A leading topic identification method, system and computer device

Abstract

The application provides a front theme identification method, system and computer equipment. The method comprises the following steps: selecting field-oriented literature as a data set, and preprocessing the data set; constructing a citation coupling matrix and performing standardization according to the literature co-citation relationship to generate a literature relationship standard matrix; performing text modeling by using an LDA model to generate a literature similarity matrix; linearly fusing the literature similarity matrix and the literature relationship standardization matrix to form a text similarity matrix; obtaining a theme cluster by using community discovery; and visually describing the theme cluster to obtain an identification result of a front theme of the field-oriented literature. The method has high efficiency in simultaneously analyzing the structure and semantic levels, high accuracy in keyword disambiguation and artificial interpretation, overcomes the disadvantages of previous single-dimensional text analysis, simultaneously solves the problems of text similarity and fusion, and can be widely applied to structural and semantic text analysis.

Description

Technical Field

[0001] This invention relates to the field of information retrieval technology, and in particular to a cutting-edge topic identification method, system, and computer device. Background Technology

[0002] The world today is on the eve of a technological innovation breakthrough and a new technological revolution. The interdisciplinary, complex, and diverse characteristics of technological development are becoming increasingly prominent, and the crossover and integration between disciplines and fields are becoming more and more common. The Scientific Research Front refers to the latest trends and conceptual status of a research field. Exploration of the research front began in the 1960s, but the surge in research on the scientific front only began after 2005. It encompasses a wide range of content, integrating disciplines such as science science, library science, information science, artificial intelligence, content visualization, and the internet. In recent years, academic attention to frontier research has been continuously increasing, and similar or related terms include emerging topic / field research, topic tracking, trend detection, scientific research hotspots, disruptive technology prediction, and breakthrough technology prediction.

[0003] Scientific research frontiers often originate from new scientific discoveries or advancements. These discoveries or advancements can quickly attract the attention of scientists in the field, leading to further publications and an expansion of the field. Therefore, the emergence of research frontiers is often accompanied by corresponding bibliometric characteristics, typically including: a large number of new keywords appearing at a rapid pace; changes in the relationships between keywords or themes due to the influence of other disciplines; changes in the meaning of keywords due to the emergence of new phenomena; abnormal changes in the number of articles published on a particular topic, such as sudden increases or decreases; the emergence of new journals representing new fields; and abnormal changes in scientific citation networks or journal networks, as well as significant changes in knowledge structures. Monitoring these external characteristics may help detect potential research frontiers.

[0004] For a long time, methods for exploring the scientific frontier have included the Delphi method and bibliometrics. With the development of intelligent technology, machine learning, knowledge graphs, and other technologies and methods have also been gradually applied to frontier identification. All of these methods have been widely used in research practice. Methodologically, they can be categorized into two types: one based on citation clustering and the other based on keyword bursts; the other focuses on the quantitative expression and screening of indicator meanings (e.g., novelty, innovativeness, forward-looking perspectives). With the increasing richness of data sources, facing multiple data sources and the need for data complementarity and correction, research on the fusion of heterogeneous data to explore scientific development has begun to emerge. This shifts from single papers to the fusion of open-source data such as papers, patents, funding projects, and reports. This fusion is not merely a fusion of bibliometric features, but a fusion that delves into the knowledge unit level.

[0005] In econometric methods, those targeting cutting-edge scientific research are primarily based on high co-occurrence or relational clustering of paper citations or keyword features. Through temporal evolution, these methods reveal the scientific frontiers in relevant fields. Such high co-occurrence or strong relationships typically indicate a high degree of interaction among actors, with close relationships in certain existing forms of interaction. Therefore, they are more easily identified and confirmed in the detection of scientific frontiers.

[0006] Meanwhile, due to the complexity of disciplinary knowledge, some phenomena have emerged in the exploration of cutting-edge research that deserve in-depth study. For example, the use of single methods: many data mining algorithms often focus on one aspect of the data model, discarding other dimensions. However, the important information hidden in other dimensions, especially those topics with low support and high confidence, deserves more attention. Secondly, in the process of clustering topics at the forefront of research, since co-word analysis uses high-frequency words, the clusters composed of relatively low-frequency topics cannot be reflected in the clustering, resulting in an inability to reflect the full picture of the discipline. Finally, the micro-content revealed by a single method or relationship is one-sided and limited. In the open-source intelligence data environment, domain-oriented judgment and early warning require more data information to be aggregated and integrated. Summary of the Invention

[0007] This invention discloses a cutting-edge topic identification method, system, and computer device to address the problems that traditional quantitative analysis cannot delve into semantics and that topic discovery cannot provide better interpretation.

[0008] In a first aspect, embodiments of the present invention propose a method for identifying cutting-edge topics, comprising: Step 1, selecting domain-specific literature as a dataset and preprocessing the dataset; Step 2, constructing a citation coupling matrix based on co-citation relationships and standardizing it to generate a standard document relationship matrix; Step 3, using an LDA model to perform text modeling, obtaining the topic probability distribution of documents, and calculating document similarity at the content level to generate a document similarity matrix; Step 4, linearly fusing the document similarity matrix and the standardized document relationship matrix to form a text similarity matrix; Step 5, using community discovery to obtain topic clusters, visually describing the topic clusters, and obtaining the identification results of cutting-edge topics in the domain-specific literature.

[0009] Secondly, this invention proposes a cutting-edge topic identification system, comprising: a preprocessing module for selecting domain-specific literature as a dataset and preprocessing the dataset; a literature relationship matrix generation module for constructing and standardizing a citation coupling matrix based on co-citation relationships to generate a standard literature relationship matrix; a literature similarity matrix generation module for using an LDA model to perform text modeling, obtaining the topic probability distribution of documents, and calculating the literature similarity at the content level to generate a literature similarity matrix; a data fusion module for linearly fusing the literature similarity matrix and the standardized literature relationship matrix to form a text similarity matrix; and a data analysis module for using community discovery to obtain topic clusters, visually describing the topic clusters, and obtaining the identification results of cutting-edge topics in the domain-specific literature.

[0010] Thirdly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the frontier topic identification method as described in the above embodiments.

[0011] Fourthly, embodiments of the present invention provide a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the frontier topic identification method as described in the above embodiments.

[0012] The cutting-edge topic identification method, system, and computer equipment disclosed in this invention are oriented towards the structural and semantic fusion of text. Based on the combined concepts of "entity-relationship" and "semantics-structure," and grounded in text modeling, similarity construction, semantic structure fusion, and automatic topic identification, it detects cutting-edge topics while simultaneously researching structural analysis methods and exploring engineering implementation issues in intelligence research. The cutting-edge topic identification method of this invention exhibits high efficiency in simultaneously analyzing both structural and semantic levels, and high accuracy in keyword disambiguation and manual interpretation. It overcomes the shortcomings of previous methods that relied on single-dimensional text analysis, and simultaneously solves the problems of text similarity and fusion. It can be widely applied to text analysis involving both structure and semantics.

[0013] Specific embodiments of the invention are disclosed in detail with reference to the following description and accompanying drawings, indicating how the principles of the invention can be employed. It should be understood that the embodiments of the invention are not therefore limited in scope. Within the spirit and scope of the appended claims, embodiments of the invention include many changes, modifications, and equivalents.

[0014] Features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, combined with features in other embodiments, or substituted for features in other embodiments.

[0015] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, whole, step, or component, but does not exclude the presence or addition of one or more other features, wholes, steps, or components. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the frontier topic identification method according to an embodiment of the present invention.

[0018] Figure 2 This is a partial screenshot of the Topic-doc matrix obtained based on LDA modeling in a specific embodiment of the present invention;

[0019] Figure 3 This is a partial screenshot of the document similarity matrix in a specific embodiment of the present invention;

[0020] Figure 4 This is a schematic diagram illustrating the effect of a topic visualization generated by the application of a community identification algorithm in a specific embodiment of the present invention;

[0021] Figure 5 This is a schematic diagram of the frontier topic recognition system according to an embodiment of the present invention;

[0022] Figure 6 A schematic diagram of the structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Those skilled in the art will recognize that embodiments of the present invention can be implemented as a system, apparatus, device, method, or computer program product. Therefore, this disclosure can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0025] The principles and spirit of the present invention will be explained in detail below with reference to several representative embodiments.

[0026] In this invention, it is important to understand that:

[0027] NPG: Nature Publication Group.

[0028] NI: Nature Index is a database that tracks the number of papers published by universities, research institutions (countries) in the world's most influential research journals.

[0029] LDA: Latent Dirichlet Allocation, is a set probability model primarily used to process discrete datasets. Currently, it is mainly used in text mining and natural language processing to reduce dimensionality.

[0030] JCR: Journal Citation Reports.

[0031] C-value: The basic idea of ​​the C-value method is to first obtain a set of candidate terms using language rules, and then use statistical information for filtering.

[0032] TF-IDF (Term Frequency-Inverse Document Frequency) is a statistical method used to assess the importance of a word to a document within a set of documents or a corpus. A word's importance increases proportionally to the number of times it appears in a document, but decreases inversely proportionally to its frequency in the corpus.

[0033] LSA: Latent Semantic Analysis.

[0034] LSI: Latent Semantic Index.

[0035] PLSA: Probabilistic Latent Semantic Analysis.

[0036] ESI: Essential Science Indicators.

[0037] Citation coupling network: This refers to the network formed by the citation relationships of references in papers within a specific field. The more common references two documents share, the greater their citation coupling, the higher their static association, and the stronger their relevance. Greater citation coupling strength indicates a closer connection in content and subject matter between the two documents.

[0038] Community identification: This involves analyzing complex networks formed by entities such as keywords, documents, and authors, based on relationships like co-occurrence and citations, using algorithms such as clustering and methods related to topic identification.

[0039] Clustering: Clustering is a typical algorithm for unsupervised learning that does not require labeling results. It attempts to explore and discover certain patterns to find common groups, dividing data into multiple categories based on intrinsic similarity, such that intra-intra-intra-similarity is high and inter-intra-similarity is low.

[0040] Text similarity: refers to the degree of similarity between texts when mining text.

[0041] Structural analysis: generally refers to the analysis of the external features of texts or documents.

[0042] Currently, at the semantic and structural level, there are two main approaches: one in the library and information science field focuses on describing the external characteristics of documents at the scientometric level, including analysis of co-occurrence and citation relationships such as keywords, authors, citations, and journals; the other in the computer science field focuses on semantic and content analysis. Both approaches suffer from a lack of dimension. In addition, traditional feature extraction and text similarity methods have the following drawbacks: the resulting topic clusters are composed of articles, requiring a set of identification methods to determine the content characteristics of the clustered documents; the preprocessing requirements for words are high, making it difficult to handle word meaning disambiguation; and the method relies solely on statistical word frequency or co-word frequency, with less consideration given to the semantic relationships among the topic, keywords, and articles.

[0043] Based on the above shortcomings, this invention combines the concepts of "entity-relationship" and "semantics-structure," using text modeling, similarity construction, semantic structure fusion, and automatic topic identification to explore cutting-edge topics. Simultaneously, it studies structural analysis methods, explores engineering implementation issues in intelligence research, and proposes current research topics and cross-disciplinary fields. The aim is to provide a more comprehensive, in-depth, and detailed interpretation of domain-specific knowledge structures, addressing the limitations of traditional quantitative analysis in reaching semantic depth and the inability to provide adequate interpretation at the topic discovery level. This will help researchers identify hot topics, future research directions, and cutting-edge fields.

[0044] Figure 1 This is a flowchart illustrating the frontier topic identification method according to an embodiment of the present invention. Figure 1 As shown, the method in this embodiment includes:

[0045] Step S101: Select domain-specific literature as the dataset and preprocess the dataset;

[0046] Step S102: Based on the co-citation relationship of the documents, construct the citation coupling matrix and standardize it to generate the standard document relationship matrix;

[0047] Step S103: Use the LDA model to perform text modeling, obtain the topic probability distribution of the document, calculate the document similarity from the content level, and generate a document similarity matrix.

[0048] Step S104: Linearly fuse the document similarity matrix and the document relationship standardization matrix to form a text similarity matrix;

[0049] Step S105: Using community discovery, topic clusters are obtained, and the topic clusters are visualized to obtain the identification results of the cutting-edge topics of the literature in the target field.

[0050] In specific implementation, step S101 involves preprocessing the dataset, specifically including vectorizing the text to generate a text-feature word matrix and a text-citation matrix. Text vectorization is a fundamental prerequisite for natural language processing, such as semantic analysis and data fusion.

[0051] In step S102, to avoid the time lag caused by co-citation, citation coupling is used in establishing the document relationship, and standardization is performed. Based on the coupling relationship between the documents, the standard co-citation degree of the two documents is calculated:

[0052]

[0053] Where, N x Let N be the number of references for document x. y N represents the number of references in document y. xy This represents the number of references shared by document x and document y.

[0054] At the level of literature content analysis, Whittaker was the first to propose the assumptions of co-word analysis. Traditional co-word analysis, using methods such as inclusion index and proximity index, strategic coordinates, and multidimensional scales, has been widely applied to reveal the development of research topics in a field, and has been used to analyze research hotspots and trends in a discipline. In recent years, researchers have used methods such as keyword weighting, word segmentation techniques based on supplementary dictionaries to add candidate keywords based on author keywords, and combining C-value and TF-IDF (Term Frequency-Inverse Document Frequency) to improve keyword extraction calculation models, and integrating linguistic rules and statistical information to compensate for the semantic degree of words in order to express text vectors. Knowledge models for mining latent semantics include LSA (Latent Semantic Analysis) / LSI (Latent Semantic Index), PLSA (Probabilitistic Latent Semantic Analysis), and LDA (Latent Dirichlet Allocation). Compared with LSA, PLSA defines a probabilistic model and its assumptions are more consistent with the characteristics of text. PLSA itself grows linearly with the number of documents and keywords, becoming increasingly large. Blei's LDA, based on a three-layer Bayesian model, is a fully probabilistic model and is more complete than PLSA. Its parameter space size is determined by the number of latent topics and words, and is independent of the document size. Therefore, LDA is more suitable for building models on large-scale corpora.

[0055] LDA (Latent Distributed Aspects) is a type of topic model in text modeling, and it is also a landmark topic model because it was the first to formally introduce topics as latent variables, forming a three-layer Bayesian model. Compared to the PLSA (Latent Partial Scale) text model, which was the closest to it before, LDA's topic selection is no longer constrained by the content of the training set text. It is a completely unsupervised algorithm in the fields of machine learning and data mining that clusters multiple topics.

[0056] Modeling a text set using the LDA model can yield document-topic and topic-word distribution matrices, allowing for the extraction of latent semantic knowledge. In this embodiment, keywords are extracted from titles and abstracts, and then deduplicated and cleaned by comparing them with the authors' own keywords. The modeling process employs the Gibbs random sampling algorithm, using perplexity to determine the optimal number of topics. The parameter α = 50 / K indicates the relative strength of latent topics within the text set, and the parameter β = 0.01 characterizes the probability distribution of latent topics on words. The optimal number of topics is selected based on the perplexity of the computational model.

[0057] The topic vector for article mapping based on the LDA model is d. i =(t1,t2,……,t K The similarity between the two articles is calculated using the cosine values ​​of two topic vectors:

[0058]

[0059] Where, d i and d j Sim(d) represents the topic vectors mapped from the two documents. i d j The similarity between two documents is measured by the cosine value of the topic vector based on the LDA model.

[0060] The obtained document similarity matrix is ​​shown in Table 1:

[0061] Table 1. Document Similarity Matrix

[0062] doci docj sim … … … … … … … … …

[0063] Where doci and docj represent unique identifiers of documents, and sim is the similarity between documents measured by the cosine value of topic vectors based on the LDA model.

[0064] Considering the motivation behind citations, not all cited references necessarily match the core ideas of the article. Based on this assumption, this invention, in addition to citation analysis, considers the implicit semantics of the article. From another perspective, the information contained in scientific literature is not merely a collection of words. To fully identify the subject matter within a document, citations, as a crucial component, should naturally participate in the composition of the document's subject matter. Incorporating citations into the subject matter not only enriches the theme but also comprehensively demonstrates the connections between topics. A large category of deep semantic algorithms is based on citation networks, but the calculation of subject similarity based on citations has certain problems. On the one hand, factors such as similar publication dates may result in no citation relationship between two documents; on the other hand, even if two documents do have a citation relationship, considering the assumption of citation motivation, this citation relationship might be meaningless.

[0065] In this embodiment of the invention, the standard document relationship matrix formed based on citation coupling in step S102 and the document similarity matrix formed based on LDA in step S103 are semantically and structurally fused. Based on the citation coupling network (normalized value >= 0.06), the network association relationships are adjusted according to semantic information, and its weight λ is also adjusted:

[0066]

[0067] Where λ is the set weight, a variable that can be adjusted at any time by those skilled in the art based on needs and work experience in actual work; Sim LDA The Sim value represents the document-to-document similarity value in the document similarity matrix; cite couple The standard co-citation degree in the document relationship standardization matrix is ​​represented; cite(i,j) refers to whether any two nodes i,j (nodes are mappings of documents) are adjacent in the basic citation network. If cite(i,j) = 0, they are not adjacent; if cite(i,j) > 0, they are adjacent.

[0068] a and b are two boundary values ​​of similarity in the document similarity matrix, and the values ​​of a and b can be determined by examining the constructed LDA modeling and document similarity matrix.

[0069] View the LDA modeling and literature similarity matrix based on the constructed model, sort them in order of similarity from high to low, select the top 10%, and the minimum value in the top 10% is a;

[0070] View the LDA modeling and literature similarity matrix based on the constructed model, sort them in descending order of similarity from low to high, select the top 10%, and the maximum value in the top 10% is b.

[0071] Based on the basic network, if there is no link between two nodes, and the semantic similarity between the two nodes is higher than the top 10% of the pairwise similarity of all nodes (denoted as 'a', sorted in descending order of similarity), then a link is added to the basic network; if there is a link between two nodes, and the semantic similarity between the two nodes is higher than the top 10% of the semantic similarity of adjacent nodes in the basic network (denoted as 'b', sorted in descending order of similarity), then a link is added to the basic network.

[0072] Traverse the dataset, find the Sim values ​​between documents matched in the constructed LDA model and document similarity matrix, and add them to the database table to form the form shown in Table 2 below, so as to adjust the network association relationship.

[0073] Table 2 shows the new database table formed by adjusting the relationship network.

[0074] ESI_ID1 ESI_ID2 normalized_cocited sim 5 7 0.08 0.95

[0075] In one specific embodiment, Table 3 is the dataset obtained through citation co-citation, i.e., the standard matrix of document relationships formed based on citation coupling, as shown below:

[0076] Table 3 shows the dataset (standard matrix of document relationships) obtained through co-citation.

[0077] ESI_ID1 ESI_ID2 normalized_cocited 1 7977 23008 0.119098266835083 2 8834 36261 0.0683408533502272 3 16079 46814 0.108253175473055 4 28990 44633 0.107211253483779 5 6962 12988 0.109108945117996 6 32334 33719 0.0816496580927726 7 28454 49202 0.0603984338889136 8 43396 46823 0.114285714285714 9 10672 35566 0.0753778361444409

[0078] Table 4 shows the document similarity matrix formed by LDA using cosine pairs, as follows:

[0079] Table 4. Document Similarity Matrix

[0080]

[0081]

[0082] The two matrices (shown in Tables 3 and 4) are semantically and structurally fused using the linear fusion formula (3). Based on the citation coupling network (normalized value >= 0.06), the network association is adjusted according to the semantic information, and its weight λ is also adjusted.

[0083] In this specific embodiment, λ is temporarily set to 0.5. Of course, λ can be set as a variable and adjusted at any time. Based on the basic network, if there is no link between two nodes, and the semantic similarity between the two points is higher than the top 10% of the pairwise similarity of all nodes (denoted as 'a', sorted in descending order of similarity), then a link is added to the basic network. If there is a link between two nodes, and the semantic similarity between the two points is higher than the top 10% of the semantic similarity of adjacent nodes in the basic network (denoted as 'b', sorted in descending order of similarity), then a link is added to the basic network.

[0084] After traversing the dataset, i.e., after linear fusion, the dataset is presented in the form shown in Table 5 below:

[0085] Table 5. Similarity of merged documents

[0086] doci docj sim 1 2 x 1 3 y … … …

[0087] Where x and y are both values ​​between 0 and 1.

[0088] In specific implementation, in step S105, community discovery is used to obtain topic clusters, the topic clusters are visualized and described, and structural analysis is performed to obtain the identification results of the cutting-edge topics of the literature in the field, so as to identify the current research hotspots and cutting-edge directions.

[0089] The difference between community partitioning and traditional clustering lies in the fact that community partitioning focuses on the relationships between nodes and links in a complex network, while traditional clustering emphasizes the similarity of text content itself. The data source is standardized using a citation network matrix; based on the LDA topic model, document similarity is calculated at the content level to obtain a document-document relationship matrix; based on a fusion network of citation relationships and semantic content, a comprehensive analysis is performed from both semantic and structural perspectives. The Louvain algorithm, a large-scale community discovery algorithm, is used for community partitioning. This algorithm is a modularity-based graph algorithm model. Unlike ordinary modularity-based algorithms and modularity gain algorithms, this algorithm is very fast and has a particularly significant effect on clustering graphs with many nodes and few edges, resulting in a significant speed improvement in graph rendering. The Louvain algorithm process consists of two stages: the first stage, called ModularityOptimization, mainly assigns each node to the community of its neighboring nodes to continuously increase the modularity value; the second stage, called CommunityAggregation, mainly aggregates the communities partitioned in the first step into a single point, i.e., reconstructs the network based on the community structure generated in the previous step. This process is repeated until the network structure no longer changes.

[0090] After identifying topic clusters using community discovery, topic visualization and description will be performed. The relationships between and within clusters are represented by a fused matrix obtained by combining two matrices from the previous steps. This fused matrix describes the relationships between nodes, revealing relationships between documents within a cluster and connections outside the cluster. The coordinates of the cluster's central node can be determined by randomly selecting a node and calculating the node with the highest degree (most links) within that cluster. Based on the overall layout, the position of each node is calculated. Links within a cluster can be obtained using the similarity matrix generated earlier.

[0091] Links between class groups require the following calculations: Assume two class groups A and B, with x nodes in group A and y nodes in group B. If all nodes in both groups A and B are linked, the maximum number of links is x*y. Let the actual number of links be z, then z / (x*y) is the normalized link between class groups. Taking a class group as the center (let's say group M), calculate the normalized links between groups connected to M. The layout of these links can be based on the principle that the distance to group M is inversely proportional to the size of the group. This method allows us to first determine the position of the central node of a class group, and then determine the positions of other central nodes. The correspondence between class groups, domains, and topics in the literature can be found in the generated process file results, which allows for automatic description of class group topics.

[0092] The following specific embodiment further illustrates the cutting-edge topic identification method of the present invention.

[0093] The experimental data for this invention were selected from data of 68 journals covered by the Nature Index from 2014 to 2016, totaling over 170,000 papers. To avoid time lag caused by co-citation, citation coupling was used to establish article relationships, and standardization was applied. The data for each year is divided into two tables: one table shows the topic and external characteristics of each paper in that year, and the other table shows the paper coupling relationships and standard co-citations. The database table has three fields: ESI_ID1 and ESI_ID2 represent any two papers in the collection, and normalized_cocited represents the standard co-citation.

[0094] LDA-based modeling and literature similarity matrix construction:

[0095] This experiment selected 50 topics (K=50) and 2000 iterations. The resulting Topic-doc screenshot is shown below. Figure 2 As shown.

[0096] Based on the LDA model, the similarity between documents is calculated according to the above formula (2), and a document similarity matrix is ​​generated, such as... Figure 3 As shown in the document similarity matrix, doci and docj represent unique identifiers of documents, and sim is the document similarity measure based on the cosine value of topic vectors in the LDA model.

[0097] Finally, the application of community identification and topic visualization yields cutting-edge topic identification results, such as... Figure 4 The image shows visualizations of 2014, 2015, and 2016 generated using a community identification algorithm.

[0098] It should be noted that, in this embodiment of the invention, literature data from five fields—physics, chemistry, life sciences, interdisciplinary fields, and earth and environmental sciences—were selected as the dataset for modeling and analysis. Those skilled in the art will understand that... Figure 4 This is merely a schematic diagram for visualization purposes. The results of data from different fields, their community divisions, and visualization descriptions can be distinguished by different shades of gray in the image. In practical applications, those skilled in the art can intuitively see the visualization results of cutting-edge topic recognition through various methods such as color differentiation, shape differentiation, or perspective differentiation.

[0099] Referring to the above renderings, and taking the 2015 life science analysis as an example, those skilled in the art can analyze the following from the visualizations generated by the community identification algorithm:

[0100] (1) The major categories of life science research have evolved from basic research to clinical translational research.

[0101] (2) In the major category of biology, protein science research has reached a considerable scale, with three important frontier directions being the simulation and prediction of protein structure using molecular dynamics at the molecular level, the intersection of postgenomics and protein science, and biofuels.

[0102] (3) In the medical category, cancer research is the hot topic with the widest range of topics. Brain science, research on nervous system diseases, autoimmune diseases, and regenerative medicine with induced pluripotent stem cells as the main topic have all become research hotspots and frontiers in the year.

[0103] (4) Social medical research has focused more on public health, and characteristic terms related to preventive healthcare such as health care, self-testing health, and mental health have emerged.

[0104] ……etc.

[0105] Therefore, through the above description of the embodiments of the cutting-edge topic identification method, it can be seen that the method of the present invention has high efficiency in analyzing both structural and semantic levels, high accuracy in keyword disambiguation and human interpretation, overcomes the drawbacks of previous methods that analyzed text in a single dimension, and solves the problems of text similarity and fusion. It can be widely applied to text analysis of structure and semantics.

[0106] After introducing the method of exemplary embodiments of the present invention, the following references are made. Figure 5 A frontier topic recognition system according to an exemplary embodiment of the present invention will be described. Implementation of this system can refer to the implementation of the method described above, and repetitions will not be repeated. The terms "module" and "unit" used below can refer to software and / or hardware that implements a predetermined function. Although the modules described in the following embodiments are preferably implemented in software, hardware implementations, or a combination of software and hardware, are also possible and contemplated.

[0107] Figure 5 This is a schematic diagram of the frontier topic recognition system according to an embodiment of the present invention. Figure 5 As shown, the frontier topic recognition system in this embodiment includes:

[0108] The preprocessing module 501 is used to select domain-specific literature as the dataset and preprocess the dataset;

[0109] The document relationship matrix generation module 502 is used to construct and standardize the citation coupling matrix based on the co-citation relationship of documents, and generate a standard document relationship matrix.

[0110] The document similarity matrix generation module 503 is used to perform text modeling using the LDA model, obtain the topic probability distribution of documents, calculate document similarity from the content level, and generate a document similarity matrix.

[0111] Data fusion module 504 is used to linearly fuse the document similarity matrix and the document relationship standardization matrix to form a text similarity matrix;

[0112] The data analysis module 505 is used to obtain topic clusters by utilizing community discovery, and to visualize and describe the topic clusters to obtain the identification results of the cutting-edge topics of the literature in the field.

[0113] The specific implementation of each module can be found in the description of the above method embodiments, and repeated details will not be elaborated upon. Furthermore, although several units of the frontier topic recognition system have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more units described above can be embodied in one unit. Similarly, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0114] This invention also provides a computer device, such as... Figure 6 As shown, it includes a memory 61, a processor 62, and a computer program stored on the memory 61 and executable on the processor 62. When the processor 62 executes the computer program, it implements the frontier topic identification method as described in the above embodiments.

[0115] This invention also provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the frontier topic identification method as described in the above embodiments.

[0116] This invention discloses a cutting-edge topic identification method, system, computer equipment, and storage medium. Oriented towards the structural and semantic fusion of text, it combines the concepts of "entity-relationship" and "semantics-structure," and utilizes text modeling, similarity construction, semantic structure fusion, and automatic topic identification to detect cutting-edge topics. Simultaneously, it studies structural analysis methods and explores engineering implementation issues in intelligence research. The cutting-edge topic identification method of this invention exhibits high efficiency in simultaneously analyzing structural and semantic levels, and high accuracy in keyword disambiguation and manual interpretation. It overcomes the shortcomings of previous methods that relied on single-dimensional text analysis, while also solving the problems of text similarity and fusion. It can be widely applied to structural and semantic text analysis. Furthermore, the model implemented in this invention is a model that combines citation and semantic information for cutting-edge topic identification and structural detection. It can provide researchers with a visual representation of domain structure, cutting-edge directions, and hot topics, offering a convenient and efficient tool for scientists to comprehensively, meticulously, and deeply interpret domain research content, discover hot topics, future research directions, and cutting-edge fields.

[0117] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0118] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0121] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A leading topic identification method characterized by, The method comprises the following steps: Step 1, selecting field-oriented literature as a data set, the data set comprising literature in the field covered by natural index indicators, and preprocessing the data set; Step 2, constructing a citation coupling matrix and standardizing it according to the co-citation relationship of the literature to generate a literature relationship standardized matrix; Step 3, performing text modeling using an LDA model to obtain the topic probability distribution of the document, and calculating the similarity of the literature from the content level to generate a literature similarity matrix; Step 4, linearly fusing the literature similarity matrix and the literature relationship standardized matrix to form a text similarity matrix; Step 5, obtaining a topic cluster by community discovery, and visually describing the topic cluster to obtain the identification result of the frontier topic of the field-oriented literature. In step 4, the linear fusion of the literature relationship standardized matrix and the literature similarity matrix to form a text similarity matrix comprises: The linear fusion algorithm is: in, The weights are set; This represents the document-to-document similarity value in the document similarity matrix. value; This represents the standard co-citation degree in the document relationship standardization matrix; It refers to the nodes of any two mapped documents. Are they adjacent in the basic reference network? Then they are not adjacent, if Then they are adjacent; a and b are two boundary values ​​of similarity in the document similarity matrix; For the determination of a and b, it comprises: viewing the LDA modeling and the literature similarity matrix based on the construction, sorting in the order of similarity from high to low, and selecting top 10%, and the minimum value in top 10% is a; viewing the LDA modeling and the literature similarity matrix based on the construction, sorting in the order of similarity from low to high in descending order, and selecting top 10%, and the maximum value in top 10% is b; The merged values are updated to the literature relation standardized matrix to generate the text similarity matrix. The merged values are updated to the literature relation standardized matrix to generate the text similarity matrix. 2.The front topic identification method of claim 1, wherein, In step 1, the preprocessing of the data set comprises: Vectorizing the text to generate a text-feature word matrix and a text-citation matrix. 3.The front topic identification method of claim 1, wherein, In step 2, the construction of a citation coupling matrix and the standardization thereof according to the co-citation relationship of the literature to generate a literature relationship standard matrix comprises: According to the coupling relationship of the literature, the standard co-citation degree of two literatures is calculated : ； wherein, is the number of references of document x, is the number of references of document y, is the number of common references of document x and document y. 4.The front topic identification method of claim 1, wherein, In step 3, the text modeling using an LDA model to obtain the topic probability distribution of the document, and calculating the similarity of the literature from the content level to generate a literature similarity matrix comprises: Using the LDA model to perform text modeling to obtain a document-topic and topic-word distribution matrix, and mining the latent semantic knowledge of the text; Calculating the similarity between documents according to the cosine direction to generate a literature similarity matrix.

5. The leading topic identification method of claim 4, wherein, Calculating the similarity between documents according to the cosine direction to generate a literature similarity matrix comprises: The topic vector of the literature mapping based on the LDA model is The similarity of two literatures is calculated by the cosine value of the two topic vectors: ； wherein, and denote the mapped topic vectors of the two documents, respectively, is the similarity of the two documents based on the cosine value measure of the LDA model topic vectors.

6. A leading topic identification system characterized by, It comprises: A preprocessing module for selecting field-oriented literature as a data set, the data set comprising literature in the field covered by natural index indicators, and preprocessing the data set; A literature relationship matrix generation module for constructing a citation coupling matrix and standardizing it according to the co-citation relationship of the literature to generate a literature relationship standardized matrix; A literature similarity matrix generation module for performing text modeling using an LDA model to obtain the topic probability distribution of the document, and calculating the similarity of the literature from the content level to generate a literature similarity matrix; A data fusion module for linearly fusing the literature similarity matrix and the literature relationship standardized matrix to form a text similarity matrix; A data analysis module for obtaining a topic cluster by community discovery, and visually describing the topic cluster to obtain the identification result of the frontier topic of the field-oriented literature. The algorithm for linearly fusing the document relationship standardized matrix and the document similarity matrix is: in, The weights are set; This represents the document-to-document similarity value in the document similarity matrix. value; This represents the standard co-citation degree in the document relationship standardization matrix; It refers to the nodes of any two mapped documents. Are they adjacent in the basic reference network? Then they are not adjacent, if Then they are adjacent; a and b are two boundary values ​​of similarity in the document similarity matrix; For the determination of a and b, the following steps are included: viewing the LDA modeling based on the construction and the document similarity matrix, sorting in descending order of similarity, selecting top 10%, and the minimum value in top 10% is a; viewing the LDA modeling based on the construction and the document similarity matrix, sorting in descending order of similarity, selecting top 10%, and the maximum value in top 10% is b; Iterate through the dataset and merge the data. The document relationship normalization matrix is ​​updated to generate the text similarity matrix.

7. A computer device, comprising: The computer program product comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, and the processor implements the front topic identification method according to any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that, The computer program product comprises instructions, which, when executed on a computer, cause the computer to perform the front topic identification method according to any one of claims 1 to 5.

Images