Dynamic topic evolution mining method, system and medium based on dynamic academic network

Abstract

The present invention discloses a method, system and medium for mining dynamic topic evolution based on a dynamic academic network. The method includes: extracting phrase-level technical words from text information, using the extracted technical words and papers to build a dynamic academic network; obtaining a technical word dynamic vector with time series characteristics based on the dynamic academic network, and generating a technical word distribution of the paper in a global sense based on the academic heterogeneous network of the final time series of the dynamic academic network; inputting the technical word dynamic vector and the technical word distribution into a dynamic topic model, generating the distribution of paper topics at different times, modeling the time series characteristics of paper topics, and obtaining the dynamic evolution trajectory of paper topics at different times. The present invention improves the recognition accuracy of technical words, enhances the modeling ability of the dynamic topic model, and enhances the cross-paper dissemination ability of technical words, which is conducive to generating more interpretable and higher quality topics.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning and data mining, and in particular to a dynamic topic evolution mining method, system and medium based on a dynamic academic network. Background Art

[0002] The topic evolution task is a text mining task that conducts research and analysis based on a given document set to explore the temporal evolution of a topic. The task is generally divided into two parts: topic model and topic evolution analysis. Topic models can generally be divided into traditional topic models and topic models based on deep learning. The most intuitive and widely used method for dynamic evolution of topics is to cluster related documents to form research topics, and then calculate the frequency of technical words in each time document set according to time slices, which can be used as the topic research content. In the early stages of the task, more emphasis was placed on the analysis of text content. Subsequently, some methods using citation networks and topic modeling gradually emerged, which more efficiently and scientifically considered the structural information and semantic information between documents in the topic identification and modeling process, in order to discover the evolutionary trends of scientific research hotspots and form the development trajectory of new technologies.

[0003] Topic modeling technology is an important part of topic evolution. It is a technology in the field of text mining that solves the problem of difficulty in extracting hidden information in corpus. Topic modeling can be used to identify research topics from document objects. Topic modeling is mainly used to analyze large unstructured text collections. In the early days of the technology, it mainly used traditional probability methods and statistical methods for calculation and analysis. Usually, the bag-of-words matrix is ​​used as an input to generate the distribution matrix of topics under each document and the distribution matrix of words under each topic. The inner product can reproduce the word matrix of the document with relatively small errors, and obtain the research topic in the document generation process. However, there are complex variational inference contents, which makes it difficult to use deep learning methods.

[0004] With the development of deep learning, neural topic models (NTMs) have been proposed, which use deep neural networks to improve the flexibility and scalability of topic models. Neural topic models can be divided into metadata-based neural topic models, short text neural topic models, sequence neural topic models, neural topic models based on pre-trained language models, neural topic models based on autoregression, neural topic models based on generative adversarial networks, neural topic models based on graph neural networks, and other types. These models use various deep learning methods to solve various problems that exist in traditional topic modeling, but there is also the problem of not being able to fully utilize the heterogeneous information of academic networks.

[0005] In order to analyze the evolution of the topics of a collection of papers over time, more and more studies have begun to consider introducing word vectors into the topic modeling process to obtain more semantically coherent topics. Some studies have proposed using the Transformer-based pre-trained language model Bert model to generate text representations of papers, and then clustering them through TF-IDF (Term Frequency–Inverse Document Frequency, a statistical algorithm) to generate a global vector representation of the topic. The product of the technical word frequency of documents at different timestamps and the pre-calculated global IDF value (used to reflect the importance of a certain word) can be used to obtain a local representation of each topic and model the dynamic evolution of the topic. However, previous methods did not incorporate the rich semantics of academic networks into time series modeling to obtain a more semantically coherent topic evolution trajectory, and the interpretability of the topic was poor.

[0006] Therefore, the prior art still needs to be improved and enhanced. Summary of the invention

[0007] The technical problem to be solved by the present invention is that, in view of the above-mentioned defects of the prior art, a dynamic topic evolution mining method, system and medium based on a dynamic academic network are provided, aiming to solve the problems of the lack of utilization of rich heterogeneous information of academic networks, lack of node feature extraction of academic heterogeneous network neural networks, and the resulting lack of semantic information in the methods of obtaining topics using topic models in the prior art.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0009] In a first aspect, the present invention provides a method for mining dynamic topic evolution based on a dynamic academic network, wherein the method comprises:

[0010] Extract phrase-level technical terms from text information, and use the extracted technical terms and papers to build a dynamic academic network;

[0011] Based on the dynamic academic network, a dynamic vector of technical words with time series characteristics is obtained, and based on the academic heterogeneous network of the final time series of the dynamic academic network, a global distribution of technical words in the paper is generated;

[0012] The technical word dynamic vector and the technical word distribution are input into the dynamic topic model to generate the distribution of paper topics at different times, model the temporal characteristics of paper topics, and obtain the dynamic evolution trajectory of paper topics at different times.

[0013] In one implementation, extracting phrase-level technical terms from text information and building a dynamic academic network using the extracted technical terms and papers includes:

[0014] The named entity method based on the Bert model extracts phrase-level technical terms from text information;

[0015] The extracted technical terms and papers are used as nodes to build an academic heterogeneous network and model a dynamic academic network. The academic heterogeneous network includes the citation and citation relationships between papers, and the occurrence and inclusion relationships between papers and technical terms.

[0016] In one implementation, obtaining a technical word dynamic vector with time series characteristics based on the dynamic academic network includes:

[0017] Group academic papers by year and model the changing characteristics of dynamic academic networks;

[0018] The neighbor node features of the paper nodes and the technical word nodes in the dynamic academic network are weighted and fused, and the node vectors with time series features are extracted through a recurrent neural network. The node vectors with time series features include technical word dynamic vectors with time series features.

[0019] In one implementation, the initial features of the nodes in the dynamic academic network are obtained using a pre-trained language model, wherein the initial features of the technical word nodes are obtained by pre-training using a word2vec model, and the initial features of the paper nodes are generated based on a Bert model.

[0020] In one implementation, the academic heterogeneous network based on the final time sequence of the dynamic academic network generates the global distribution of technical terms in the paper, including:

[0021] Using the academic heterogeneous network of the last time slice of the dynamic academic network, the technical words are propagated on the graph network through the citation relationship between papers;

[0022] The graph attention mechanism model is used to realize message passing aggregation and generate the technical word distribution of the paper in a global sense.

[0023] In one implementation, the technical word dynamic vector and the technical word distribution are input into a dynamic topic model to generate corresponding time variables, model the temporal characteristics of the paper topic, and obtain the dynamic evolution trajectory of the paper topic at different times, including:

[0024] Define dynamic word embedding and topic embedding, and perform inner product operations on dynamic word embedding and topic embedding based on the dynamic topic model to obtain the distribution of paper topics at different times, wherein the dynamic word embedding is initialized with a dynamic vector of a technical word;

[0025] The distribution of technical words and the latent features of the paper are spliced ​​and input into the dynamic topic model, so as to introduce the deep semantic information obtained through feature aggregation in the dynamic academic network into the dynamic topic model;

[0026] The heavy parameter method is used to infer the distribution of paper topics, and the dynamic evolution trajectory of paper topics at different times is obtained based on the distribution of paper topics and technical terms.

[0027] In one implementation, the dynamic evolution trajectory of paper topics at different times is obtained based on the paper topic distribution and the technical word distribution, including:

[0028] Based on the distribution of technical terms and the distribution of paper topics, the inner product operation method is used to reconstruct the distribution of technical terms in the paper;

[0029] The reconstruction loss of technical word distribution is calculated using the technical word distribution of the reconstructed paper;

[0030] Based on the reconstruction loss of the calculated distribution, the overall loss of the dynamic topic model is determined, and the dynamic topic model is optimized based on the overall loss, and the dynamic evolution trajectory of the paper topic at different times is obtained based on the optimized dynamic topic model.

[0031] In a second aspect, an embodiment of the present invention further provides a dynamic topic evolution mining system based on a dynamic academic network, wherein the system comprises:

[0032] Dynamic academic network construction module, which is used to extract phrase-level technical terms from text information and build a dynamic academic network using the extracted technical terms and papers;

[0033] A technical word distribution generation module is used to obtain a technical word dynamic vector with time series characteristics based on the dynamic academic network, and generate a technical word distribution of the paper in a global sense based on the academic heterogeneous network of the final time series of the dynamic academic network;

[0034] The dynamic topic modeling module is used to input the dynamic vector of the technical word and the distribution of the technical word into the dynamic topic model, the distribution of the paper topic at different times, model the temporal characteristics of the paper topic, and obtain the technical dynamic evolution trajectory of the paper topic at different times.

[0035] In the third aspect, an embodiment of the present invention also provides a computer device, wherein the computer device includes a memory, a processor, and a dynamic topic evolution mining program based on a dynamic academic network stored in the memory and executable on the processor, and when the processor executes the dynamic topic evolution mining program based on a dynamic academic network, the steps of the dynamic topic evolution mining method based on a dynamic academic network of any one of the above-mentioned schemes are implemented.

[0036] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein a dynamic topic evolution mining program based on a dynamic academic network is stored on the computer-readable storage medium, and when the dynamic topic evolution mining program based on a dynamic academic network is executed by a processor, the steps of the dynamic topic evolution mining method based on a dynamic academic network described in any one of the above-mentioned schemes are implemented.

[0037] Beneficial effects: Compared with the prior art, the present invention provides a dynamic topic evolution mining method based on a dynamic academic network. The present invention first extracts phrase-level technical words from text information, and then uses the extracted technical words and papers to build a dynamic academic network. Then, a dynamic vector of technical words with time series characteristics is obtained based on the dynamic academic network, and the distribution of technical words in the paper in a global sense is generated based on the academic heterogeneous network of the final time series of the dynamic academic network. Finally, the dynamic vector of technical words and the distribution of technical words are input into the dynamic topic model to generate the distribution of paper topics at different times, model the time series characteristics of paper topics, and obtain the dynamic evolution trajectory of paper topics at different times. The present invention adopts a phrase-level technical word extraction method to improve the recognition accuracy of technical words, and combines the time series characteristics of the dynamic academic network to improve the modeling ability of the dynamic topic model, and also solves the sparsity problem of technical word distribution and enhances the cross-paper dissemination ability of technical words. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flowchart of a preferred embodiment of a dynamic topic evolution mining method based on a dynamic academic network provided in an embodiment of the present invention.

[0039] Figure 2 An algorithm flow chart of a dynamic topic evolution mining method based on a dynamic academic network provided in an embodiment of the present invention.

[0040] Figure 3 A schematic diagram of a dynamic topic model of a dynamic topic evolution mining method based on a dynamic academic network provided in an embodiment of the present invention.

[0041] Figure 4 The invention provides an academic heterogeneous network in a dynamic topic evolution mining method based on a dynamic academic network.

[0042] Figure 5 A schematic diagram of the architecture of a dynamic topic evolution mining system based on a dynamic academic network provided in an embodiment of the present invention.

[0043] Figure 6 A functional block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0045] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations or steps, nor must they be executed in the order described. For example, some operations or steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0046] It should be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0047] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, words such as "first" and "second" are used to distinguish between identical or similar items with substantially identical functions and effects. For example, the first control information and the second control information are only used to distinguish different control information, and their order is not limited.

[0048] Those skilled in the art can understand that the words "first", "second", etc. do not limit the quantity and execution order, and the words "first", "second", etc. do not necessarily limit the differences.

[0049] It should be further understood that the term “and / or” used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0050] The existing dynamic topic model method mainly uses the topic model method to obtain the topic, and models the time characteristics of the paper collection to extract deep semantic information, which helps to generate more consistent topics. However, there is a lack of utilization of rich heterogeneous information in the academic network, a lack of node feature extraction of the academic heterogeneous network neural network, and the resulting lack of semantic information, and there is also a lack of further thinking on the dynamic nature of the topic. Therefore, in order to solve the problems of the prior art, this embodiment provides a dynamic topic evolution mining method based on a dynamic academic network. First, phrase-level technical words are extracted from text information to solve the problem that technical words are only unclearly described as words. Then, a dynamic academic network is built using the extracted technical words and papers, and a dynamic vector of technical words with time series characteristics is obtained based on the dynamic academic network, and the distribution of technical words in the paper in a global sense is generated based on the academic heterogeneous network of the final time series of the dynamic academic network. To address the problem of sparse distribution of technical terms, this embodiment uses a dynamic academic network to adopt a graph attention mechanism to propagate technical terms across papers, inputs the dynamic vector of the technical terms and the distribution of the technical terms into a dynamic topic model, generates the distribution of paper topics at different times, models the temporal features of paper topics, and obtains the dynamic evolution trajectory of paper topics at different times, which solves the problems existing in modeling the dynamic evolution of topics and is conducive to generating topics with stronger explainability and higher quality.

[0051] like Figure 1 As shown, a dynamic topic evolution mining method based on a dynamic academic network is provided. The method is executed by a server or a terminal, or can be executed by a server and a terminal in collaboration. The terminal can be an intelligent product device such as a computer. This embodiment takes the method executed by a terminal as an example for explanation, and includes the following steps:

[0052] Step S100: extract phrase-level technical terms from text information, and use the extracted technical terms and papers to build a dynamic academic network.

[0053] Combination Figure 2 and Figure 3 As shown, Figure 2 This is the algorithm flow of the dynamic topic evolution mining method based on dynamic academic network in this embodiment. Figure 3 is the dynamic topic model of this embodiment, from Figure 3It can be seen that the dynamic topic model of this embodiment includes three modules, namely, building a dynamic academic network, generating a technical word distribution, and a dynamic topic model with labels. Specifically, this embodiment first processes the data to obtain the relevant data information required for the academic network. This embodiment can extract phrase-level technical words from text information based on the named entity method of the Bert model. The Bert model is a language model. By extracting phrase-level technical words, it can solve the problem that technical words are only unclearly described as words. Next, the extracted technical words and papers are used as nodes to build an academic heterogeneous network and model a dynamic academic network. The academic heterogeneous network includes the citation and citation relationships between papers, and the appearance relationship and inclusion relationship between papers and technical words, such as Figure 4 As shown in . The modeled dynamic academic network is conducive to extracting the temporal features of nodes so as to determine the dynamic representation of technical words in the subsequent steps.

[0054] Step S200: obtaining a dynamic vector of technical terms with time series characteristics based on the dynamic academic network, and generating a global distribution of technical terms in the paper based on the academic heterogeneous network of the final time series of the dynamic academic network.

[0055] Since the word vector representation of the same technical word should also change at different times, and subtle semantic changes will occur as the research progresses, this embodiment uses a dynamic academic network to capture the temporal characteristics of technical words. This embodiment can group academic papers by year and model the changing characteristics of the dynamic academic network to extract node features, where the dynamic academic network is , T is time. Then, the neighbor node features of the extracted paper nodes and technical word nodes are weighted and fused, and the node vector with time series features is extracted through a recurrent neural network. The node vector with time series features includes a technical word dynamic vector with time series features. The initial features of the nodes in the dynamic academic network are obtained by using a pre-trained language model, wherein the initial features of the technical word nodes are obtained by pre-training using the word2vec model, and the initial features of the paper nodes are generated based on the Bert model.

[0056] In this embodiment, the dynamic academic network adopts the academic heterogeneous network neural network model (HGT model) for feature fusion, uses the structural characteristics of the constructed dynamic academic network to model the node relationship in the network, and updates the characteristics of each node.

[0057] In practical applications, a node can fuse one-hop neighbor information in one HGT calculation process to update node features. By stacking multiple layers of HGT models, node high-order neighbor features can be fused. However, using too many layers will cause the model to be over-smoothed. Therefore, this embodiment uses a two-layer HGT model to generate node features that fuse high-order neighbor information. , the formula is as follows.

[0058] , where l is a technical word node and i is a paper node. is the feature fusion function of the HGT model. By modeling the academic heterogeneous network t years before the paper was published, the features of each node on the dynamic academic network are obtained. Currently, there is no paper node. The feature aggregation of the node’s neighbor nodes according to the node type can generate the vector representation of the paper in year t. The vector representation of technical word nodes is generated in the same way. , we can get the temporal representation sequence of paper i: , and the dynamic vector representation sequence of technical word nodes The temporal feature sequence of the paper nodes obtained by the HGT model is input into the recurrent neural network model (GRU) to model the temporal characteristics of the dynamic academic network.

[0059] Taking into account the different importance of technical words to papers and the mutual dependence of technical words between mutually cited papers, this embodiment uses the heterogeneous academic network of the last time slice of the dynamic academic network to propagate technical words on the graph network through the citation relationship between papers and adopts the graph attention mechanism model (GAT) to realize message transmission aggregation and generate the technical word distribution of the paper.

[0060] Step S300: input the technical word dynamic vector and the technical word distribution into a dynamic topic model, generate the distribution of paper topics at different times, model the temporal characteristics of paper topics, and obtain the dynamic evolution trajectory of paper topics at different times.

[0061] In order to study the evolution of paper topics and explore the dynamic characteristics of paper topics, this embodiment refers to the dynamic topic model to map technical word embedding and topic embedding to a common semantic space. This embodiment can define dynamic word embedding and theme embed , Dynamic Word Embedding The dynamic vectors of technical words generated by the dynamic academic network can be initialized. They are all three-dimensional tensors. Represents the vector representation of technology words in year t. The vector representation of paper topic k in year t is represented by the following formula:

[0062] ;

[0063] This embodiment can obtain the distribution of paper topics at different times based on the dynamic topic model through the inner product operation of word embedding and topic embedding. .

[0064] Next, this embodiment splices the distribution of technical terms generated on the dynamic academic network with the latent features of the paper as the input of the dynamic topic model, so as to introduce the deep semantic information obtained by feature aggregation in the dynamic academic network into the dynamic topic model. The specific formula is as follows:

[0065] ;

[0066] Where M is a transformation matrix used to map the vector representation of the paper node to the same space as the technical word distribution. is the technical word distribution of paper i, is the hidden layer feature of paper i, This is the input of paper i in the dynamic topic model.

[0067] Next, this embodiment uses a heavy parameter method to infer the paper topic distribution, and obtains the dynamic evolution trajectory of the paper topic at different times based on the paper topic distribution and the technical word distribution. This embodiment can reconstruct the technical word distribution of the paper based on the technical word distribution and the inferred paper topic distribution using the inner product operation method. Then, the reconstruction loss of the technical word distribution is calculated using the reconstructed technical word distribution of the paper. Finally, the dynamic topic evolution is optimized based on the reconstruction loss of the calculated distribution, and the dynamic evolution trajectory of the paper topic at different times is obtained based on the optimized dynamic topic evolution model.

[0068] Specifically, this embodiment uses the method of heavy parameters to infer the paper topic distribution of paper i published in year t, With hidden time variables The concatenation is used as the input of a two-layer fully connected network (MLP) to generate the mean of the Gaussian distribution used to fit the topic of paper i and standard deviation , the specific formula is as follows:

[0069] ;

[0070] ;

[0071] Using the mean and standard deviation Generate the paper topic distribution of paper i , the formula is as follows:

[0072] ;

[0073] ;

[0074] in , That is the result of re-parameterization The topic distribution of paper i.

[0075] Based on the obtained subject technical word distribution and paper subject distribution, the inner product operation method is used to reconstruct the technical word distribution of the paper. yes The formula for the transposed and reconstructed technical word distribution of the paper is as follows:

[0076] ;

[0077] in, The technical word distribution of paper i reconstructed by the dynamic topic model can be used to calculate the reconstruction loss of the technical word distribution. The formula for the reconstruction loss of the technical word distribution is as follows:

[0078] ;

[0079] Where N is the number of papers in a batch when calculating in batches.

[0080] Furthermore, due to the variable , , They all evolve over time, so in order to make their changes smoother, the KL divergence is used to calculate the changes in adjacent time as the loss function component of the dynamic topic model, and the dynamic topic model is optimized. The relevant calculation formula is as follows:

[0081] ;

[0082] ;

[0083] ;

[0084] Where T represents the length of the time series. Referring to the dynamic topic model, the topic label of each paper has been given, that is, the field classification annotated during data extraction. This embodiment requires the loss between the topic distribution reasoning of the measurement model and the topic label of the document. This is a multi-label classification task, and the cross entropy loss is calculated as the classification loss. Since the above loss values ​​are of different magnitudes, in order to unify them for model training optimization, the overall modeling loss of the dynamic topic model in this embodiment is defined as 、 、 、 as well as The weighted sum of is as follows:

[0085] ;

[0086] in, , , and is the weight of each loss value.

[0087] After determining the overall loss of the dynamic topic model, the dynamic topic model is optimized based on the overall loss, and finally the dynamic evolution trajectory of the paper topic at different times is obtained based on the optimized dynamic topic model.

[0088] It can be seen that the dynamic topic evolution mining method based on the dynamic academic network proposed in this embodiment can solve the problems of sparse corpus, difficult to identify technical words and vague themes of words by using the named entity recognition method based on the Bert model to extract technical words, which can obtain phrase-level technical words, and design a dynamic academic network to model the temporal features of technical words, and generate a dynamic vector representation of technical words. For the problem of sparse distribution of technical words, this embodiment uses the academic network to use the graph attention mechanism to propagate technical words across papers, generate the distribution of technical words in the global sense of the paper, and use the distribution of technical words with time characteristics for the parameters of the dynamic topic model to model the time characteristics, solve the problems existing in the modeling of dynamic evolution of topics, and generate more interpretable and higher quality topics.

[0089] Based on the above embodiments, the present invention also provides a dynamic topic evolution mining system based on a dynamic academic network, such as Figure 5 As shown, the system includes: a dynamic academic network construction module 10, a technical word distribution generation module 20 and a dynamic topic modeling module 30. Specifically, the dynamic academic network construction module 10 is used to extract phrase-level technical words from text information, and use the extracted technical words and papers to build a dynamic academic network. The technical word distribution generation module 20 is used to obtain a dynamic vector of technical words with time series characteristics based on the dynamic academic network, and generate a global distribution of technical words in the paper based on the academic heterogeneous network of the final time series of the dynamic academic network. The dynamic topic modeling module 30 is used to input the dynamic vector of technical words and the technical word distribution into the dynamic topic model, the distribution of paper topics at different times, model the time series characteristics of paper topics, and obtain the technical dynamic evolution trajectory of paper topics at different times.

[0090] The working principles of each module in the dynamic topic evolution mining system based on dynamic academic network in this embodiment are the same as the principles of each step in the above method embodiment, and will not be repeated here.

[0091] Each module in the above-mentioned dynamic topic evolution mining system based on dynamic academic network can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0092] In some embodiments, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a dynamic topic evolution mining method based on a dynamic academic network is implemented. The display unit of the computer device is used to form a visually visible image, and can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse, etc.

[0093] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0094] In some embodiments, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0095] Extract phrase-level technical terms from text information, and use the extracted technical terms and papers to build a dynamic academic network;

[0096] Based on the dynamic academic network, a dynamic vector of technical words with time series characteristics is obtained, and based on the heterogeneous graph of the final time series of the dynamic academic network, a global distribution of technical words in the paper is generated;

[0097] The technical word dynamic vector and the technical word distribution are input into the dynamic topic model to generate the distribution of paper topics at different times, model the temporal characteristics of paper topics, and obtain the dynamic evolution trajectory of paper topics at different times.

[0098] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0099] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0100] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A dynamic topic evolution mining method based on dynamic academic network, characterized in that: The method comprises: Extract phrase-level technical terms from text information, and use the extracted technical terms and papers to build a dynamic academic network; Based on the dynamic academic network, a dynamic vector of technical words with time series characteristics is obtained, and based on the academic heterogeneous network of the final time series of the dynamic academic network, a global distribution of technical words in the paper is generated; Input the technical word dynamic vector and the technical word distribution into the dynamic topic model to generate the distribution of paper topics at different times, model the temporal characteristics of paper topics, and obtain the dynamic evolution trajectory of paper topics at different times; The method of extracting phrase-level technical terms from text information and building a dynamic academic network using the extracted technical terms and papers includes: The named entity method based on the Bert model extracts phrase-level technical terms from text information; Using the extracted technical terms and papers as nodes, respectively, to build an academic heterogeneous network and model a dynamic academic network. The academic heterogeneous network includes the citation and citation relationships between papers, and the occurrence and inclusion relationships between papers and technical terms. The obtaining of a technical word dynamic vector with time series characteristics based on the dynamic academic network includes: Group academic papers by year and model the changing characteristics of dynamic academic networks; The neighbor node features of the paper nodes and the technical word nodes in the dynamic academic network are weighted and fused, and the node vectors with time series features are extracted through a recurrent neural network. The node vectors with time series features include technical word dynamic vectors with time series features.

2. The method for mining dynamic topic evolution based on dynamic academic network according to claim 1, characterized in that: The initial features of the nodes in the dynamic academic network are obtained by using a pre-trained language model, wherein the initial features of the technical word nodes are obtained by pre-training using a word2vec model, and the initial features of the paper nodes are generated based on a Bert model.

3. The method for mining dynamic topic evolution based on dynamic academic network according to claim 1, characterized in that: The academic heterogeneous network based on the final time series of the dynamic academic network generates the global distribution of technical terms in the paper, including: Using the academic heterogeneous network of the last time slice of the dynamic academic network, the technical words are propagated on the graph network through the citation relationship between papers; The graph attention mechanism model is used to realize message passing aggregation and generate the technical word distribution of the paper in a global sense.

4. The method for mining dynamic topic evolution based on dynamic academic network according to claim 1, characterized in that: The technical word dynamic vector and the technical word distribution are input into the dynamic topic model to generate corresponding time variables, model the temporal characteristics of the paper topic, and obtain the dynamic evolution trajectory of the paper topic at different times, including: Define dynamic word embedding and topic embedding, and perform inner product operation on dynamic word embedding and topic embedding based on the dynamic topic model to obtain the distribution of paper topics at different times, wherein the dynamic word embedding is initialized with a dynamic vector of a technical word; The distribution of technical words and the latent features of the paper are spliced ​​and input into the dynamic topic model, so as to introduce the deep semantic information obtained through feature aggregation in the dynamic academic network into the dynamic topic model; The heavy parameter method is used to infer the distribution of paper topics, and the dynamic evolution trajectory of paper topics at different times is obtained based on the distribution of paper topics and technical terms.

5. The method for mining dynamic topic evolution based on dynamic academic network according to claim 4, characterized in that: The dynamic evolution track of the paper topics at different times is obtained based on the paper topic distribution and the technical word distribution, including: Based on the distribution of technical terms and the distribution of paper topics, the inner product operation method is used to reconstruct the distribution of technical terms in the paper; The reconstruction loss of technical word distribution is calculated using the technical word distribution of the reconstructed paper; Based on the reconstruction loss of the calculated distribution, the overall loss of the dynamic topic model is determined, and the dynamic topic model is optimized based on the overall loss, and the dynamic evolution trajectory of the paper topic at different times is obtained based on the optimized dynamic topic model.

6. A dynamic topic evolution mining system based on dynamic academic network, characterized in that: The system is used to implement the steps of the dynamic topic evolution mining method based on a dynamic academic network according to any one of claims 1 to 5, and the system comprises: Dynamic academic network construction module, which is used to extract phrase-level technical terms from text information and build a dynamic academic network using the extracted technical terms and papers; A technical word distribution generation module is used to obtain a technical word dynamic vector with time series characteristics based on the dynamic academic network, and generate a technical word distribution of the paper in a global sense based on the academic heterogeneous network of the final time series of the dynamic academic network; The dynamic topic modeling module is used to input the dynamic vector of the technical word and the distribution of the technical word into the dynamic topic model, the distribution of the paper topic at different times, model the temporal characteristics of the paper topic, and obtain the technical dynamic evolution trajectory of the paper topic at different times.

7. A computer device, characterized in that: The computer device includes a memory, a processor, and a dynamic topic evolution mining program based on a dynamic academic network stored in the memory and executable on the processor. When the processor executes the dynamic topic evolution mining program based on a dynamic academic network, the steps of the dynamic topic evolution mining method based on a dynamic academic network as described in any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a dynamic topic evolution mining program based on a dynamic academic network. When the dynamic topic evolution mining program based on a dynamic academic network is executed by a processor, the steps of the dynamic topic evolution mining method based on a dynamic academic network as described in any one of claims 1-5 are implemented.

Images