the topic author. Based on the number of citations of the main author and the academic influence indicators (such as the h-index), evaluate the authority and academic status of researchers related to the topic in the field; by counting the citations of papers related to the topic in other literature, calculate the citation frequency of the topic to measure the influence and recognition of the topic in the academic community; in addition, the influence of the journal on which the topic relies is analyzed by analyzing the impact factor of the journal where the relevant papers are published, reflecting the authority and importance of the journal in the academic community. The centrality of the topic in the network is used to evaluate the importance and connectivity of each topic in the topic co-occurrence network. The higher the centrality, the more significant the core position of the topic in the network. Secondly, the change rate of the centrality of the topic network in different time periods is calculated to reflect the magnitude of the change in the centrality of the topic in a specific period, thereby revealing whether the relative influence of the topic has fluctuated significantly over time; then, the PageRank change rate of the topic network is calculated. The network PageRank is calculated by the PageRank algorithm to determine the influence of the topic, and its degree of change reflects the diffusion rate of the topic's influence. Finally, the degree of structural mutation of the topic network is calculated by comparing the degree of structural evolution of the egocentric network in different periods.

[0086] Among them, the degree of network structure mutation is calculated by the WL subtree kernel algorithm to calculate the similarity between each subtree in the topic network structure diagram, and then measure the degree of mutation of the topic. The calculation formula is as follows:

[0087] S i (tn,t)=k WL (G i,t-n ,G i,t )

[0088] Where t represents time, n represents the time span; S i(tn,t) represents the mutation degree of the topic network from time tn to time t; k WL (G i,t-n ,G i,t ) is the WL subtree kernel algorithm used to calculate the topic network G i,t-n and G i,t 's structural similarity.

[0089] Based on the obtained characteristic values ​​of the number of topic-related papers published, the influence of authors focusing on the topic, the frequency of citations of the topic, the influence of the journal carrying the topic, network centrality, network centrality change rate, network PageRank, network PageRank change rate and network structure mutation degree, the mean and standard deviation corresponding to each characteristic value of each topic in the experimental set and the control set can be calculated. And the mean and standard deviation of the characteristic values ​​of each topic in the experimental set and the control set can be statistically analyzed.

[0090] S205. Use at least TOPSIS method, grey correlation method, entropy weight method, and rank sum ratio method to process the first multi-dimensional features and the second multi-dimensional features, and then obtain a comprehensive score for each topic.

[0091] Specifically, the TOPSIS method calculates the distance between the multi-dimensional features of each topic and the ideal solution and the negative ideal solution, obtains the relative closeness of each topic, and sorts them according to the relative closeness. The grey correlation method calculates the correlation coefficient between each topic and the reference sequence, obtains the grey correlation, and then sorts them. The entropy weight method determines the weight by calculating the information entropy of each feature of each topic, and ranks the topics based on the weighted comprehensive score. The rank sum ratio method ranks and standardizes each feature of each topic, and finally calculates the total score of each topic.

[0092] S206. The topic samples are ranked by influence based on the comprehensive scores, and the proportion of high-impact topics is obtained based on the Top-n ranking.

[0093] Specifically, through the comprehensive scores of these four methods, the influence of the topic samples in all experimental sets and control sets was ranked, and the proportion of high-impact topics was obtained based on the Top-n ranking. Figure 4 The results of the proportion of high-impact topics obtained by ranking the topics by influence using four ranking methods show that the TOPSIS method and the grey correlation method are particularly good at identifying high-impact topics, especially at P@5 and P@10 (the accuracy of the top 5 and top 10 items searched), which can identify more than 90% of high-impact topics, while the entropy weight method and the rank sum ratio method are relatively weak in performance.

[0094] S207. Mark the topic samples in the experimental set as 1, mark the topic samples in the control set as 0, and use a five-fold cross-validation strategy to train multiple machine learning models separately to learn the characteristics of high-impact topics.

[0095] It should be noted that the experimental group and the control group constructed in step S202 are divided into training sets and test sets, and the subject samples in the control set are marked as negative samples, that is, marked as "0", and the subject samples in the experimental set are marked as positive samples, that is, marked as "1", which can be abstracted as a binary classification task of machine learning.

[0096] In addition, in order to improve the robustness of the model and prevent overfitting, this embodiment can use a 5-fold crossover strategy to divide the training set data into 5 equal parts, use 4 of them for model training each time, and use the remaining 1 as a test set. The same machine learning model is trained 5 times to ensure that each subset is used as a test set. Finally, the performance of the model is evaluated by averaging the results of 5 experiments.

[0097] Among them, the machine learning model can use 8 classification algorithm models, including logistic regression (Linear), support vector machine (SVM), decision tree (DecisionTree), K neighbor (KNeighbors), random forest (RandomForest) and gradient boosting (GradientBoosting), multi-layer perceptron and TabNet. These models are trained through training sets, and the choice of loss function varies according to the model. For example, logistic regression and support vector machine models use cross-entropy loss (Cross-Entropy Loss) to minimize classification errors, while decision trees and random forests use Gini index or information gain to split nodes. For the gradient boosting model, the loss function used is mean square error (MSE) or log likelihood loss to improve the prediction effect of the tree.

[0098] S208. Use a variety of classification model performance evaluation indicators to evaluate the classification results of multiple machine learning models, compare the evaluation results with the results obtained by ranking the topic influence in the topic network to obtain the performance of each machine learning model, and select the optimal machine learning model as the subject topic recognition model.

[0099] After the model training was completed, multiple classification model performance evaluation indicators were used to evaluate the performance of 8 binary classification models in the high-impact topic identification task. These evaluation indicators include accuracy, precision, recall, and standard F1 value (F1 Score), which comprehensively evaluate the performance of each model in different evaluation dimensions. As shown in Table 1 below, the classification results of 8 machine learning models under 4 indicators show that for the main data of the medical topic selected in this embodiment, the performance indicators under the random forest model are better than those of other recognition models. By comparing these indicators, the best performing model can be effectively identified, providing a more accurate tool for the identification of high-impact topics.

[0100] Table 1 Performance evaluation results of topic identification model

[0101] Classification Model Accuracy (Acc) Precision Recall F1 value Logistic Regression 65.17 / 65.11 67.20 / 66.11 49.00 / 49.51 55.95 / 56.18 Support Vector Machine 62.71 / 62.88 63.67 / 67.46 46.99 / 39.58 53.44 / 48.78 Decision Tree 64.79 / 67.92 62.53 / 67.92 58.42 / 59.74 60.34 / 62.90 K Neighborhood 62.17 / 60.64 60.98 / 59.68 51.63 / 46.94 55.73 / 52.15 Random Forest 68.16 / 68.48 66.73 / 67.67 62.77 / 60.67 64.47 / 63.81 GBoosting 62.92 / 66.97 60.79 / 65.61 57.58 / 61.16 58.82 / 63.14 MLP 64.41 / 65.48 66.70 / 66.60 48.42 / 51.26 55.66 / 57.48 TabNet 62.56 / 61.57 60.82 / 58.25 58.13 / 59.22 58.86 / 58.76

[0102] Embodiment 3

[0103] Figure 5 This is a schematic diagram of the structure of a subject topic identification system integrating network structure features provided in the third embodiment of the present invention. Figure 5 As shown, the device comprises:

[0104] The subject extraction module 501 is used to obtain a number of text data of a subject, extract the subject in the text data, and generate a subject data set;

[0105] A topic selection module 502 is used to select a number of highly influential topics from the topic data set as an experimental set, and randomly select a number of topics from the topic data set as a control set, wherein the text data corresponding to the topics in the control set and the topics in the experimental set have the same publication start and end time;

[0106] A network construction module 503 is used to connect related topics with topics as nodes, and respectively construct a topic network of the experimental set and the control set within a preset time period;

[0107] A feature calculation module 504 is used to calculate the multi-dimensional features of the experimental set topic network and the control set topic network respectively;

[0108] An influence calculation module 505, used to obtain a comprehensive score for each topic based on the multi-dimensional features to determine the influence ranking of the topics in the topic network of the experimental set and the influence ranking of the topics in the control set;

[0109] The training module 506 is used to divide the topic data in the experimental set and the control set into a training set and a test set, and to train the machine learning model to learn the characteristics of high-influence topics, and to compare the results output by the machine learning model with the results obtained by ranking the topic influence in the topic network to obtain the accuracy of the machine learning model.

[0110] The message data processing resource scheduling device provided in the embodiment of the present invention can execute the message data processing resource scheduling device method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0111] Embodiment 4

[0112] Figure 6 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0113] like Figure 6 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0114] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0115] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a subject topic identification method integrating network structure features.

[0116] In some embodiments, a method for identifying a subject matter by integrating network structure features may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method for identifying a subject matter by integrating network structure features described above may be executed. Alternatively, in other embodiments, the processor 11 may be configured to execute a method for identifying a subject matter by integrating network structure features in any other appropriate manner (e.g., by means of firmware).

[0117] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0118] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0119] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0120] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0121] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0122] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0123] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0124] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A subject topic identification method integrating network structure features, characterized in that: include: Acquire a number of text data of a subject, extract themes from the text data, and generate a theme data set; Selecting several highly influential topics from the topic data set as an experimental set, and randomly selecting several topics from the topic data set as a control set, wherein the start and end time of publication of the text data corresponding to the topics in the control set and the topics in the experimental set are the same; Taking the topics as nodes, connecting related topics, and respectively constructing the topic networks of the experimental set and the control set within a preset time period; Calculating multi-dimensional features of the experimental set subject network and the control set subject network respectively; Obtaining a comprehensive score for each topic based on the multi-dimensional features to determine the influence ranking of topics in the topic network of the experimental set and the influence ranking of topics in the control set; The topic data in the experimental set and the control set are divided into a training set and a test set, which are used to train machine learning models respectively to learn the characteristics of high-influence topics, and the results output by the machine learning model are compared with the results obtained by ranking the topic influence in the topic network to obtain the performance of the machine learning model.

2. The subject topic identification method integrating network structure features according to claim 1 is characterized in that: The method of selecting a number of highly influential topics from the topic data set as an experimental set and randomly selecting a number of topics from the topic data set as a control set, wherein the start and end time of publication of the text data corresponding to the topics in the control set and the topics in the experimental set are the same, includes: Selecting several highly influential topics from the topic dataset as experimental sets; A random matching algorithm is used to generate topics with the same start and end time of the text data publication as the topics in the experimental set as a control set.

3. The subject identification method integrating network structure features according to claim 1 is characterized in that: The step of obtaining a plurality of text data of a subject and extracting themes from the text data includes: A plurality of text data of a subject are obtained, and the knowledge entities in the text data are subject-coded using an edit distance algorithm to obtain a plurality of coded subjects.

4. The subject identification method integrating network structure features according to claim 1 is characterized in that: Taking the topics as nodes, the related topics are connected to construct the topic networks of the experimental set and the control set in the preset time period, respectively, including: Taking topics as nodes and the mean of the probabilities of any two topics appearing in the same text data as the length of the edge between the two topics, two related topics are connected to construct a topic network of the topic data in the experimental set and the control set within a preset time period.

5. The subject identification method integrating network structure features according to claim 1 is characterized in that: The multi-dimensional features include at least the number of papers related to the topic, the influence of authors focusing on the topic, the frequency of citations of the topic, the influence of journals carrying the topic, network centrality, network centrality change rate, network PageRank, network PageRank change rate and the degree of network structure mutation.

6. The subject identification method integrating network structure features according to claim 5 is characterized in that: The degree of network structure mutation is calculated by the WL subtree kernel algorithm to calculate the similarity between the subtrees in the topic network structure diagram, and then measure the degree of mutation of the topic. The calculation formula is as follows: S i (t-n,t)=k WL (G i,t-n ,G i,t ) Where t represents time, n represents the time span; S i (tn,t) represents the mutation degree of the topic network from time tn to time t; k WL (G i,t-n ,G i,t ) is the WL subtree kernel algorithm used to calculate the topic network G i,t-n and G i,t 's structural similarity.

7. The subject identification method integrating network structure features according to claim 1 is characterized in that: The comprehensive score of each topic is obtained based on the multi-dimensional features to determine the influence ranking of the topics in the topic network of the experimental set and the influence ranking of the topics in the control set, including: At least the TOPSIS method, the grey correlation method, the entropy weight method, and the rank sum ratio method are used to process the first multi-dimensional feature and the second multi-dimensional feature, thereby obtaining a comprehensive score for each topic; Based on the comprehensive score, the topic samples were ranked by influence, and the proportion of high-impact topics was obtained based on the Top-n ranking.

8. The subject identification method integrating network structure features according to claim 1 is characterized in that: The subject data in the experimental set and the control set are divided into a training set and a test set for respectively training machine learning models to learn the characteristics of high-impact topics, and the results output by the machine learning model are compared with the results obtained by ranking the subject influence in the subject network to obtain the performance of the machine learning model, including: The topic samples in the experimental set are marked as 1, and the topic samples in the control set are marked as 0, and a five-fold cross-validation strategy is used to train multiple machine learning models respectively to learn the characteristics of high-impact topics; A variety of classification model performance evaluation indicators are used to evaluate the classification results of multiple machine learning models, and the evaluation results are compared with the results obtained by ranking the topic influence in the topic network to obtain the performance of each machine learning model, and the optimal machine learning model is selected as the subject topic recognition model.

9. The subject identification method integrating network structure features according to claim 1 is characterized in that: The machine learning models include at least logistic regression, support vector machine, decision tree, K-nearest neighbor, random forest and gradient boosting, multi-layer perceptron and TabNet models.

10. A subject identification system integrating network structure features, characterized in that: include: A subject extraction module is used to obtain a number of text data of a subject, extract the subject in the text data, and generate a subject data set; A topic selection module is used to select several highly influential topics from the topic data set as an experimental set, and randomly select several topics from the topic data set as a control set, wherein the start and end time of the publication of the text data corresponding to the topics in the control set and the topics in the experimental set are the same; A network construction module, used to connect related topics with topics as nodes, and respectively construct a topic network of the experimental set and the control set within a preset time period; A feature calculation module, used to calculate the multi-dimensional features of the experimental set subject network and the control set subject network respectively; An influence calculation module, used to obtain a comprehensive score for each topic based on the multi-dimensional features, so as to determine the influence ranking of the topics in the topic network of the experimental set and the influence ranking of the topics in the control set; A training module is used to divide the topic data in the experimental set and the control set into a training set and a test set, and to train a machine learning model to learn the characteristics of high-impact topics, and to compare the results output by the machine learning model with the results obtained by ranking the topic influence in the topic network to obtain the accuracy of the machine learning model.

Images