Data mining method, device, equipment and storage medium

Abstract

The present invention provides a data mining method device, equipment and storage medium, wherein the method includes: performing data preprocessing on multi-source heterogeneous scientific research data, generating a data set in a unified format, and mining a multi-dimensional scientific research portrait feature set therefrom. Based on these feature sets, pattern mining is performed on the academic output data of researchers to obtain research interest pattern sets and research trend indicators. Through predictive mining, a research interest prediction model and project matching score are generated for historical research data, and the model is used to perform graph structure mining on the researcher cooperation network to predict potential cooperation relationships. Finally, based on the project matching score, network structure characteristics and cooperation relationship prediction results, knowledge mining is performed on the scientific research project data to generate a personalized project recommendation list and its explainable recommendation basis. This method can comprehensively consider the evolution of researchers' interests, project characteristics, and cooperation potential, and improve the accuracy and interpretability of recommendations.

Description

Technical Field

[0001] The present invention relates to the field of data mining, and in particular to a data mining method, device, equipment and storage medium. Background Art

[0002] With the increasing complexity and diversification of scientific research activities, researchers face great challenges in choosing research directions and projects. Traditional methods for identifying scientific research interests and recommending projects mainly rely on keyword matching and simple statistical analysis, which cannot fully utilize multi-source heterogeneous scientific research data. In the existing technology, some studies have tried to use machine learning and data mining techniques to improve the accuracy of scientific research interest identification and project recommendation. For example, some studies use topic models to analyze researchers' published literature to identify their research interests; some studies use collaborative filtering algorithms to recommend scientific research projects. Some methods also try to combine social network analysis to consider the cooperative relationship between researchers to optimize the recommendation results. These existing methods have a major flaw: they usually treat interest identification, project matching, and cooperative relationship analysis as independent tasks, lacking a unified framework to integrate information from these different dimensions. This results in the recommendation results often failing to fully consider multiple aspects such as the evolution of researchers' interests, project characteristics, and cooperation potential, thereby limiting the accuracy and interpretability of the recommendations. Summary of the invention

[0003] The main purpose of the present invention is to solve the technical problem that the existing data mining methods treat interest identification, project matching and cooperative relationship analysis as independent tasks, which limits the accuracy and interpretability of recommendations;

[0004] A first aspect of the present invention provides a data mining method, the data mining method comprising:

[0005] Perform data mining preprocessing on multi-source heterogeneous scientific research data to obtain a preprocessed unified format data set, and perform feature mining on the unified format data set to obtain a multi-dimensional scientific research portrait feature set;

[0006] Based on the academic output data in the multi-dimensional scientific research portrait feature set, pattern mining is performed on the research activities of researchers to obtain a research interest pattern set and research trend indicators;

[0007] According to the research interest pattern set and research trend indicators, predictive mining is performed on the historical research data in the multi-dimensional scientific research portrait feature set to obtain a research interest prediction model and a project matching score;

[0008] According to the research interest prediction model, the graph structure mining is performed on the researcher cooperation network in the multi-dimensional scientific research portrait feature set to obtain the network structure characteristics and potential cooperation relationship prediction results;

[0009] Based on the project matching scores, network structure characteristics and potential cooperation relationship prediction results, knowledge mining is performed on the research project data of the multi-source heterogeneous scientific research data to obtain mining results, and the mining results are comprehensively processed and knowledge represented to obtain a personalized project recommendation list and explainable recommendation basis.

[0010] Optionally, in a first implementation of the first aspect of the present invention, the pattern mining of the researchers' research activities is performed based on the academic output data in the multi-dimensional scientific research portrait feature set to obtain a research interest pattern set and research trend indicators including:

[0011] Performing topic modeling and keyword extraction on the academic output data in the multi-dimensional scientific research portrait feature set to obtain research topic distribution and keyword set;

[0012] According to the research topic distribution and keyword set, a time series cluster analysis is performed on the academic achievements of researchers in the academic output data to obtain the evolution trajectory of research interests;

[0013] Based on the evolution trajectory of the research interests, a frequent pattern mining algorithm is used to analyze the research behavior sequence in the academic output data to obtain a research interest pattern set;

[0014] The trend prediction algorithm and anomaly detection method are applied to the research interest pattern set, combined with preset external discipline development data, to obtain research trend indicators.

[0015] Optionally, in a second implementation of the first aspect of the present invention, based on the research interest evolution trajectory, a frequent pattern mining algorithm is used to analyze the research behavior sequence in the academic output data to obtain a research interest pattern set including:

[0016] Segmenting the evolution trajectory of the research interest into time windows to obtain multiple time series segments;

[0017] According to the time series segments, the research behavior sequences in the academic output data are coded and serialized to obtain a standardized behavior sequence set;

[0018] Applying a preset sequence pattern mining algorithm to the behavior sequence set to extract frequent subsequence patterns and obtain candidate research interest patterns;

[0019] In combination with the research topic distribution, a hierarchical clustering algorithm is used to classify and merge the candidate research interest patterns to obtain a research interest pattern set.

[0020] Optionally, in a third implementation of the first aspect of the present invention, predictive mining is performed on the historical research data in the multi-dimensional scientific research portrait feature set according to the research interest pattern set and the research trend index to obtain a research interest prediction model and a project matching score, including:

[0021] Performing feature fusion on the research interest pattern set and the research trend index, and combining the historical research data in the multi-dimensional scientific research portrait feature set to obtain a comprehensive feature vector;

[0022] Based on the comprehensive feature vector, the time series decomposition technique is used to separate the trend, cycle and random components of the historical research data to obtain multidimensional time series components;

[0023] Applying a deep learning model to the multidimensional time series component to generate a research interest prediction model, and performing interest prediction for a preset future time point based on the research interest prediction model and the research activity data in the multidimensional scientific research portrait feature set to obtain a predicted interest vector;

[0024] According to the predicted interest vector, combined with the preset project feature vector, the project matching score is obtained through cosine similarity calculation and multi-factor weighting.

[0025] Optionally, in a fourth implementation of the first aspect of the present invention, the research interest prediction model is used to perform graph structure mining on the researcher cooperation network in the multi-dimensional scientific research portrait feature set to obtain network structure features and potential cooperation relationship prediction results, including:

[0026] Performing graph embedding processing on the researcher cooperation network in the multi-dimensional scientific research portrait feature set to obtain a low-dimensional node representation vector;

[0027] According to the node representation vector and the output of the research interest prediction model, the attributes of the researcher nodes in the researcher cooperation network are enhanced to obtain an enhanced node representation that incorporates interest information;

[0028] Applying a graph attention network to the enhanced node representation to perform multi-level information propagation and aggregation to obtain dynamically updated node features;

[0029] According to the node characteristics, the community detection algorithm is used to analyze the researcher cooperation network to obtain the network structure characteristics;

[0030] Based on the network structure characteristics and the node characteristics, link prediction is performed through a graph neural network to obtain a potential cooperation relationship prediction result.

[0031] Optionally, in a fifth implementation of the first aspect of the present invention, the research project data of the multi-source heterogeneous scientific research data is subjected to knowledge mining according to the project matching score, network structure characteristics and potential cooperative relationship prediction results to obtain mining results, and the mining results are subjected to comprehensive processing and knowledge representation to obtain a personalized project recommendation list and an explainable recommendation basis, including:

[0032] Performing multimodal feature fusion on the project matching scores, network structure characteristics and potential cooperative relationship prediction results to obtain a comprehensive feature representation;

[0033] According to the comprehensive feature representation, semantic analysis and topic modeling are performed on the research project data of the multi-source heterogeneous scientific research data to obtain a project semantic network;

[0034] Based on the semantic network and comprehensive feature representation of the project, a knowledge graph construction algorithm is used to mine and infer the associations between projects to obtain a scientific research project knowledge graph;

[0035] Using graph neural networks and attention mechanisms, multi-dimensional correlation calculation and sorting are performed on the scientific research project knowledge graph to obtain a personalized project recommendation list;

[0036] For each item in the personalized project recommendation list, relationship path analysis and feature importance quantification are performed based on the scientific research project knowledge graph to obtain an explainable recommendation basis.

[0037] Optionally, in a sixth implementation of the first aspect of the present invention, for each item in the personalized project recommendation list, performing relationship path analysis and feature importance quantification based on the scientific research project knowledge graph to obtain an explainable recommendation basis includes:

[0038] Perform a bidirectional breadth-first search on the nodes related to the recommended projects in the scientific research project knowledge graph to obtain a multi-level relationship path set;

[0039] Performing semantic embedding processing on the nodes and edges of the paths in the relationship path set to obtain a path representation vector;

[0040] Applying an attention mechanism and a path pruning algorithm to the path representation vector to obtain a core interpretation path, and quantifying feature contributions based on the core interpretation path and the comprehensive feature representation to obtain a feature importance ranking;

[0041] According to the core explanation path and feature importance ranking, a structured explanation template is constructed through natural language generation technology to obtain an explainable recommendation basis.

[0042] A second aspect of the present invention provides a data mining device, the data mining device comprising:

[0043] A feature mining module is used to perform data mining preprocessing on multi-source heterogeneous scientific research data to obtain a preprocessed unified format data set, and perform feature mining on the unified format data set to obtain a multi-dimensional scientific research portrait feature set;

[0044] A pattern mining module is used to perform pattern mining on the research activities of researchers based on the academic output data in the multi-dimensional scientific research portrait feature set to obtain a research interest pattern set and research trend indicators;

[0045] A predictive mining module is used to perform predictive mining on the historical research data in the multi-dimensional scientific research portrait feature set according to the research interest pattern set and research trend indicators to obtain a research interest prediction model and a project matching score;

[0046] A graph structure mining module is used to perform graph structure mining on the researcher cooperation network in the multi-dimensional scientific research portrait feature set according to the research interest prediction model, and obtain network structure characteristics and potential cooperation relationship prediction results;

[0047] The knowledge mining module is used to perform knowledge mining on the research project data of the multi-source heterogeneous scientific research data according to the project matching scores, network structure characteristics and potential cooperation relationship prediction results, obtain mining results, and perform comprehensive processing and knowledge representation on the mining results to obtain a personalized project recommendation list and explainable recommendation basis.

[0048] The third aspect of the present invention provides a data mining device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the data mining device executes the steps of the above-mentioned data mining method.

[0049] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the above-mentioned data mining method.

[0050] The above-mentioned data mining method device, equipment and storage medium generate a unified format data set by preprocessing multi-source heterogeneous scientific research data, and mine multi-dimensional scientific research portrait feature sets from them. Based on these feature sets, pattern mining is performed on the academic output data of researchers to obtain research interest pattern sets and research trend indicators. Through predictive mining, a research interest prediction model and project matching score are generated for historical research data, and the model is used to perform graph structure mining on the researcher cooperation network to predict potential cooperation relationships. Finally, based on the project matching score, network structure characteristics and cooperation relationship prediction results, knowledge mining is performed on the scientific research project data to generate a personalized project recommendation list and its explainable recommendation basis. This method can comprehensively consider the evolution of researchers' interests, project characteristics, and cooperation potential, and improve the accuracy and interpretability of recommendations.

[0051] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0052] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of a first embodiment of a data mining method in an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of an embodiment of a data mining device in an embodiment of the present invention;

[0055] Figure 3 The figure is a schematic diagram of an embodiment of a data mining device in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0058] To facilitate understanding of this embodiment, a data mining method disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps:

[0059] 101. Perform data mining preprocessing on multi-source heterogeneous scientific research data to obtain a unified format data set after preprocessing, and perform feature mining on the unified format data set to obtain a multi-dimensional scientific research portrait feature set;

[0060] In one embodiment of the present invention, it is first necessary to determine the source of multi-source heterogeneous scientific research data. These data generally include academic paper databases, patent databases, scientific research project applications and funding records, personal academic homepages, social media platforms, academic conferences and seminar records, and scientific research management systems within institutions. For each data source, it is necessary to develop a corresponding data acquisition module, which involves methods such as web crawler technology, API interface calls, database queries, and natural language processing technologies. The collected raw data often has problems such as inconsistent formats, missing or redundant information, so data cleaning and preprocessing are required. This process includes removing duplicate and invalid data, unifying data formats, processing missing values ​​and outliers, and performing data standardization and normalization processing. In particular, for text data, natural language processing operations such as word segmentation, stop word removal, and part-of-speech tagging are also required. In addition, it is also necessary to solve data inconsistencies between different data sources, such as name spelling differences, changes in institution names, etc., which can be achieved through entity alignment technology. The preprocessed data needs to be integrated to form a data set in a unified format. This process involves data integration technology, which organizes and stores data from different sources in a predefined format. Relational databases can be used to store structured data, document databases can be used to store semi-structured data, and for large-scale raw data, distributed file systems can be considered for storage. Next, feature mining is performed on the data set in a unified format to construct a multi-dimensional scientific research portrait feature set. This process first involves feature extraction, extracting meaningful features from different types of data. For text data, semantic features can be extracted using technologies such as TF-IDF and Word2Vec; for numerical data, various statistical indicators can be calculated; for time series data, trends, periodicity and other features can be extracted. Then, the most relevant and representative features are selected through feature selection techniques such as chi-square test and information gain. In order to capture the dynamic characteristics of researchers' academic activities, time series features need to be constructed. This can be achieved by performing time series analysis on the academic output of researchers, such as using sliding window technology to calculate feature changes in different time periods. At the same time, in order to reflect the position and influence of researchers in the academic network, network features need to be constructed. This can be achieved by performing social network analysis on the cooperative relationship network and calculating indicators such as centrality and structural holes. Finally, all extracted features are integrated and standardized to form a multi-dimensional scientific research portrait feature set. This feature set should be able to fully reflect the researcher's academic background, research interests, cooperation model, influence, and other aspects. In order to facilitate subsequent analysis and model training, these features can be represented as high-dimensional vectors or matrices. At the same time, considering the dynamic nature of scientific research activities, it is necessary to design an incremental update mechanism so that the feature set can be updated efficiently when new data arrives.

[0061] 102. Based on the academic output data in the multi-dimensional scientific research portrait feature set, the research activities of researchers are pattern mined to obtain research interest pattern sets and research trend indicators;

[0062] In one embodiment of the present invention, pattern mining is performed on the research activities of researchers based on the academic output data in the multi-dimensional scientific research portrait feature set to obtain a research interest pattern set and a research trend indicator, including: subject modeling and keyword extraction are performed on the academic output data in the multi-dimensional scientific research portrait feature set to obtain a research topic distribution and a keyword set; based on the research topic distribution and the keyword set, a time series clustering analysis is performed on the academic achievements of researchers in the academic output data to obtain a research interest evolution trajectory; based on the research interest evolution trajectory, a frequent pattern mining algorithm is used to analyze the research behavior sequence in the academic output data to obtain a research interest pattern set; trend prediction algorithms and anomaly detection methods are applied to the research interest pattern set, combined with preset external discipline development data, to obtain research trend indicators.

[0063] Specifically, first, the academic output data in the multi-dimensional scientific research portrait feature set is subjected to topic modeling and keyword extraction. The topic modeling can use the latent Dirichlet allocation (LDA) algorithm, which can discover potential topic structures from a large number of documents. In the specific implementation, the researcher's academic papers, patent texts, etc. are used as input documents, a suitable number of topics is predefined, and then the document-topic distribution and topic-word distribution are estimated through iterative optimization. In this way, the topic distribution of each document can be obtained, and then the overall research topic distribution of the researcher can be obtained by aggregating these distributions. At the same time, keyword extraction can use the TF-IDF (term frequency-inverse document frequency) method, which can identify words with high importance in a specific document. By setting a threshold or selecting the top-N method, a set of keywords can be extracted for each document, and then the keyword set of the researcher can be summarized. Next, according to the obtained research topic distribution and keyword set, the academic achievements of the researcher in the academic output data are subjected to time series clustering analysis. The purpose of this step is to track the evolution of research interests. First, the academic achievements need to be arranged in chronological order, and then the dynamic time warping (DTW) algorithm can be used to measure the similarity between academic achievements at different time points. The DTW algorithm can handle the problem of inconsistent time series lengths and is suitable for analyzing research activities in different periods. Based on the DTW distance metric, hierarchical clustering or density clustering algorithms (such as DBSCAN) can be used to identify research interest clusters in different periods. By analyzing the evolution of these clusters, the evolution trajectory of research interests can be obtained, which reflects the changes and shifts of researchers' interests over time. Based on the obtained research interest evolution trajectory, the next step is to use frequent pattern mining algorithms to analyze the research behavior sequences in academic output data. Projection-based sequence pattern mining algorithms such as the PrefixSpan algorithm can be used here. First, the research behavior needs to be encoded into a sequence. For example, each research topic or keyword can be regarded as a project, and then the sequence is constructed in chronological order. The PrefixSpan algorithm discovers frequent sequence patterns by recursively constructing a prefix projection database. It can efficiently process long sequences and large-scale data sets. By setting the minimum support threshold, a series of frequently occurring research behavior patterns can be obtained, which constitute the research interest pattern set. Finally, the trend prediction algorithm and anomaly detection method are applied to the obtained research interest pattern set, and combined with the preset external discipline development data to obtain research trend indicators. Trend forecasting can use time series analysis methods such as ARIMA (Autoregressive Integrated Moving Average) model or Prophet model. These models can capture the trend, periodicity and seasonality characteristics in time series data, so as to predict future development trends. For each research interest pattern, its frequency changes in the past period of time can be analyzed and future development trends can be predicted.Anomaly detection methods can use statistical or machine learning-based methods, such as One-Class SVM or Isolation Forest algorithm, which can identify research behaviors that are significantly different from normal patterns and may represent emerging research directions or innovations. When performing trend prediction and anomaly detection, it is also necessary to combine preset external discipline development data. These data include the overall development trend of the discipline, hot topics of important conferences or journals, and priority funding directions of research funding agencies. By comparing and integrating the personal research interest patterns of researchers with these external data, more comprehensive and objective research trend indicators can be obtained. Specifically, the matching degree between personal research interests and disciplinary hotspots can be calculated, the cutting-edge nature and potential influence of research directions can be evaluated, and the future development potential of certain research topics can be predicted.

[0064] Furthermore, based on the research interest evolution trajectory, a frequent pattern mining algorithm is used to analyze the research behavior sequences in the academic output data to obtain a research interest pattern set, including: segmenting the research interest evolution trajectory into time windows to obtain multiple time series segments; encoding and serializing the research behavior sequences in the academic output data according to the time series segments to obtain a standardized behavior sequence set; applying a preset sequence pattern mining algorithm to the behavior sequence set to extract frequent subsequence patterns to obtain candidate research interest patterns; and combining the research topic distribution with a hierarchical clustering algorithm to classify and merge the candidate research interest patterns to obtain a research interest pattern set.

[0065] Specifically, first, the evolution trajectory of research interests is segmented into time windows. This step aims to divide the continuous time series data into discrete segments for subsequent pattern mining. In specific implementation, the sliding window technology can be used to set a fixed-size time window (for example, 6 months or 1 year), and then slide the window along the time axis, moving a certain step size (such as 1 month) each time. In this way, a series of overlapping time series segments can be obtained, each of which represents the state of research interests in a specific period. In order to capture patterns at different scales, multiple time windows of different sizes can also be set to obtain a multi-scale set of time series segments. Next, based on these time series segments, the research behavior sequence in the academic output data is encoded and serialized. The key to the encoding process is to convert complex research behaviors into concise and meaningful symbolic or digital representations. For example, a unique identifier can be assigned to each research topic or keyword, and then the main research activities of researchers in each time window can be represented as a sequence of these identifiers. Specifically, the top N topics or keywords with the highest frequency in each time window can be selected as representatives of the window, and sorted according to their importance to form an ordered symbol sequence. This encoding method not only retains the temporal information of the research behavior, but also greatly simplifies the data structure, which is convenient for subsequent pattern mining. The serialization process is to organize these encoded sequences in chronological order to form a standardized behavior sequence set. This sequence set contains the evolution process of the researchers' research behavior during the entire observation period. After obtaining the standardized behavior sequence set, the next step is to apply the preset sequence pattern mining algorithm to extract frequent subsequence patterns. Here, you can choose to use a projection-based sequence pattern mining algorithm, such as PrefixSpan or its improved version. The core idea of ​​the PrefixSpan algorithm is to discover frequent sequence patterns by recursively building a prefix projection database, which can effectively handle long sequences and large-scale data sets. The algorithm first needs to set a minimum support threshold, and only subsequences with a frequency exceeding this threshold will be regarded as frequent patterns. Then, the algorithm starts with a sequence of length 1 and gradually expands the prefix. Each expansion will generate a projection database and search for frequent items in this database. This process will be recursively performed until no longer frequent sequences can be found. In this way, the algorithm can efficiently discover all frequent subsequences that meet the minimum support requirements, and these subsequences constitute the candidate research interest pattern set. Finally, in order to further refine and organize these candidate research interest patterns, it is necessary to classify and merge them using a hierarchical clustering algorithm in combination with the research topic distribution. First, a feature vector needs to be calculated for each candidate research interest pattern, which can be constructed based on the distribution of research topics contained in the pattern. Specifically, the frequency of each topic in the pattern can be counted, and these frequency values ​​can be normalized to form a topic distribution vector.Then, these feature vectors are clustered using a hierarchical clustering algorithm. The advantage of the hierarchical clustering algorithm is that it does not require the number of clusters to be specified in advance, but rather gradually merges or splits clusters in a bottom-up or top-down manner. In this process, metrics such as Euclidean distance or cosine similarity can be used to calculate the similarity between patterns. The clustering process can be controlled by setting a distance threshold, and the merging is stopped when the distance between any two clusters is greater than this threshold. Each cluster finally represents a class of similar research interest patterns, and the center of the cluster can be used as a representative of this class of patterns. In this way, a set of classified and merged research interest patterns is obtained, which not only reflects the temporal characteristics of research behavior, but also takes into account the semantic similarity of research topics.

[0066] 103. Based on the research interest pattern set and research trend indicators, predictive mining is performed on the historical research data in the multi-dimensional scientific research portrait feature set to obtain the research interest prediction model and project matching score;

[0067] In one embodiment of the present invention, predictive mining is performed on the historical research data in the multidimensional scientific research portrait feature set according to the research interest pattern set and research trend indicators to obtain a research interest prediction model and a project matching score, including: feature fusion of the research interest pattern set and research trend indicators, combined with the historical research data in the multidimensional scientific research portrait feature set, to obtain a comprehensive feature vector; based on the comprehensive feature vector, time series decomposition technology is used to separate the trend, cycle and random components of the historical research data to obtain a multidimensional time series component; a deep learning model is applied to the multidimensional time series component to generate a research interest prediction model, and based on the research interest prediction model and the research activity data in the multidimensional scientific research portrait feature set, interest prediction is performed for a preset future time point to obtain a predicted interest vector; based on the predicted interest vector, combined with a preset project feature vector, a project matching score is obtained through cosine similarity calculation and multi-factor weighting.

[0068] Specifically, first, the research interest pattern set and the research trend index are feature fused. This step aims to integrate the research interest pattern and trend information obtained in the previous step into a unified feature representation. In specific implementation, the feature splicing method can be used to represent each pattern in the research interest pattern set as a vector, in which each element represents a certain feature of the pattern (such as the frequency and duration of the pattern); at the same time, the research trend index is also converted into a vector form, and each element corresponds to a specific trend indicator value. Then, the two vectors are spliced ​​together to form a preliminary fused feature vector. Next, this fused feature vector needs to be combined with the historical research data in the multi-dimensional scientific research portrait feature set. Historical research data includes information in multiple dimensions such as the number of published papers, the number of citations, and the characteristics of the cooperative relationship network. By converting these historical data into vector form and further splicing or weighted fusion with the fused feature vector obtained previously, a comprehensive comprehensive feature vector is finally obtained. This comprehensive feature vector contains both the researcher's past research interests and trend information and various aspects of his historical research performance. After obtaining the comprehensive feature vector, the next step is to use time series decomposition technology to separate the trend, cycle and random components of the historical research data. The purpose of this step is to deeply understand the inherent structure and changing laws of historical research data. Common time series decomposition methods include classical decomposition and STL (Seasonal and Trend decomposition using Loess) decomposition. Taking STL decomposition as an example, it can effectively handle nonlinear trends and multiple seasonality. In specific implementation, it is first necessary to extract the time-related features in the comprehensive feature vector to form one or more time series. Then, the STL decomposition algorithm is applied to each time series to decompose it into trend components, seasonal components, and residual (random) components. The trend component reflects the long-term trend of research activities, the seasonal component captures possible periodic patterns (such as the annual academic conference cycle), and the random component contains short-term fluctuations and noise. Through this decomposition, multidimensional time series components are obtained, each of which represents an important aspect of historical research data. Next, a deep learning model is applied to the obtained multidimensional time series components to generate a research interest prediction model. Considering the characteristics of time series data, it is suitable to use recurrent neural networks (RNN) or their variants, such as long short-term memory networks (LSTM) or gated recurrent units (GRU). These models can effectively capture long-term dependencies in time series data. In the specific implementation, we first need to organize the multi-dimensional time series components into a format suitable for deep learning model input, usually a three-dimensional tensor, whose dimensions represent the number of samples, time steps, and number of features. Then, we build a network structure containing multiple layers of LSTM or GRU units, and consider adding an attention mechanism to enhance the model's perception of key time points.The training process of the model adopts supervised learning, using part of the historical data as input to predict the research interest of the next time step. Through the back-propagation algorithm and the gradient descent optimizer, the model parameters are continuously adjusted until the performance of the model on the validation set reaches the best. After the training is completed, a deep learning model that can predict future research interests is obtained, namely the research interest prediction model. With the research interest prediction model, the next step is to use this model to predict the interest at a preset future time point. Here, it is necessary to combine the latest research activity data in the multi-dimensional scientific research portrait feature set as the input of the model. Specifically, the research activity data of the recent period (such as the past year) can be selected, and the model input is formed after the same preprocessing and feature extraction process as the training data. Then, the trained deep learning model is used to forward propagate these input data to obtain the prediction results for several future time points (such as the next three years). These prediction results can be expressed as a series of vectors, each of which represents the predicted interest distribution at a future time point, collectively referred to as the predicted interest vector. The last step is to calculate the project matching score based on the obtained predicted interest vector and the preset project feature vector. The project feature vector is a characteristic description of each possible research project, including the research topic, required skills, funding scale and other factors of the project. The process of calculating the matching degree first uses the cosine similarity method to calculate the similarity between the predicted interest vector and each project feature vector. Cosine similarity can effectively measure the directional similarity of two vectors in high-dimensional space, and is suitable for measuring the matching degree between research interests and project characteristics. Then, considering that different factors may have different importance, a multi-factor weighting mechanism is introduced. The factors here include the professional background of researchers, past project experience, cooperation network, etc. By assigning different weights to these factors and multiplying them with the cosine similarity calculation results, a comprehensive project matching score is finally obtained. This score takes into account both the future interest prediction of researchers and the characteristics of the project and other relevant factors, and can provide researchers with more personalized and forward-looking project recommendations.

[0069] 104. Based on the research interest prediction model, the graph structure mining of the researcher cooperation network in the multi-dimensional scientific research portrait feature set is carried out to obtain the network structure characteristics and potential cooperation relationship prediction results;

[0070] In one embodiment of the present invention, the research interest prediction model is used to perform graph structure mining on the researcher cooperation network in the multi-dimensional scientific research portrait feature set to obtain network structure characteristics and potential cooperation relationship prediction results, including: performing graph embedding processing on the researcher cooperation network in the multi-dimensional scientific research portrait feature set to obtain a low-dimensional node representation vector; according to the node representation vector and the output of the research interest prediction model, the researcher nodes in the researcher cooperation network are attribute enhanced to obtain an enhanced node representation that integrates interest information; a graph attention network is applied to the enhanced node representation to perform multi-level information dissemination and aggregation to obtain dynamically updated node characteristics; according to the node characteristics, a community detection algorithm is used to analyze the researcher cooperation network to obtain network structure characteristics; based on the network structure characteristics and the node characteristics, link prediction is performed through a graph neural network to obtain potential cooperation relationship prediction results.

[0071] Specifically, the researcher cooperation network in the multi-dimensional scientific research portrait feature set is first processed by graph embedding. Graph embedding is a technology that maps high-dimensional graph structure data to low-dimensional vector space. The purpose is to reduce the dimension of the data while retaining the graph structure information, so as to facilitate subsequent processing and analysis. In specific implementation, algorithms such as DeepWalk, node2vec or GraphSAGE can be used. Taking node2vec as an example, the algorithm generates a node sequence by performing biased random walks on the graph, and then uses a method similar to Word2Vec to learn the vector representation of the node. When processing the researcher cooperation network, each researcher is regarded as a node, and the cooperation relationship is represented as an edge between nodes. By setting appropriate walking strategies (such as balancing depth-first and breadth-first search), node2vec can capture the local and global structural characteristics of the network. Finally, each researcher node is mapped to a low-dimensional vector, which constitutes a set of node representation vectors. After obtaining the node representation vectors, the next step is to perform attribute enhancement on the researcher nodes in the researcher cooperation network based on these vectors and the output of the research interest prediction model. This step aims to integrate the predicted interest information of researchers into the network structure. Specifically, we first need to use the previously trained research interest prediction model to generate a predicted interest vector for each researcher. Then, this predicted interest vector is concatenated or weightedly fused with the node representation vector to form an enhanced node representation. This fusion not only retains the original network structure information, but also introduces dynamic interest prediction information, making the node representation richer and more multidimensional. The fusion method can be simple vector concatenation or a more complex attention mechanism to dynamically adjust the weights according to the importance of different features. Next, the graph attention network (GAT) is applied to the obtained enhanced node representation for multi-level information propagation and aggregation. The core idea of ​​GAT is to adaptively assign different importance weights to neighboring nodes through the self-attention mechanism, thereby achieving more refined information aggregation. In the implementation process, a multi-layer GAT structure is first constructed, and each layer contains multiple attention heads. Each attention head independently calculates the attention coefficient between node pairs, and then weighted sums the features of neighboring nodes according to these coefficients. By stacking multiple layers of GAT, multi-hop information propagation is achieved, so that each node can not only obtain the information of direct neighbors, but also capture the influence of more distant nodes. This process is dynamic because the attention weights are calculated in real time based on the features of the current node and neighboring nodes, so they can adapt to changes in network structure and node attributes. After GAT processing, each node obtains a dynamically updated feature representation, which not only contains the node's own information, but also incorporates its structural position and neighbor information in the network. Based on the dynamically updated node features obtained, the next step is to use the community detection algorithm to analyze the researcher collaboration network to obtain the network structure characteristics.The purpose of community detection is to identify closely connected groups of nodes in the network, which represent different research teams or academic circles in the research collaboration network. Common community detection algorithms include Louvain method, label propagation algorithm or spectral clustering. Taking Louvain method as an example, it discovers community structure by optimizing modularity. In specific implementation, the algorithm first regards each node as an independent community, and then iteratively moves the node to the neighboring community that can maximize the overall modularity. This process is repeated until the modularity no longer increases significantly. Through community detection, network structure features such as the number of communities, community size distribution, and community affiliation of nodes can be obtained, which reflect the organizational structure and grouping pattern of the researcher collaboration network. Finally, based on the obtained network structure features and node features, link prediction is performed through graph neural network to obtain the prediction results of potential collaboration relationships. Link prediction aims to predict edges that have not yet been formed in the network but may appear in the future. In the research collaboration network, this represents potential new collaboration relationships. In the implementation process, graph convolutional network (GCN) or graph attention network (GAT) can be used as the basic model. First, the network structure features (such as node degree, centrality index, community affiliation, etc.) and the previously obtained dynamically updated node features are combined as the input features of each node. Then, a multi-layer graph neural network is constructed, and each layer aggregates and transforms the node features. In the last layer of the model, a link prediction head is used, which accepts the final representation of two nodes as input and outputs the probability of forming an edge between the two nodes. The training of the model can adopt negative sampling technology, that is, randomly selecting some non-existent edges as negative samples, and using them together with known cooperative relationships (positive samples) for training. By minimizing the binary cross entropy loss function, the model learns how to distinguish between existing and non-existent edges. After the training is completed, all unconnected node pairs in the network are predicted to obtain the probability of them forming a cooperative relationship in the future, thereby obtaining the prediction results of potential cooperative relationships.

[0072] 105. Based on the project matching scores, network structure characteristics and potential cooperation relationship prediction results, knowledge mining is carried out on the research project data of multi-source heterogeneous scientific research data to obtain mining results, and the mining results are comprehensively processed and knowledge represented to obtain a personalized project recommendation list and explainable recommendation basis.

[0073] In one embodiment of the present invention, the research project data of the multi-source heterogeneous scientific research data is subjected to knowledge mining according to the project matching scores, network structure characteristics and potential cooperation relationship prediction results to obtain mining results, and the mining results are subjected to comprehensive processing and knowledge representation to obtain a personalized project recommendation list and an explainable recommendation basis, including: multimodal feature fusion of the project matching scores, network structure characteristics and potential cooperation relationship prediction results to obtain a comprehensive feature representation; according to the comprehensive feature representation, semantic analysis and topic modeling are performed on the research project data of the multi-source heterogeneous scientific research data to obtain a project semantic network; based on the project semantic network and the comprehensive feature representation, a knowledge graph construction algorithm is used to mine and infer the associations between projects to obtain a scientific research project knowledge graph; using a graph neural network and an attention mechanism, multi-dimensional correlation calculation and sorting are performed on the scientific research project knowledge graph to obtain a personalized project recommendation list; for each project in the personalized project recommendation list, relationship path analysis and feature importance quantification are performed based on the scientific research project knowledge graph to obtain an explainable recommendation basis.

[0074] Specifically, first, multimodal feature fusion is performed on the project matching score, network structure features, and potential cooperation relationship prediction results. This step aims to integrate features from different sources and forms into a unified representation. In specific implementation, feature splicing or attention mechanism can be used for fusion. The feature splicing method directly connects different feature vectors together to form a high-dimensional vector. The attention mechanism can dynamically adjust the importance weights of different features according to the needs of the current task. For example, using a multi-head attention mechanism, an attention head is assigned to each feature, and then the weighted sum of different features is performed through the learned weights. This method can adaptively adjust the contribution of different features to obtain a more refined and information-rich comprehensive feature representation. After obtaining the comprehensive feature representation, the next step is to perform semantic analysis and topic modeling on the research project data in multi-source heterogeneous scientific research data based on this representation. The purpose of this step is to deeply understand the content and topic structure of the research project. First, the project description text needs to be preprocessed, including word segmentation, stop word removal, word form restoration, etc. Then, word embedding technology (such as Word2Vec or BERT) can be used to convert each word into a dense vector representation. Next, a topic modeling algorithm is applied, such as Latent Dirichlet Allocation (LDA) or its variants. LDA assumes that each document is a mixture of multiple topics, and each topic is a probability distribution over words. Through iterative optimization, LDA can discover potential topic structures from a large number of documents. When processing research project data, each project description is regarded as a document, and LDA can be used to obtain the topic distribution of each project and the keyword distribution of each topic. Based on these results, a project semantic network can be constructed, in which nodes represent projects or topics, and edges represent the strength of semantic associations between them. Next, based on the project semantic network and comprehensive feature representation, the knowledge graph construction algorithm is used to mine and reason about the associations between projects to obtain the scientific research project knowledge graph. The knowledge graph construction process includes several key steps: entity recognition, relationship extraction, and knowledge fusion. First, key entities such as research topics, methods, and application fields are identified from project descriptions using named entity recognition (NER) technology. Then, relationship extraction algorithms such as remote supervised learning or neural network-based methods are used to identify semantic relationships between entities from the text. In addition, the comprehensive feature representation obtained previously can be used to infer potential associations between projects through feature similarity calculations. In the knowledge fusion stage, it is necessary to solve the problems of entity alignment and relationship alignment and eliminate duplicate and contradictory information. Finally, the knowledge graph is expanded through reasoning mechanisms, such as using transitive rules to infer new relationships. The constructed scientific research project knowledge graph not only contains the direct relationship between projects, but also contains rich semantic information and implicit knowledge structure. With the scientific research project knowledge graph, the next step is to use graph neural networks and attention mechanisms to calculate and sort its multi-dimensional correlations to obtain a personalized project recommendation list.The core of this step is to design a deep learning model that can reason on the knowledge graph. The graph attention network (GAT) can be used as the basic model because it can adaptively assign importance weights to different neighbor nodes. First, each node in the knowledge graph is initialized as a feature vector, which can be a combination of the node's attribute information and the previously obtained comprehensive feature representation. Then, a multi-layer GAT structure is constructed, with each layer containing multiple attention heads. In each layer, each node updates its own representation by aggregating the information of neighbor nodes. The attention mechanism allows the model to dynamically adjust the importance of different relationships and neighbors according to the current task. After multiple layers of propagation, each project node obtains a representation containing rich contextual information. Finally, the similarity is calculated based on the user's research interest (which can be the final representation of the user node) and the representation of the project node to obtain a recommendation score. The projects are sorted according to these scores to obtain a personalized project recommendation list. Finally, for each project in the personalized project recommendation list, the relationship path analysis and feature importance quantification are performed based on the scientific research project knowledge graph to obtain an explainable recommendation basis. This step aims to provide an intuitive and understandable explanation for the recommendation results. In specific implementation, explanations can be generated by finding the shortest or most meaningful path from the user node to the recommended item node in the knowledge graph. At the same time, attention weights or technologies like SHAP (SHapley Additive exPlanations) are used to quantify the contribution of different features to the recommendation results. These explanatory information are finally integrated into easy-to-understand text or visualization forms and presented to users as the basis for recommendation.

[0075] Furthermore, for each item in the personalized project recommendation list, relationship path analysis and feature importance quantification are performed based on the scientific research project knowledge graph to obtain an explainable recommendation basis, including: performing a bidirectional breadth-first search on the nodes related to the recommended projects in the scientific research project knowledge graph to obtain a multi-level relationship path set; performing semantic embedding processing on the nodes and edges of the paths in the relationship path set to obtain a path representation vector; applying an attention mechanism and a path pruning algorithm to the path representation vector to obtain a core explanation path, and based on the core explanation path and the comprehensive feature representation, quantifying the feature contribution to obtain a feature importance ranking; based on the core explanation path and the feature importance ranking, a structured explanation template is constructed through natural language generation technology to obtain an explainable recommendation basis.

[0076] Specifically, we first start with a bidirectional breadth-first search of the nodes related to the recommended projects in the scientific research project knowledge graph. This step aims to explore the multi-level relationships between the recommended projects and other entities. In the specific implementation, we start with the recommended project node as the starting point, and perform breadth-first search forward (incoming edge) and backward (outgoing edge) at the same time. During the search process, it is necessary to record the level and path information of each visited node. In order to control the breadth and depth of the search, the maximum search depth and the maximum number of expanded nodes per layer can be set. This bidirectional search strategy can fully capture the contextual information of the project in the knowledge graph, including its preconditions, related topics, potential applications, etc. The search results form a multi-level set of relationship paths, each of which represents a semantic association between the recommended project and other entities. After obtaining the set of relationship paths, the next step is to perform semantic embedding processing on the nodes and edges in these paths to obtain the vector representation of the path. The purpose of this step is to convert the discrete graph structure information into a continuous vector space representation, which is convenient for subsequent similarity calculation and machine learning processing. In the implementation process, it is first necessary to pre-train the embedding vectors for the nodes and edges in the knowledge graph. For nodes, knowledge graph embedding methods such as TransE and RotatE can be used; for edges, specific vector representations can be assigned according to the type of edge. Then, for each relationship path, the embedding vectors of the nodes and edges in it are concatenated in sequence order to form a preliminary path representation. In order to capture the overall semantics of the path, a sequence model such as LSTM (Long Short-Term Memory Network) or Transformer can be used to further encode this preliminary representation to obtain the final path representation vector. This method not only retains the semantic information of each element in the path, but also captures the sequential relationship and long-term dependency between elements. Next, the attention mechanism and path pruning algorithm are applied to the obtained path representation vector to screen out the most representative core explanation path. The application of the attention mechanism allows the model to automatically learn the importance weights of different paths. In specific implementation, a multi-head self-attention mechanism can be used, each attention head calculates the correlation between paths, and then the results of multiple heads are merged. In this way, the model can capture the complex interactions between paths. At the same time, in order to reduce the computational complexity and improve the simplicity of the explanation, a path pruning algorithm needs to be applied. The pruning process can retain paths with higher weights based on the attention weights, while considering the diversity of paths to ensure that different types of explanations can be retained. The core explanatory path set obtained after pruning is highly correlated and can fully reflect the different aspects of the recommendation reasons. After obtaining the core explanatory paths, the next step is to quantify the feature contributions based on these paths and the comprehensive feature representation obtained previously, and obtain the feature importance ranking. This step aims to clarify the degree of influence of each feature on the recommendation results. In the implementation process, the SHAP (SHapleyAdditive exPlanations) value calculation method can be used.SHAP value is based on the concept of Shapley value in game theory, which can fairly distribute the contribution of each feature to the prediction result. Specifically, it is necessary to build an explanation model with the core explanation path and comprehensive feature representation as input and the recommendation score as output. Then, by calculating the SHAP value of each feature, its contribution to the recommendation result is obtained. These contributions can be normalized and sorted to form a feature importance ranking. This method not only considers the influence of a single feature, but also captures the interaction between features, providing a more comprehensive and accurate explanation. Finally, according to the core explanation path and feature importance ranking, a structured explanation template is constructed through natural language generation technology to obtain an explainable recommendation basis. The goal of this step is to convert the mathematical representation obtained in the previous step into a natural language description that is easy for users to understand. In the implementation process, first, a series of explanation templates need to be designed, which should be able to cover different types of recommendation reasons, such as similarity, complementarity, novelty, etc. Then, based on the semantic information of the core explanation path and the feature importance ranking, the most suitable template is selected and filled with relevant content. In the filling process, a pre-trained language model can be used to generate fluent and natural text. In order to enhance the personalization of explanations, the language style and professionalism can be adjusted according to the user's background knowledge and preferences. In addition, you can also consider adding visualization elements, such as drawing the core explanation path into a simplified knowledge graph to intuitively show the relationship between the recommended items and user interests. In this way, the final generated recommendation basis not only accurately reflects the decision logic of the recommendation system, but also can be presented in a user-friendly way, which helps to enhance the user's understanding and trust in the recommendation results.

[0077] In this embodiment, data preprocessing is performed on multi-source heterogeneous scientific research data to generate a data set in a unified format, and a multi-dimensional scientific research portrait feature set is mined from it. Based on these feature sets, the academic output data of researchers is pattern mined to obtain research interest pattern sets and research trend indicators. Through predictive mining, a research interest prediction model and project matching score are generated for historical research data, and the model is used to perform graph structure mining on the researcher cooperation network to predict potential cooperation relationships. Finally, based on the project matching score, network structure characteristics and cooperation relationship prediction results, knowledge mining is performed on the scientific research project data to generate a personalized project recommendation list and its explainable recommendation basis. This method can comprehensively consider the evolution of researchers' interests, project characteristics, and cooperation potential, and improve the accuracy and interpretability of recommendations.

[0078] The above describes the data mining method in the embodiment of the present invention. The following describes the data mining device in the embodiment of the present invention. Figure 2 , an embodiment of the data mining device in the embodiment of the present invention includes:

[0079] The feature mining module 201 is used to perform data mining preprocessing on multi-source heterogeneous scientific research data to obtain a preprocessed unified format data set, and perform feature mining on the unified format data set to obtain a multi-dimensional scientific research portrait feature set;

[0080] A pattern mining module 202 is used to perform pattern mining on the research activities of researchers based on the academic output data in the multi-dimensional scientific research portrait feature set to obtain a research interest pattern set and research trend indicators;

[0081] A predictive mining module 203 is used to perform predictive mining on the historical research data in the multi-dimensional scientific research portrait feature set according to the research interest pattern set and the research trend index, to obtain a research interest prediction model and a project matching score;

[0082] A graph structure mining module 204 is used to perform graph structure mining on the researcher cooperation network in the multi-dimensional scientific research portrait feature set according to the research interest prediction model to obtain network structure features and potential cooperation relationship prediction results;

[0083] The knowledge mining module 205 is used to perform knowledge mining on the research project data of the multi-source heterogeneous scientific research data according to the project matching scores, network structure characteristics and potential cooperation relationship prediction results, obtain mining results, and perform comprehensive processing and knowledge representation on the mining results to obtain a personalized project recommendation list and explainable recommendation basis.

[0084] In an embodiment of the present invention, the data mining device runs the above-mentioned data mining method, and the data mining device generates a data set in a unified format by performing data preprocessing on multi-source heterogeneous scientific research data, and mines a multi-dimensional scientific research portrait feature set therefrom. Based on these feature sets, pattern mining is performed on the academic output data of researchers to obtain research interest pattern sets and research trend indicators. Through predictive mining, a research interest prediction model and project matching score are generated for historical research data, and the model is used to perform graph structure mining on the researcher cooperation network to predict potential cooperation relationships. Finally, based on the project matching score, network structure characteristics and cooperation relationship prediction results, knowledge mining is performed on the scientific research project data to generate a personalized project recommendation list and its explainable recommendation basis. This method can comprehensively consider the evolution of researchers' interests, project characteristics, and cooperation potential, and improve the accuracy and interpretability of recommendations.

[0085] above Figure 2 The data mining apparatus in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The data mining device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0086] Figure 33 is a structural diagram of a data mining device provided by an embodiment of the present invention. The data mining device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the data mining device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the data mining device 300 to implement the steps of the above-mentioned data mining method.

[0087] The data mining device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The data mining device structure shown does not constitute a limitation on the data mining device provided by the present invention, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0088] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the data mining method.

[0089] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0090] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0091] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data mining method, characterized in that: The data mining method comprises: Perform data mining preprocessing on multi-source heterogeneous scientific research data to obtain a preprocessed unified format data set, and perform feature mining on the unified format data set to obtain a multi-dimensional scientific research portrait feature set; Based on the academic output data in the multi-dimensional scientific research portrait feature set, pattern mining is performed on the research activities of researchers to obtain a research interest pattern set and research trend indicators; According to the research interest pattern set and research trend indicators, predictive mining is performed on the historical research data in the multi-dimensional scientific research portrait feature set to obtain a research interest prediction model and a project matching score; According to the research interest prediction model, the graph structure mining is performed on the researcher cooperation network in the multi-dimensional scientific research portrait feature set to obtain the network structure characteristics and potential cooperation relationship prediction results; According to the project matching scores, network structure characteristics and potential cooperation relationship prediction results, knowledge mining is performed on the research project data of the multi-source heterogeneous scientific research data to obtain mining results, and the mining results are comprehensively processed and knowledge represented to obtain a personalized project recommendation list. For each project in the personalized project recommendation list, relationship path analysis and feature importance quantification are performed based on the scientific research project knowledge graph to obtain an explainable recommendation basis, including: performing a bidirectional breadth-first search on the nodes related to the recommended projects in the scientific research project knowledge graph to obtain a multi-level relationship path set; semantic embedding of the nodes and edges of the paths in the relationship path set Input processing to obtain a path representation vector; apply the attention mechanism and path pruning algorithm to the path representation vector to obtain the core explanation path, and based on the core explanation path and the comprehensive feature representation, quantify the feature contribution to obtain the feature importance ranking; according to the core explanation path and the feature importance ranking, a structured explanation template is constructed through natural language generation technology to obtain an explainable recommendation basis, wherein the scientific research project knowledge graph is obtained based on mining and reasoning the correlation relationship between projects, and the comprehensive feature representation is obtained based on multimodal feature fusion of the project matching score, network structure characteristics and potential cooperation relationship prediction results.

2. The data mining method according to claim 1, characterized in that: According to the academic output data in the multi-dimensional scientific research portrait feature set, the research activities of researchers are pattern mined to obtain the research interest pattern set and research trend indicators including: Performing topic modeling and keyword extraction on the academic output data in the multi-dimensional scientific research portrait feature set to obtain research topic distribution and keyword set; According to the research topic distribution and keyword set, a time series cluster analysis is performed on the academic achievements of researchers in the academic output data to obtain the evolution trajectory of research interests; Based on the evolution trajectory of the research interests, a frequent pattern mining algorithm is used to analyze the research behavior sequence in the academic output data to obtain a research interest pattern set; The trend prediction algorithm and anomaly detection method are applied to the research interest pattern set, combined with preset external discipline development data, to obtain research trend indicators.

3. The data mining method according to claim 2, characterized in that: Based on the evolution trajectory of the research interest, the frequent pattern mining algorithm is used to analyze the research behavior sequence in the academic output data, and the research interest pattern set obtained includes: Segmenting the evolution trajectory of the research interest into time windows to obtain multiple time series segments; According to the time series segments, the research behavior sequences in the academic output data are encoded and serialized to obtain a standardized behavior sequence set; Applying a preset sequence pattern mining algorithm to the behavior sequence set to extract frequent subsequence patterns and obtain candidate research interest patterns; In combination with the research topic distribution, a hierarchical clustering algorithm is used to classify and merge the candidate research interest patterns to obtain a research interest pattern set.

4. The data mining method according to claim 1, characterized in that: The predictive mining of historical research data in the multi-dimensional scientific research portrait feature set based on the research interest pattern set and research trend indicators to obtain a research interest prediction model and project matching score includes: Performing feature fusion on the research interest pattern set and the research trend index, and combining the historical research data in the multi-dimensional scientific research portrait feature set to obtain a comprehensive feature vector; Based on the comprehensive feature vector, the time series decomposition technique is used to separate the trend, cycle and random components of the historical research data to obtain multidimensional time series components; Applying a deep learning model to the multidimensional time series component to generate a research interest prediction model, and performing interest prediction for a preset future time point based on the research interest prediction model and the research activity data in the multidimensional scientific research portrait feature set to obtain a predicted interest vector; According to the predicted interest vector, combined with the preset project feature vector, the project matching score is obtained through cosine similarity calculation and multi-factor weighting.

5. The data mining method according to claim 1, characterized in that: According to the research interest prediction model, the graph structure mining is performed on the researcher cooperation network in the multi-dimensional scientific research portrait feature set to obtain the network structure characteristics and potential cooperation relationship prediction results including: Performing graph embedding processing on the researcher cooperation network in the multi-dimensional scientific research portrait feature set to obtain a low-dimensional node representation vector; According to the node representation vector and the output of the research interest prediction model, the attributes of the researcher nodes in the researcher cooperation network are enhanced to obtain an enhanced node representation that incorporates interest information; Applying a graph attention network to the enhanced node representation to perform multi-level information propagation and aggregation to obtain dynamically updated node features; According to the node characteristics, the community detection algorithm is used to analyze the researcher cooperation network to obtain the network structure characteristics; Based on the network structure characteristics and the node characteristics, link prediction is performed through a graph neural network to obtain a potential cooperation relationship prediction result.

6. The data mining method according to claim 1, characterized in that: According to the project matching score, network structure characteristics and potential cooperation relationship prediction results, the research project data of the multi-source heterogeneous scientific research data is subjected to knowledge mining to obtain mining results, and the mining results are subjected to comprehensive processing and knowledge representation to obtain a personalized project recommendation list and an explainable recommendation basis, including: Performing multimodal feature fusion on the project matching scores, network structure characteristics and potential cooperative relationship prediction results to obtain a comprehensive feature representation; According to the comprehensive feature representation, semantic analysis and topic modeling are performed on the research project data of the multi-source heterogeneous scientific research data to obtain a project semantic network; Based on the semantic network and comprehensive feature representation of the project, a knowledge graph construction algorithm is used to mine and infer the associations between projects to obtain a scientific research project knowledge graph; Using graph neural networks and attention mechanisms, multi-dimensional correlation calculation and sorting are performed on the scientific research project knowledge graph to obtain a personalized project recommendation list; For each item in the personalized project recommendation list, relationship path analysis and feature importance quantification are performed based on the scientific research project knowledge graph to obtain an explainable recommendation basis.

7. A data mining device, characterized in that: The data mining device comprises: A feature mining module is used to perform data mining preprocessing on multi-source heterogeneous scientific research data to obtain a preprocessed unified format data set, and perform feature mining on the unified format data set to obtain a multi-dimensional scientific research portrait feature set; A pattern mining module is used to perform pattern mining on the research activities of researchers based on the academic output data in the multi-dimensional scientific research portrait feature set to obtain a research interest pattern set and research trend indicators; A predictive mining module, for performing predictive mining on the historical research data in the multi-dimensional scientific research portrait feature set according to the research interest pattern set and the research trend index, to obtain a research interest prediction model and a project matching score; A graph structure mining module is used to perform graph structure mining on the researcher cooperation network in the multi-dimensional scientific research portrait feature set according to the research interest prediction model, and obtain network structure characteristics and potential cooperation relationship prediction results; The knowledge mining module is used to perform knowledge mining on the research project data of the multi-source heterogeneous scientific research data according to the project matching score, network structure characteristics and potential cooperation relationship prediction results, obtain mining results, and perform comprehensive processing and knowledge representation on the mining results to obtain a personalized project recommendation list. For each project in the personalized project recommendation list, relationship path analysis and feature importance quantification are performed based on the scientific research project knowledge graph to obtain an explainable recommendation basis, including: performing a bidirectional breadth-first search on the nodes related to the recommended projects in the scientific research project knowledge graph to obtain a multi-level relationship path set; and performing a search on the nodes and edges of the paths in the relationship path set. Perform semantic embedding processing to obtain a path representation vector; apply an attention mechanism and a path pruning algorithm to the path representation vector to obtain a core explanation path, and based on the core explanation path and the comprehensive feature representation, quantify the feature contribution to obtain a feature importance ranking; based on the core explanation path and the feature importance ranking, a structured explanation template is constructed through natural language generation technology to obtain an explainable recommendation basis, wherein the scientific research project knowledge graph is obtained based on mining and reasoning about the correlation between projects, and the comprehensive feature representation is obtained based on multimodal feature fusion of the project matching score, network structure characteristics and potential cooperation relationship prediction results.

8. A data mining device, characterized in that: The data mining device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the data mining device to perform the steps of the data mining method according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by a processor, the steps of the data mining method according to any one of claims 1 to 6 are implemented.

Images