Research direction recommendation method and related equipment based on student portraits

Abstract

The present invention relates to the field of computer technology, and discloses a method for recommending scientific research directions based on student portraits and related equipment. The method includes: obtaining multi-source heterogeneous data of students to be recommended, and pre-processing the multi-source heterogeneous data, extracting and fusing multimodal features of the pre-processed data to obtain a comprehensive feature representation, and dynamically modeling the student portraits of the students to be recommended to obtain student dynamic portraits; extracting graph relationships from the acquired scientific research literature data and scientific research project data to obtain a scientific research direction knowledge graph; cross-connecting the student dynamic portraits and the scientific research direction knowledge graph to obtain a recommendation candidate set, and performing multi-dimensional recommendation scoring and sorting on the recommendation candidate set to obtain a scientific research direction recommendation list; interpreting, generating, and dynamically adjusting the scientific research direction recommendation list to generate a recommendation plan. This application fully combines student information with relevant scientific research information to improve the accuracy of the scientific research directions recommended to students.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method for recommending scientific research directions based on student portraits and related equipment. Background Art

[0002] With the rapid development of higher education and the continuous expansion of scientific research, providing students with personalized research guidance has become increasingly important. Traditional research recommendation methods often rely on the instructor's experience and the student's subjective interests, often failing to fully consider multiple factors such as the student's academic background, research ability, and development potential. Furthermore, the rapid changes and cross-fertilization of scientific research fields pose challenges in accurately grasping research frontiers and development trends.

[0003] Currently, several data analysis-based methods have been applied to research direction recommendations. For example, some studies have utilized text mining techniques to analyze academic literature and extract hot research topics; others have attempted to use collaborative filtering algorithms to make recommendations based on the research interests of similar students. However, these methods often suffer from the following problems: a single data source makes it difficult to fully characterize students' academic characteristics and research potential; they overlook the integration of students' personal development trajectories with research trends; they fail to fully consider the potential for interdisciplinary research; and they tend to limit recommendations to established research paradigms. Consequently, the recommended research directions for students are inaccurate. In other words, existing research direction recommendation methods fail to effectively integrate student profiles with relevant research information, resulting in low accuracy in recommended research directions. Summary of the Invention

[0004] The main purpose of the present invention is to solve the problem that the existing method for recommending scientific research directions to students cannot well combine student information with relevant scientific research information, resulting in low accuracy of the scientific research directions recommended to students.

[0005] The first aspect of the present invention provides a method for recommending scientific research directions based on student portraits, and the method for recommending scientific research directions based on student portraits includes: obtaining multi-source heterogeneous data of students to be recommended, and preprocessing the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data; extracting and fusing multimodal features of the preprocessed multi-source heterogeneous data to obtain a comprehensive feature representation, and based on the comprehensive feature representation, dynamically modeling the student portrait of the student to be recommended to obtain a student dynamic portrait; obtaining scientific research literature data and scientific research project data to be recommended, and extracting graph relationships between the scientific research literature data and the scientific research project data to obtain a scientific research direction knowledge graph; cross-connecting the student dynamic portrait and the scientific research direction knowledge graph to obtain a recommendation candidate set, and performing multi-dimensional recommendation scoring and sorting on the recommendation candidate set to obtain a scientific research direction recommendation list; interpreting, generating and dynamically adjusting the scientific research direction recommendation list to generate a final scientific research direction recommendation plan.

[0006] Optionally, in a first implementation method of the first aspect of the present invention, the multi-source heterogeneous data is preprocessed to obtain preprocessed multi-source heterogeneous data, including: performing regular expression matching and outlier marking on text fields in the multi-source heterogeneous data to obtain outlier marking data, and performing median replacement and multiple interpolation on the marked outliers in the outlier marking data to obtain complete multi-source data; performing multi-dimensional feature extraction and cluster analysis on the corresponding student information in the complete multi-source data to obtain a student unique identifier, and based on the student unique identifier and a preset log record time sequence, performing structured extraction of scientific research results information and annotation of academic fields on the complete multi-source data to obtain preprocessed multi-source heterogeneous data.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the multimodal features of the preprocessed multi-source heterogeneous data are extracted and fused to obtain a comprehensive feature representation, including: performing word segmentation on the student text data in the preprocessed multi-source heterogeneous data to obtain a word vector sequence, and performing time series analysis on the student numerical data in the preprocessed multi-source heterogeneous data to obtain learning performance trend features, and performing structured extraction on the scientific research activity data in the preprocessed multi-source heterogeneous data to obtain scientific research experience features; performing vectorized mapping on the word vector sequence to obtain a word vector set of student-related features, and clustering the word vector set by interest topics to obtain a student interest topic distribution; integrating the student interest topic distribution, the learning performance trend features and the scientific research experience features to obtain a multimodal feature matrix, and normalizing and reducing the dimensionality of the multimodal feature matrix to obtain a comprehensive feature representation of the students to be recommended.

[0008] Optionally, in a third implementation method of the first aspect of the present invention, the dynamic modeling of the student portrait to be recommended is performed based on the comprehensive feature representation to obtain a dynamic student portrait, including: constructing a student scientific research portrait framework of the student to be recommended under multiple preset key portrait dimensions based on the comprehensive feature representation, and performing a cluster analysis on the academic interest characteristics in the student scientific research portrait framework to obtain the research topic preference distribution of the student to be recommended, and performing a quantitative evaluation of the research ability characteristics in the student scientific research portrait framework to obtain the research ability score of the student to be recommended in different research methods and skills, and performing a predictive analysis on the scientific research potential characteristics in the student scientific research portrait framework to obtain the development potential index of the student to be recommended in different research directions; performing correlation calculation on the research topic preference distribution, the research ability score and the development potential index to obtain a correlation coefficient matrix, and constructing an internal association network corresponding to the student scientific research characteristics based on the correlation coefficient matrix; performing time series decomposition on the characteristic values ​​of each node in the internal association network to obtain the time series components of the student scientific research characteristics, and performing weighted combination on the time series components based on the historical academic data in the preprocessed multi-source heterogeneous data to obtain a dynamic student portrait.

[0009] Optionally, in a fourth implementation method of the first aspect of the present invention, the graph relationship extraction of the scientific research literature data and the scientific research project data is performed to obtain a scientific research direction knowledge graph, including: extracting a variety of key scientific research concepts from the scientific research literature data and the scientific research project data to obtain a core entity set in the scientific research field, and extracting keywords and subject descriptions of scientific research entities in the scientific research project data to obtain display relationship types between scientific research entities, and performing association strength calculation and subject clustering on the core entity set to obtain preliminary scientific research topics; performing semantic similarity calculation on the scientific research entities in each scientific research topic group in the preliminary scientific research topics to obtain sub-topic division results, and performing standardized mapping and division of topics on the preliminary scientific research topics and the sub-topic division results. The balance of the results is adjusted to obtain a scientific research direction classification system; the entity co-occurrence in the scientific research direction classification system is counted and the point mutual information value between entities is calculated to obtain the entity semantic association, and based on the entity semantic association and the display relationship type, a preliminary association graph between scientific research entities is constructed; the entity importance score of each scientific research entity in the preliminary association graph is calculated, and based on the entity importance score and the entity semantic association, the preliminary association graph is adjusted to obtain a semantic association network, and the semantic association network is structured to obtain a preliminary scientific research direction knowledge graph; the graph nodes and edges in the preliminary scientific research direction knowledge graph are vectorized and searched for across domains of meta-paths to obtain a final scientific research direction knowledge graph.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, the student dynamic portrait and the scientific research direction knowledge graph are cross-connected to obtain a recommendation candidate set, including: multi-dimensional decomposition of the academic interest characteristics, research ability characteristics and development potential index in the student dynamic portrait to obtain a student research feature tensor, and using the student research feature tensor to project the scientific research direction knowledge graph to obtain a personalized weighted knowledge graph; using a preset topic random walk model, performing random walk extraction of high-frequency research topics on the personalized weighted knowledge graph to obtain a research topic sequence corresponding to the student to be recommended, and performing topic clustering on the research topic sequence to obtain a potential research direction set; based on the potential research direction set, expanding the local network of the personalized weighted knowledge graph and identifying related research sub-fields to obtain candidate research communities, and performing time series correlation analysis on the candidate research communities based on the historical research trajectory corresponding to the student to be recommended to obtain a development path map; based on the development path map, predicting the potential breakthrough points and multi-research objective optimization of the student to be recommended to obtain a recommendation candidate set.

[0011] Optionally, in a sixth implementation method of the first aspect of the present invention, the recommendation candidate set is subjected to multi-dimensional recommendation scoring and sorting to obtain a list of recommended scientific research directions, including: constructing a set of scoring dimensions for the students to be recommended based on the preprocessed multi-source heterogeneous data, and performing multi-dimensional quantitative evaluation on each scientific research direction in the recommendation candidate set to obtain an original scoring matrix, and performing hierarchical clustering and hierarchical tree node importance quantification on the scoring dimension set to obtain an initial weight vector; performing time series analysis on the students' recent research behavior data in the preprocessed multi-source heterogeneous data to obtain an interest drift model, and fusing the initial weight vector and the interest drift model to obtain a dynamic weight function; performing time series decomposition and periodic pattern convolution operations on the original scoring matrix and the dynamic weight function to obtain a time-varying weighted scoring matrix, and performing non-dominated sorting and hierarchical analysis on the time-varying weighted scoring matrix to obtain a comprehensive ranking score; normalizing the comprehensive ranking score and constructing an adjacency matrix to obtain a state transition matrix, and performing steady-state distribution calculation on the state transition matrix to obtain a list of recommended scientific research directions.

[0012] The second aspect of the present invention provides a scientific research direction recommendation device based on student portraits, and the scientific research direction recommendation device based on student portraits includes: a preprocessing module for obtaining multi-source heterogeneous data of students to be recommended, and preprocessing the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data; a feature representation module for extracting and fusing multimodal features of the preprocessed multi-source heterogeneous data to obtain a comprehensive feature representation, and based on the comprehensive feature representation, dynamically modeling the student portrait of the student to be recommended to obtain a student dynamic portrait; a relationship extraction module for obtaining scientific research literature data and scientific research project data to be recommended, and performing graph relationship extraction on the scientific research literature data and the scientific research project data to obtain a scientific research direction knowledge graph; a cross-connection module for cross-connecting the student dynamic portrait and the scientific research direction knowledge graph to obtain a recommendation candidate set, and performing multi-dimensional recommendation scoring and sorting on the recommendation candidate set to obtain a scientific research direction recommendation list; a dynamic adjustment module for interpreting, generating and dynamically adjusting the scientific research direction recommendation list to generate a final scientific research direction recommendation plan.

[0013] The third aspect of the present invention provides a device for recommending scientific research directions based on student portraits, comprising: 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 so that the device for recommending scientific research directions based on student portraits performs the various steps of the above-mentioned method for recommending scientific research directions based on student portraits.

[0014] The fourth aspect of the present invention provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, it enables the computer to execute the various steps of the above-mentioned method for recommending scientific research directions based on student portraits.

[0015] The above-mentioned method for recommending scientific research directions based on student portraits and related equipment. In an embodiment of the present invention, by obtaining multi-source heterogeneous data of the students to be recommended and pre-processing the multi-source heterogeneous data, the pre-processed multi-source heterogeneous data is obtained; multi-modal features are extracted and fused on the pre-processed multi-source heterogeneous data to obtain a comprehensive feature representation, and based on the comprehensive feature representation, dynamic modeling of the student portrait of the recommended student is performed to obtain a student dynamic portrait; scientific research literature data and scientific research project data to be recommended are obtained, and graph relationship extraction is performed on the scientific research literature data and scientific research project data to obtain a scientific research direction knowledge graph; the student dynamic portrait and the scientific research direction knowledge graph are cross-connected to obtain a recommendation candidate set, and the recommendation candidate set is scored and sorted in a multi-dimensional recommendation manner to obtain a scientific research direction recommendation list; the scientific research direction recommendation list is interpreted, generated, and dynamically adjusted to generate a final scientific research direction recommendation plan. Compared with the existing technology, this application obtains multi-source heterogeneous data of students to be recommended, extracts and fuses multimodal features of the multi-source heterogeneous data, and dynamically models student portraits to obtain student dynamic portraits, and extracts graph relationships of the obtained scientific research literature data and scientific research project data to be recommended to obtain a scientific research direction knowledge graph, and then cross-connects the student dynamic portraits and the scientific research direction knowledge graph to obtain a recommendation candidate set, and performs multi-dimensional recommendation scoring and sorting on the recommendation candidate set to obtain a scientific research direction recommendation list, thereby interpreting, generating and dynamically adjusting the recommendation list to generate the final recommendation plan for the students to be recommended, thereby fully combining student information with relevant scientific research information, and improving the accuracy of the scientific research directions recommended to students.

[0016] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the present invention. The purposes and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0017] 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

[0018] Figure 1 This is a schematic diagram of a first embodiment of a method for recommending research directions based on student portraits in an embodiment of the present invention;

[0019] Figure 2 Schematic diagram of an embodiment of a device for recommending research directions based on student portraits in an embodiment of the present invention;

[0020] Figure 3 This is a schematic diagram of an embodiment of a device for recommending scientific research directions based on student portraits in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0022] The terms "including," "having," and any variations thereof, as used in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to the process, method, product, or device.

[0023] To facilitate understanding of this embodiment, the specific process of the embodiment of the present invention is described below. Figure 1 The first embodiment of the method for recommending scientific research directions based on student portraits in the embodiment of the present invention includes:

[0024] 101. Acquire multi-source heterogeneous data of the students to be recommended, and pre-process the multi-source heterogeneous data to obtain pre-processed multi-source heterogeneous data;

[0025] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results.

[0026] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0027] In this embodiment, the multi-source heterogeneous data here refers to the personal information data of the students to be recommended, student scientific research activity data, student behavior data, scientific research literature data, scientific research project data, subject area knowledge data, and corresponding historical data, etc. Regular expression matching and outlier marking are performed on the text fields in the multi-source heterogeneous data to obtain outlier marked data, and median replacement and multiple interpolation are performed on the marked outliers in the outlier marked data to obtain complete multi-source data; multi-dimensional feature extraction and cluster analysis are performed on the corresponding student information in the complete multi-source data to obtain a student unique identifier. Based on the student unique identifier and the preset log record time sequence, the complete multi-source data is subjected to structured extraction of scientific research results information and academic field annotation to obtain pre-processed multi-source heterogeneous data.

[0028] In practical applications, relevant information about the recommended students is obtained from multiple data sources, which may include student management systems, scientific research project databases, academic literature databases, and online learning platforms. The types of data obtained cover student personal information, academic performance, scientific research activity records, published paper information, and corresponding historical data. Regular expression matching is then performed on text fields in the data to identify and extract information with specific patterns, such as email addresses, phone numbers, or identifiers in a specific format. At the same time, outlier detection and labeling are performed on the data (outliers caused by data entry errors, system failures, or real extreme situations are identified and labeled by setting reasonable thresholds or using statistical methods (such as Z-score or interquartile range method)). ; Then, for the marked outliers, median replacement is used to process the outliers of continuous variables and multiple imputation is used to create multiple possible fill-in values ​​to replace missing or abnormal data, thus obtaining a relatively complete and reasonable data set; then, multi-dimensional feature extraction is performed on these complete multi-source data, including text feature extraction (such as TF-IDF or word embedding technology), standardization or normalization of numerical features, unique-hot encoding of categorical features, etc., and cluster analysis is performed on the results of feature extraction to discover the inherent structure and pattern in the data, thereby generating a unique identifier for each student. This identifier is not just a simple ID, but can reflect the combination of features of the student in multiple dimensions, and is combined with the preset log record time series to further process the complete multi-source data. The log record time series here may refer to the timeline of the student's learning and scientific research activities; then, based on this time series information, structured extraction of the student's scientific research results information is performed, including extracting keywords from the paper title and abstract, identifying research methods, determining research topics, etc. At the same time, these scientific research results are labeled in academic fields using a pre-defined subject classification system, and finally the pre-processed multi-source heterogeneous data is obtained.

[0029] 102. Extract and fuse multimodal features from the preprocessed multi-source heterogeneous data to obtain a comprehensive feature representation. Based on the comprehensive feature representation, dynamically model the student profiles of the recommended students to obtain dynamic student profiles.

[0030] In this embodiment, the student text data in the preprocessed multi-source heterogeneous data is segmented to obtain a word vector sequence, and the student numerical data in the preprocessed multi-source heterogeneous data is time-series analyzed to obtain learning performance trend characteristics, and the scientific research activity data in the preprocessed multi-source heterogeneous data is structured extracted to obtain scientific research experience characteristics; the word vector sequence is vectorized to obtain a word vector set of student-related characteristics, and the word vector set is clustered by interest topics to obtain the student interest topic distribution; the student interest topic distribution, learning performance trend characteristics and scientific research experience characteristics are integrated to obtain a multimodal feature matrix, and the multimodal feature matrix is ​​normalized and dimensionally reduced to obtain a comprehensive feature representation of the students to be recommended. Based on the comprehensive feature representation, a student scientific research portrait framework is constructed under multiple preset key portrait dimensions for the recommended students, and a cluster analysis is performed on the academic interest characteristics in the student scientific research portrait framework to obtain the research topic preference distribution of the recommended students, and a quantitative evaluation is performed on the research ability characteristics in the student scientific research portrait framework to obtain the research ability scores of the recommended students in different research methods and skills, and a predictive analysis is performed on the scientific research potential characteristics in the student scientific research portrait framework to obtain the development potential index of the recommended students in different research directions; the research topic preference distribution, research ability score and development potential index are correlated to obtain the correlation coefficient matrix, and based on the correlation coefficient matrix, an internal association network corresponding to the student scientific research characteristics is constructed; the characteristic values ​​of each node in the internal association network are decomposed into time series to obtain the time series components of the student scientific research characteristics, and based on the historical academic data in the pre-processed multi-source heterogeneous data, the time series components are weighted combined to obtain a dynamic student portrait.

[0031] In practical applications, first, we use specialized word segmentation tools, such as the jieba word segmenter for Chinese and NLTK or spaCy libraries for English, to perform word segmentation on the student text data in the pre-processed multi-source heterogeneous data. In the word segmentation process, we need to consider not only general language rules but also professional terms and compound words in the academic field. We need to introduce custom dictionaries to improve the accuracy of word segmentation, thereby obtaining a word vector sequence. We also use the time series analysis technology of the ARIMA model to perform time series analysis on the students' numerical data (including indicators that change over time, such as academic performance, frequency of scientific research activities, number of published papers, etc.) to capture the long-term trends and short-term fluctuations in students' learning performance, and obtain information reflecting the speed of students' learning progress. The method uses Word2Vec technology to vectorize the word vector sequence, mapping each word to a high-dimensional vector space so that words with similar semantics are closer in this space, and obtains a word vector set of student-related features. The topic model algorithm is used to cluster the word vector set by interest topics to discover potential topics in the word vector set and generate a topic distribution for each student. This distribution reflects the students' interest levels in different research topics. Multimodal fusion technology is then used to integrate the above-mentioned student interest topic distributions, learning performance trend characteristics, and scientific research experience characteristics to organize these features into a multimodal feature matrix, in which each row represents a student and each column represents a feature dimension. The multimodal feature matrix is ​​normalized using Z-score standardization and the normalized data is reduced in dimensionality using principal component analysis, ultimately obtaining a comprehensive feature representation of the students to be recommended.

[0032] Next, based on the comprehensive feature representation, a student scientific research portrait framework is constructed for the recommended students under multiple preset key portrait dimensions (wherein these key dimensions include academic interests, research abilities, scientific research potential, etc.), and K-means is used to perform cluster analysis on the academic interest characteristics to classify similar research topics and obtain the students' research topic preference distribution (wherein this distribution not only reflects the students' current research interests, but may also reveal potential interdisciplinary research tendencies); and a quantitative evaluation is performed on the research ability characteristics in the student scientific research portrait framework (such as students' scientific research results, course grades, project experience, etc.) to obtain students' research ability scores in different research methods and skills, and a prediction model is constructed in advance based on factors such as students' historical performance, learning curves, and development trends of research topics to predict students' development potential in different research directions. The prediction model is used to perform predictive analysis on the scientific research potential characteristics in the student scientific research portrait framework to predict the student's development potential index to reflect the student's long-term development prospects in various research directions; and then the research topic preference distribution, research ability, and research potential are calculated by calculating the Pearson correlation coefficient. The correlation between the ability score and the development potential index is calculated to obtain a correlation coefficient matrix (reflecting the strength of the relationship between each feature). Based on this correlation coefficient matrix, an internal correlation network corresponding to the student's scientific research characteristics is constructed, where this network is represented by a graph structure, in which the nodes represent each feature and the edge weights represent the strength of the correlation between the features. Then, the moving average decomposition is used to perform time series decomposition on the characteristic values ​​of each node in this internal correlation network, that is, by decomposing the time series into trend, seasonal and random components, in order to identify the long-term trend, cyclical changes and short-term fluctuations of the student's scientific research characteristics (that is, the change pattern of the student's scientific research characteristics over time), and the historical academic data in the pre-processed multi-source heterogeneous data is used to perform weighted combination of the time series components. The initial weights are set by using expert knowledge, and then these weights are optimized by machine learning algorithms (such as gradient descent or genetic algorithms). For example, the most recent data may have a higher weight than the older data. Finally, a dynamic student portrait is obtained, which not only includes the student's current scientific research characteristics, but also incorporates information in the time dimension, which can reflect the development trajectory of the student's scientific research ability and interest. The advantage of this dynamic portrait is that it can capture the dynamic process of students' academic development rather than just a static snapshot, and can better match students' long-term development needs rather than just satisfying short-term interests.

[0033] 103. Obtain the scientific research literature data and scientific research project data to be recommended, and extract the graph relationship between the scientific research literature data and scientific research project data to obtain the scientific research direction knowledge graph;

[0034] In this embodiment, a variety of key scientific research concepts are extracted from scientific research literature data and scientific research project data to obtain a core entity set in the scientific research field, and keywords and subject descriptions of scientific research entities in scientific research project data are extracted to obtain the display relationship type between scientific research entities, and the core entity set is subjected to correlation strength calculation and subject clustering to obtain preliminary scientific research topics; semantic similarity calculation is performed on the scientific research entities in each scientific research topic group in the preliminary scientific research topics to obtain sub-topic division results, and the preliminary scientific research topics and sub-topic division results are subjected to standardized mapping of topics and balance adjustment of division results to obtain a scientific research direction classification system; the scientific research direction classification system is subjected to semantic similarity calculation and subject clustering to obtain a sub-topic division result. The co-occurrence of entities in the system is counted and the point mutual information value between entities is calculated to obtain the entity semantic association, and based on the entity semantic association and the display relationship type, a preliminary association graph between scientific research entities is constructed; the entity importance score of each scientific research entity in the preliminary association graph is calculated, and based on the entity importance score and the entity semantic association, the preliminary association graph is adjusted to obtain a semantic association network, and the semantic association network is structured to obtain a preliminary scientific research direction knowledge graph; the graph nodes and edges in the preliminary scientific research direction knowledge graph are represented by vectors and cross-domain meta-path searches are performed to obtain the final scientific research direction knowledge graph.

[0035] In practical applications, scientific research literature data and scientific research project data to be recommended are obtained. These data come from multiple sources such as but not limited to academic databases, scientific research project management systems, etc. Professional term extraction algorithms such as TextRank or RAKE are used to extract a variety of key scientific research concepts from these data. These concepts constitute the core entity set in the scientific research field. At the same time, keywords and topic descriptions of each scientific research entity are extracted from the scientific research project data to determine the explicit relationship types between scientific research entities (including topic relationships (reflecting the hierarchical or inclusive relationship between scientific research topics), method relationships (indicating the association between a certain research method or technology and a specific research field), problem-solution relationships (indicating the relationship between a certain research method or technology and a specific problem), temporal relationships (reflecting the evolution of scientific research topics or methods), application relationships (indicating the application of research results in practical fields) and comparative relationships (indicating the comparative relationship between different methods or technologies), etc., and point mutual information (PMI) is used to calculate the association strength of the core entity set to measure the degree of association between entities, and a hierarchical clustering algorithm is used to cluster these entities to obtain preliminary scientific research topics. The Word2Vec model is then used to calculate the semantic similarity of the entities in the preliminary scientific research topics, so that based on these related Similarity is used to obtain the subtopic division result, that is, assuming that the word vector cosine similarity of "deep learning" and "neural network" is very high, they may be divided into the same subtopic; then the preliminary scientific research topics and subtopic division results are standardized and balanced to ensure that our scientific research direction classification system meets the requirements of standardization and maintains a reasonable balance between various topics and subtopics, and obtains the scientific research direction classification system; then the entity co-occurrence in the scientific research direction classification system is counted and the point mutual information value between entities is calculated to obtain the semantic relevance between entities, and based on the entity semantic relevance and the display relationship type, a scientific research direction classification system is constructed. A preliminary association graph between scientific research entities is constructed; the PageRank algorithm is then used to calculate the importance score of each scientific research entity in the preliminary association graph. Based on the entity importance score and semantic association, the preliminary association graph is adjusted to construct a semantic association network between scientific research entities. For example, if the entity "machine learning" is referenced or associated with many other important entities, its PageRank value will be higher, and its importance in the knowledge graph will also increase accordingly. This semantic association network is then structured to obtain a preliminary scientific research direction knowledge graph (this preliminary knowledge graph contains rich scientific research entity and relationship information).We then used knowledge graph embedding technology to vectorize the graph nodes and edges in the preliminary research direction knowledge graph, mapping the entities and relationships in the graph into a low-dimensional vector space. We then used the meta-path2vec algorithm to perform cross-domain meta-path search on the resulting vector representations to capture the complex semantic relationships between different types of nodes and discover potential interdisciplinary connections. Ultimately, we developed a comprehensive and in-depth research direction knowledge graph that effectively captures the complex structure and potential connections within the research field, reflecting the interrelationships and evolutionary trends between different research topics. This not only enables traditional content-based recommendations, but also enables deeper semantic reasoning and association discovery based on the graph structure.

[0036] 104. Cross-connect the student dynamic profile and the scientific research direction knowledge graph to obtain a recommendation candidate set, and perform multi-dimensional recommendation scoring and sorting on the recommendation candidate set to obtain a list of recommended scientific research directions;

[0037] In this embodiment, the academic interest characteristics, research ability characteristics and development potential index in the student dynamic portrait are multi-dimensionally decomposed to obtain the student research feature tensor, and the student research feature tensor is used to project the scientific research direction knowledge graph to obtain a personalized weighted knowledge graph; a preset topic random walk model is used to perform random walk extraction of high-frequency research topics on the personalized weighted knowledge graph to obtain a research topic sequence corresponding to the recommended student, and the research topic sequence is thematically clustered to obtain a potential research direction set; based on the potential research direction set, the personalized weighted knowledge graph is expanded in the local network and related research sub-fields are identified to obtain candidate research communities, and the historical research trajectory corresponding to the recommended student is used to perform time series correlation analysis on the candidate research communities to obtain a development path map; based on the development path map, the potential breakthrough points and multi-research target optimization of the recommended student are predicted to obtain a recommendation candidate set. ; Based on the preprocessed multi-source heterogeneous data, a set of scoring dimensions for the students to be recommended is constructed, and a multi-dimensional quantitative evaluation is performed on each scientific research direction in the recommendation candidate set to obtain the original scoring matrix, and the scoring dimension set is hierarchically clustered and the importance of the hierarchical tree nodes is quantified to obtain the initial weight vector; a time series analysis is performed on the students' recent research behavior data in the preprocessed multi-source heterogeneous data to obtain an interest drift model, and the initial weight vector and the interest drift model are fused to obtain a dynamic weight function; the original scoring matrix and the dynamic weight function are subjected to time series decomposition and convolution operations of periodic patterns to obtain a time-varying weighted scoring matrix, and the time-varying weighted scoring matrix is ​​subjected to non-dominated sorting and hierarchical analysis to obtain a comprehensive ranking score; the comprehensive ranking score is normalized and the adjacency matrix is ​​constructed to obtain a state transition matrix, and the state transition matrix is ​​calculated for steady-state distribution to obtain a recommended list of scientific research directions.

[0038] In practical applications, the Tucker decomposition method is used to perform multi-dimensional decomposition of the academic interest characteristics, research ability characteristics and development potential index in the student dynamic portrait to capture the potential interactions between student characteristics and obtain a more compact and information-rich student research feature tensor. Based on the student research feature tensor, the tensor-matrix multiplication is used to project the scientific research direction knowledge graph, and the student feature tensor is multiplied by the adjacency matrix of the knowledge graph to integrate the student's personal characteristics into the knowledge graph, thereby obtaining a personalized weighted knowledge graph (the weight reflects the importance of different research topics to the student). Then, a preset topic random walk model such as the Node2Vec model is used to extract high-frequency research topics from the graph. Through multiple random walks, a series of research topic sequences are obtained (where these sequences reflect the correlation between topics in the knowledge graph, and also take into account the personal characteristics of students). The obtained research topic sequences are then clustered to identify potential research direction sets, so that similar research topics can be grouped together. Each group can be regarded as a potential research direction. Then, based on the potential research direction set, the graph convolutional network is used to extract high-frequency research topics. The network expands the personalized weighted knowledge graph locally to explore related topics and subfields around each potential research direction, discovering research subfields that are closely related to the potential research direction but may have been overlooked by the initial clustering, thereby obtaining a more comprehensive candidate research community. Long-short-term memory networks are used to analyze the changing trends of students' past research interests to predict their possible future research directions. This information is integrated into candidate research communities to form a development path map. Based on the development path map, a deep Q-network is used to predict students' potential breakthrough points and perform multi-research objective optimization. That is, by modeling the research direction selection problem as a multi-step decision-making process and simulating different research paths, potential breakthrough points, i.e., research directions that may bring significant progress, are found. At the same time, a multi-objective optimization algorithm is used to consider the balance of multiple research objectives, such as academic influence, innovation, and practical value. Ultimately, a recommended candidate set for a specific student is obtained. (This candidate set not only takes into account the student's current research interests and abilities, but also incorporates the prediction of disciplinary development trends and potential breakthrough points. It reflects both the student's personal characteristics and the knowledge structure and development dynamics of the entire academic field.)

[0039] Next, based on the pre-processed multi-source heterogeneous data, a set of scoring dimensions for the recommended students is constructed (this set may include multiple dimensions such as academic influence, innovation, practical value, and interdisciplinary degree), and a multi-dimensional quantitative evaluation is performed on each scientific research direction in the recommendation candidate set (including text analysis, citation analysis, expert scoring, etc.). An original scoring matrix is ​​obtained, in which each row represents a scientific research direction and each column represents a scoring dimension. A hierarchical clustering algorithm is then used to hierarchically cluster the scoring dimension set to identify interrelated scoring dimensions, forming a hierarchical structure, and using methods such as PageRank or HITS. The algorithm considers factors such as the position of the node in the tree and the number of child nodes it contains, quantifies the importance of each node in this hierarchical tree, and obtains an initial weight vector to reflect the relative importance of different scoring dimensions; then uses a long short-term memory network to perform time series analysis on students' recent research behavior data (such as recently read papers, participated projects, published articles, etc.) to obtain an interest drift model to capture the trend of students' research interests changing over time, and uses exponential decay or sigmoid function to fuse the initial weight vector and the interest drift model. The weight can be dynamically adjusted as time passes and students' interests change. A dynamic weight function that can change with time is obtained; then the original score matrix and the dynamic weight function are decomposed into trend, cycle and random components using wavelet transform, and the periodic pattern convolution operation is performed on the result of the time series decomposition to capture the periodic changes in the score, thus obtaining a time-varying weighted score matrix; then the time-varying weighted score matrix is ​​subjected to non-dominated sorting and hierarchical analysis to find a balance between multiple goals and consider the relative importance of different dimensions, thereby obtaining a comprehensive ranking score by comprehensively considering multiple evaluation dimensions and time factors; then the comprehensive ranking score is analyzed. Normalization is performed to ensure that all scores are on the same scale. Based on these normalized scores, an adjacency matrix is ​​constructed to describe the similarities or transition probabilities between different research directions. Through certain mathematical transformations, this adjacency matrix is ​​converted into a state transition matrix, where each element represents the probability of switching from one research direction to another. The steady-state distribution of this state transition matrix is ​​then calculated by solving the matrix eigenvalue problem, thereby deriving the long-term probability of each research direction being selected. Based on this probability distribution, a final list of recommended research directions is generated, with directions with higher probabilities being ranked higher. This algorithm not only provides recommendations for the most suitable research directions currently but also predicts possible future research trends. For example, for a student majoring in machine learning, "Explainable Artificial Intelligence" might be recommended as a key research direction because it is not only relevant to the student's background, but also has high academic influence and practical value, and is a rapidly developing frontier field.

[0040] 105. Explain, generate and dynamically adjust the list of recommended scientific research directions to generate the final scientific research direction recommendation plan.

[0041] In this embodiment, the previously constructed scientific research direction knowledge graph is used to extract key concepts, important research issues and latest developments related to the direction, and combined with the student's personal characteristics, an explanation is given as to why this direction is suitable for the student, which may involve aspects such as the student's research background, skill advantages and development potential. Then, a GPT series of models is used to generate coherent text paragraphs based on given key information points. In the generation process, the personalization of the explanation needs to be considered to ensure that the explanation content not only covers general information about the scientific research direction, but also includes personalized content related to specific students. Then, a feedback mechanism is pre-designed to allow students to rate or provide opinions on the recommended scientific research direction. Based on these feedbacks, an online learning algorithm is used. To continuously update and optimize the recommendation model, a regular update mechanism should be established to regularly crawl and analyze the latest academic literature and conference information to capture the latest trends and breakthroughs in scientific research. For example, the knowledge graph and recommendation model should be updated weekly or monthly. This feedback mechanism dynamically adjusts the recommendation list based on multiple factors, including student feedback, recent developments in the scientific research field, and changes in the external environment. During this dynamic adjustment, the diversity and novelty of the recommendations should be considered, using a multi-armed bandit algorithm to appropriately introduce research directions that may be unexpected but have great potential. All these elements should then be integrated to generate a comprehensive, personalized, and dynamic research direction recommendation plan. This plan should not only include a list of recommended research directions, but also include a detailed explanation of each direction, the rationale for the recommendation, possible research questions, relevant recent literature, and potential collaborators or research teams. Furthermore, a regular review and update mechanism can be established, such as a comprehensive evaluation and adjustment of the recommendation plan at regular intervals (e.g., every semester or every year) to ensure that the recommendations remain relevant and up-to-date.

[0042] In an embodiment of the present invention, multi-source heterogeneous data of students to be recommended are obtained, and multi-modal features are extracted and integrated and student portraits are dynamically modeled for the multi-source heterogeneous data to obtain student dynamic portraits, and graph relationship extraction is performed on the obtained scientific research literature data and scientific research project data to be recommended to obtain a scientific research direction knowledge graph, and then the student dynamic portraits and the scientific research direction knowledge graph are cross-connected to obtain a recommendation candidate set, and the recommendation candidate set is multi-dimensionally recommended scored and sorted to obtain a scientific research direction recommendation list, thereby interpreting, generating and dynamically adjusting the recommendation list to generate a final recommendation plan for the students to be recommended, thereby achieving a full combination of student information and relevant scientific research information, and improving the accuracy of the scientific research directions recommended to students.

[0043] The above describes the method for recommending scientific research directions based on student portraits in an embodiment of the present invention. The following describes the device for recommending scientific research directions based on student portraits in an embodiment of the present invention. Figure 2 In one embodiment of the present invention, a device for recommending research directions based on student portraits includes:

[0044] A preprocessing module 201 is used to obtain multi-source heterogeneous data of students to be recommended, and preprocess the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data;

[0045] The feature representation module 202 is configured to extract and fuse multimodal features from the preprocessed multi-source heterogeneous data to obtain a comprehensive feature representation, and based on the comprehensive feature representation, dynamically model the student portrait of the student to be recommended to obtain a dynamic student portrait;

[0046] The relationship extraction module 203 is used to obtain the scientific research literature data and scientific research project data to be recommended, and perform graph relationship extraction on the scientific research literature data and the scientific research project data to obtain a scientific research direction knowledge graph;

[0047] A cross-connection module 204 is configured to cross-connect the student dynamic portrait and the scientific research direction knowledge graph to obtain a recommendation candidate set, and perform multi-dimensional recommendation scoring and sorting on the recommendation candidate set to obtain a scientific research direction recommendation list;

[0048] The dynamic adjustment module 205 is used to interpret, generate and dynamically adjust the scientific research direction recommendation list to generate a final scientific research direction recommendation plan.

[0049] In an embodiment of the present invention, multi-source heterogeneous data of students to be recommended are obtained, and multi-modal features are extracted and integrated and student portraits are dynamically modeled for the multi-source heterogeneous data to obtain student dynamic portraits, and graph relationship extraction is performed on the obtained scientific research literature data and scientific research project data to be recommended to obtain a scientific research direction knowledge graph, and then the student dynamic portraits and the scientific research direction knowledge graph are cross-connected to obtain a recommendation candidate set, and the recommendation candidate set is multi-dimensionally recommended scored and sorted to obtain a scientific research direction recommendation list, thereby interpreting, generating and dynamically adjusting the recommendation list to generate a final recommendation plan for the students to be recommended, thereby achieving a full combination of student information and relevant scientific research information, and improving the accuracy of the scientific research directions recommended to students.

[0050] above Figure 2 The device for recommending scientific research directions based on student portraits in an embodiment of the present invention is described in detail from the perspective of modular functional entities. The device for recommending scientific research directions based on student portraits in an embodiment of the present invention is described in detail from the perspective of hardware processing.

[0051] Figure 3 Figure 3 is a schematic diagram of the structure of a device for recommending research directions based on student profiles, provided by an embodiment of the present invention. The device 300 may vary significantly depending on configuration or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors), memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing applications 333 or data 332. The memory 320 and storage medium 330 may be either transient or persistent storage. The program stored in the storage medium 330 may include one or more modules (not shown), each of which may include a series of instructions for operating on the device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, executing the series of instructions stored in the storage medium 330 on the device 300.

[0052] The device 300 for recommending research directions based on student profiles may further 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 understood by those skilled in the art that Figure 3 The structure of the device for recommending scientific research directions based on student portraits shown does not constitute a limitation of the device for recommending scientific research directions based on student portraits, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0053] The present invention also provides a device for recommending scientific research directions based on student portraits. The computer device includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes each step of the method for recommending scientific research directions based on student portraits in the above-mentioned embodiments.

[0054] 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 run on a computer, the computer executes the various steps of the method for recommending scientific research directions based on student portraits.

[0055] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0056] If the integrated unit is implemented as 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, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0057] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0058] 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 above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can 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 method for recommending research directions based on student portraits, characterized in that: The method for recommending scientific research directions based on student portraits includes: Acquiring multi-source heterogeneous data of students to be recommended, and preprocessing the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data; The student text data in the preprocessed multi-source heterogeneous data is segmented to obtain a word vector sequence, and the student numerical data in the preprocessed multi-source heterogeneous data is time-series analyzed to obtain learning performance trend characteristics, and the scientific research activity data in the preprocessed multi-source heterogeneous data is structured extracted to obtain scientific research experience characteristics; the word vector sequence is vectorized and mapped to obtain a word vector set of student-related characteristics, and the word vector set is clustered by interest topics to obtain a student interest topic distribution; the student interest topic distribution, the learning performance trend characteristics and the scientific research experience characteristics are integrated to obtain a multimodal feature matrix, and the multimodal feature matrix is ​​normalized and dimensionally reduced to obtain a comprehensive feature representation of the student to be recommended, and based on the comprehensive feature representation, a student scientific research portrait framework of the student to be recommended under a preset plurality of key portrait dimensions is constructed, and the student scientific research is A cluster analysis is performed on the academic interest characteristics in the portrait framework to obtain the research topic preference distribution of the recommended students, and a quantitative evaluation is performed on the research ability characteristics in the student scientific research portrait framework to obtain the research ability scores of the recommended students in different research methods and skills, and a predictive analysis is performed on the scientific research potential characteristics in the student scientific research portrait framework to obtain the development potential index of the recommended students in different research directions; a correlation calculation is performed on the research topic preference distribution, the research ability score and the development potential index to obtain a correlation coefficient matrix, and based on the correlation coefficient matrix, an internal association network corresponding to the student scientific research characteristics is constructed; a time series decomposition is performed on the characteristic values ​​of each node in the internal association network to obtain the time series components of the student scientific research characteristics, and a weighted combination of the time series components is performed based on the historical academic data in the preprocessed multi-source heterogeneous data to obtain a dynamic student portrait; Acquire scientific research literature data and scientific research project data to be recommended, and extract graph relationships between the scientific research literature data and the scientific research project data to obtain a scientific research direction knowledge graph; The academic interest characteristics, research ability characteristics and development potential index in the dynamic student portrait are decomposed in multiple dimensions to obtain a student research feature tensor, and the student research feature tensor is used to project the scientific research direction knowledge graph to obtain a personalized weighted knowledge graph; a preset topic random walk model is used to perform random walk extraction of high-frequency research topics on the personalized weighted knowledge graph to obtain a research topic sequence corresponding to the recommended student, and the research topic sequence is subjected to topic clustering to obtain a set of potential research directions; based on the potential research direction set, the personalized weighted knowledge graph is expanded in local networks and related research sub-fields are identified to obtain candidate research communities, and the historical research trajectory corresponding to the recommended student is used to perform time series correlation analysis on the candidate research communities to obtain a development path map; based on the development path map, the potential breakthrough points and multi-research target optimization of the recommended student are predicted to obtain a recommendation candidate set, and the recommendation candidate set is scored and ranked in multiple dimensions to obtain a list of recommended scientific research directions; The scientific research direction recommendation list is interpreted, generated and dynamically adjusted to generate a final scientific research direction recommendation plan.

2. The method for recommending scientific research directions based on student portraits according to claim 1, characterized in that: The preprocessing of the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data includes: Performing regular expression matching and outlier marking on text fields in the multi-source heterogeneous data to obtain outlier marked data, and performing median replacement and multiple interpolation on marked outliers in the outlier marked data to obtain complete multi-source data; Multi-dimensional feature extraction and cluster analysis are performed on the corresponding student information in the integrity multi-source data to obtain a student unique identifier, and based on the student unique identifier and the preset log record time sequence, structured extraction of scientific research results information and annotation of academic fields are performed on the integrity multi-source data to obtain preprocessed multi-source heterogeneous data.

3. The method for recommending scientific research directions based on student portraits according to claim 1, characterized in that: The graph relationship extraction of the scientific research literature data and the scientific research project data to obtain a scientific research direction knowledge graph includes: Extracting multiple key scientific research concepts from the scientific research literature data and the scientific research project data to obtain a core entity set in the scientific research field, extracting keywords and subject descriptions of scientific research entities in the scientific research project data to obtain display relationship types between scientific research entities, and performing association strength calculation and subject clustering on the core entity set to obtain preliminary scientific research themes; Calculating semantic similarity of scientific research entities within each scientific research theme group in the preliminary scientific research theme to obtain sub-theme division results, and performing standardized mapping of themes and balance adjustment of the division results on the preliminary scientific research theme and the sub-theme division results to obtain a scientific research direction classification system; Counting the co-occurrence of entities in the scientific research direction classification system and calculating the point mutual information value between entities to obtain the entity semantic association, and constructing a preliminary association graph between scientific research entities based on the entity semantic association and the display relationship type; Calculating the entity importance score of each scientific research entity in the preliminary association graph, and adjusting the association of the preliminary association graph based on the entity importance score and the entity semantic association degree to obtain a semantic association network, and performing a structured representation on the semantic association network to obtain a preliminary scientific research direction knowledge graph; The graph nodes and edges in the preliminary scientific research direction knowledge graph are vectorized and subjected to cross-domain meta-path search to obtain the final scientific research direction knowledge graph.

4. The method for recommending scientific research directions based on student portraits according to claim 1, characterized in that: The recommendation candidate set is scored and sorted in multiple dimensions to obtain a list of recommended research directions, including: Based on the preprocessed multi-source heterogeneous data, a set of scoring dimensions for the students to be recommended is constructed, and a multi-dimensional quantitative evaluation is performed on each scientific research direction in the recommendation candidate set to obtain an original scoring matrix. Hierarchical clustering and importance quantification of hierarchical tree nodes are performed on the scoring dimension set to obtain an initial weight vector. Performing time series analysis on the students' recent research behavior data in the preprocessed multi-source heterogeneous data to obtain an interest drift model, and fusing the initial weight vector and the interest drift model to obtain a dynamic weight function; Performing time series decomposition and periodic pattern convolution operations on the original scoring matrix and the dynamic weight function to obtain a time-varying weighted scoring matrix, and performing non-dominated sorting and hierarchical analysis on the time-varying weighted scoring matrix to obtain a comprehensive ranking score; The comprehensive ranking scores are normalized and an adjacency matrix is ​​constructed to obtain a state transfer matrix, and a steady-state distribution calculation is performed on the state transfer matrix to obtain a list of recommended scientific research directions.

5. A device for recommending research directions based on student portraits, characterized in that: The scientific research direction recommendation device based on student portraits includes: A preprocessing module is used to obtain multi-source heterogeneous data of the students to be recommended, and preprocess the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data; A feature representation module is used to segment the student text data in the preprocessed multi-source heterogeneous data to obtain a word vector sequence, perform time series analysis on the student numerical data in the preprocessed multi-source heterogeneous data to obtain learning performance trend features, and perform structured extraction on the scientific research activity data in the preprocessed multi-source heterogeneous data to obtain scientific research experience features; perform vectorized mapping on the word vector sequence to obtain a word vector set of student-related features, and cluster the word vector set by interest topics to obtain a student interest topic distribution; integrate the student interest topic distribution, the learning performance trend features and the scientific research experience features to obtain a multimodal feature matrix, and perform normalization and dimensionality reduction on the multimodal feature matrix to obtain a comprehensive feature representation of the student to be recommended, and based on the comprehensive feature representation, construct a student scientific research portrait framework for the student to be recommended under multiple preset key portrait dimensions, and A cluster analysis is performed on the academic interest characteristics in the student scientific research portrait framework to obtain the research topic preference distribution of the recommended students, and a quantitative evaluation is performed on the research ability characteristics in the student scientific research portrait framework to obtain the research ability scores of the recommended students in different research methods and skills, and a predictive analysis is performed on the scientific research potential characteristics in the student scientific research portrait framework to obtain the development potential index of the recommended students in different research directions; a correlation calculation is performed on the research topic preference distribution, the research ability score and the development potential index to obtain a correlation coefficient matrix, and based on the correlation coefficient matrix, an internal association network corresponding to the student scientific research characteristics is constructed; a time series decomposition is performed on the characteristic values ​​of each node in the internal association network to obtain the time series components of the student scientific research characteristics, and a weighted combination of the time series components is performed based on the historical academic data in the preprocessed multi-source heterogeneous data to obtain a dynamic student portrait; A relationship extraction module is used to obtain scientific research literature data and scientific research project data to be recommended, and to extract graph relationships between the scientific research literature data and the scientific research project data to obtain a scientific research direction knowledge graph; A cross-connection module is used to perform multi-dimensional decomposition of the academic interest characteristics, research ability characteristics and development potential index in the dynamic student portrait to obtain a student research feature tensor, and use the student research feature tensor to project the scientific research direction knowledge graph to obtain a personalized weighted knowledge graph; use a preset topic random walk model to perform random walk extraction of high-frequency research topics on the personalized weighted knowledge graph to obtain a research topic sequence corresponding to the recommended student, and perform topic clustering on the research topic sequence to obtain a potential research direction set; based on the potential research direction set, the personalized weighted knowledge graph is expanded in local networks and related research sub-fields are identified to obtain candidate research communities, and the candidate research communities are subjected to time series correlation analysis based on the historical research trajectory corresponding to the recommended student to obtain a development path map; based on the development path map, the potential breakthrough points and multi-research target optimization of the recommended student are predicted to obtain a recommendation candidate set, and the recommendation candidate set is scored and ranked in multiple dimensions to obtain a scientific research direction recommendation list; The dynamic adjustment module is used to interpret, generate and dynamically adjust the scientific research direction recommendation list to generate a final scientific research direction recommendation plan.

6. A device for recommending scientific research directions based on student portraits, characterized in that: The device for recommending scientific research directions based on student portraits includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory so that the scientific research direction recommendation device based on student portraits performs the various steps of the scientific research direction recommendation method based on student portraits as described in any one of claims 1-4.

7. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the method for recommending scientific research directions based on student portraits as described in any one of claims 1 to 4 are implemented.

Images