Academic resource recommendation method and system based on knowledge graph and large model

Abstract

The invention provides an academic resource recommendation method and system based on a knowledge graph and a large model. According to the method, firstly, academic resources to be recommended are obtained in batches from a plurality of authoritative academic data sources through an advanced data collection method, and then key entities and entity relationships in the academic resources to be recommended are identified by utilizing a knowledge graph construction technology, so that a knowledge graph with a clear structure and rich contents is formed. On the basis, a deep learning large model, namely a target large model, is selected and customized based on the language understanding ability, generalization performance and training efficiency of the large model. Meanwhile, a multi-level attention mechanism and a fusion layer are introduced, and a training target of a target large model is designed by using a loss function and a multi-dimensional evaluation index. Through strategies such as task customization, training target design and knowledge graph fusion, deep fusion of text information and structured knowledge is realized, so that the accuracy of academic resource recommendation can be improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an academic resource recommendation method and system based on knowledge graphs and large models. Background Art

[0002] With the rapid development of Internet technology and the increasing frequency of global academic exchanges, academic resources are showing an explosive growth trend. Various academic resources such as academic papers, conference proceedings, data sets and patent documents are spread across various online platforms and databases, forming a huge ocean of academic information.

[0003] At present, existing academic resource recommendation methods often rely on simple text matching algorithms, that is, pre-processing operations such as word segmentation and part-of-speech tagging are performed on the text, and then matching is performed according to preset rules. This type of method is suitable for structured text scenarios such as database queries, but has limited capabilities for complex and changeable natural language processing.

[0004] However, existing technologies recommend academic resources through text matching algorithms, which makes it difficult to capture the intrinsic connections and structured knowledge between academic resources, resulting in low accuracy in academic resource recommendations. Summary of the invention

[0005] The embodiment of the present invention provides an academic resource recommendation method based on a knowledge graph and a large model, which can improve the accuracy of academic resource recommendation. The method includes:

[0006] A1: Using web crawler technology and at least one API interface, obtain academic resources to be recommended based on at least one of academic databases, open source platforms and academic conference resources;

[0007] A2: Using entity recognition and relationship extraction technology, determine the key entities and entity relationships in the academic resources to be recommended, and construct a knowledge graph based on the key entities and the entity relationships;

[0008] A3: Determine the target large model based on the language understanding ability, generalization performance and training efficiency of the large model, wherein the target large model includes a multi-level attention mechanism and a fusion layer;

[0009] A4: Determine a training target of the target large model by using at least one of a cross entropy loss function and a BPR loss function and at least one evaluation index, wherein the target large model includes a user feedback mechanism, and the user feedback mechanism is used to adjust parameters of the target large model;

[0010] A5: By integrating the key entity embedding vectors and the entity relationship embedding vectors in the knowledge graph with text vector representations, the knowledge graph is fused with the target big model to recommend target academic resources.

[0011] Preferably,

[0012] After A1 and before A2, further comprising:

[0013] Performing data preprocessing operations on the acquired academic resources to be recommended to generate preprocessed academic resources, wherein the data preprocessing operations are used to perform data cleaning on the academic resources to be recommended, and the data cleaning includes: intelligent deduplication algorithm, standardized data format, missing value filling and noise removal operations based on statistics and rules;

[0014] Determine the text semantics and citation relationships in the preprocessed academic resources using natural language processing technology, wherein the natural language processing technology includes: BERT variants and TextRank optimization algorithms;

[0015] The A2 includes:

[0016] Utilizing entity recognition and relationship extraction techniques, key entities and entity relationships in the preprocessed academic resources are determined, and a knowledge graph is constructed based on the key entities and the entity relationships.

[0017] Preferably,

[0018] After A3 and before A4, further comprising:

[0019] Based on the specific context of academic resources and user needs, the parameters of the target large model are optimized, wherein the parameter optimization includes: adjusting the model architecture and optimizing the pre-training tasks.

[0020] Preferably,

[0021] After A4 and before A5, further comprising:

[0022] Using at least one of a Bayesian optimization algorithm and a hyperparameter search algorithm, adjusting key hyperparameters of the target large model;

[0023] The academic resource recommendation performance of the target large model is evaluated by utilizing the user feedback mechanism and evaluation indicators, wherein the evaluation indicators include at least one of accuracy, recall, F1 score and ranking evaluation indicator NDCG.

[0024] Preferably,

[0025] The A5 includes:

[0026] The key entity embedding vector and the entity relationship embedding vector in the knowledge graph are represented by a text vector to construct a target knowledge graph;

[0027] The target knowledge graph is fused with the target large model based on a preset fusion strategy to recommend target academic resources, wherein the fusion strategy includes at least one of splicing, weighted summation and attention mechanism.

[0028] In a second aspect, an embodiment of the present invention provides an academic resource recommendation system based on a knowledge graph and a large model, the system comprising:

[0029] Acquisition module: used to acquire academic resources to be recommended based on at least one of academic databases, open source platforms and academic conference resources by using web crawler technology and at least one API interface;

[0030] A knowledge graph construction module is used to determine the key entities and entity relationships in the academic resources to be recommended by using entity recognition and relationship extraction technology, and to construct a knowledge graph based on the key entities and the entity relationships;

[0031] Large model selection module: used to determine the target large model based on the language understanding ability, generalization performance and training efficiency of the large model, wherein the target large model includes a multi-level attention mechanism and a fusion layer;

[0032] A training target determination module: used to determine the training target of the target large model by using at least one of the cross entropy loss function and the BPR loss function and at least one evaluation index, wherein the target large model includes a user feedback mechanism, and the user feedback mechanism is used to adjust the parameters of the target large model;

[0033] Fusion module: used to fuse the knowledge graph with the target big model by integrating the key entity embedding vector and the entity relationship embedding vector in the knowledge graph with the text vector representation, so as to recommend target academic resources.

[0034] Preferably,

[0035] After the acquisition module and before the knowledge graph construction module, it further includes: a data processing module and a semantic determination module;

[0036] The data processing module is used to perform data preprocessing operations on the acquired academic resources to be recommended to generate preprocessed academic resources, wherein the data preprocessing operations are used to perform data cleaning on the academic resources to be recommended, and the data cleaning includes: intelligent deduplication algorithm, standardized data format, missing value filling and noise removal operations based on statistics and rules;

[0037] The semantic determination module is used to determine the text semantics and citation relationships in the preprocessed academic resources using natural language processing technology, wherein the natural language processing technology includes: BERT variants and TextRank optimization algorithms;

[0038] The knowledge graph construction module is also used to use entity recognition and relationship extraction technology to determine the key entities and entity relationships in the preprocessed academic resources, and to construct a knowledge graph based on the key entities and the entity relationships.

[0039] Preferably,

[0040] After the large model selection module and before the training target determination module, it further includes:

[0041] The model optimization module is used to optimize the parameters of the target large model based on the specific context of academic resources and user needs, wherein the parameter optimization includes: adjusting the model architecture and optimizing the pre-training tasks.

[0042] Preferably,

[0043] After the training target determination module and before the fusion module, it further includes: a model evaluation module;

[0044] The model evaluation module is used to perform:

[0045] Using at least one of a Bayesian optimization algorithm and a hyperparameter search algorithm, adjusting key hyperparameters of the target large model;

[0046] The academic resource recommendation performance of the target large model is evaluated by utilizing the user feedback mechanism and evaluation indicators, wherein the evaluation indicators include at least one of accuracy, recall, F1 score and ranking evaluation indicator NDCG.

[0047] Preferably,

[0048] The fusion module is further used to execute:

[0049] The key entity embedding vector and the entity relationship embedding vector in the knowledge graph are represented by a text vector to construct a target knowledge graph;

[0050] The target knowledge graph is fused with the target large model based on a preset fusion strategy to recommend target academic resources, wherein the fusion strategy includes at least one of splicing, weighted summation and attention mechanism.

[0051] In a third aspect, the present invention provides an academic resource recommendation system based on a knowledge graph and a large model, comprising: at least one memory and at least one processor;

[0052] The at least one memory is used to store a machine-readable program;

[0053] The at least one processor is used to call the machine-readable program to execute any method described in the first aspect of claim 1.

[0054] In a fourth aspect, the present invention provides a computer-readable medium having computer instructions stored thereon, wherein when the computer instructions are executed by a processor, the processor executes any one of the methods described in the first aspect of claim 1.

[0055] The embodiment of the present invention provides a method and system for recommending academic resources based on knowledge graph and large model. The method first obtains academic papers, data sets and scholar information in batches from multiple authoritative academic data sources (e.g., academic databases, open source platforms and academic conference resources) through advanced data collection methods (e.g., web crawler technology and at least one API interface) to wait for recommended academic resources, laying a solid foundation for subsequent steps. The knowledge graph construction technology is then used to identify key entities and entity relationships in the academic resources to be recommended, forming a knowledge graph of academic fields with clear structure and rich content. On this basis, the present invention selects and customizes a deep learning large model, i.e., a target large model, based on the language comprehension ability, generalization performance and training efficiency of the large model. At the same time, a multi-level attention mechanism is innovatively introduced, which not only focuses on the key information inside the text, but also focuses on the correlation information between cross-text and cross-resources, so as to realize the deep mining of complex relationships between academic resources. In addition, an efficient fusion layer is also designed to realize the seamless connection and complementary enhancement of text information and structured knowledge. Then the training target of the target large model is designed by using the loss function and multi-dimensional evaluation indicators. Through strategies such as task customization, training objective design, and knowledge graph fusion, we have achieved a deep integration of text information and structured knowledge, enhancing the large model's ability to understand the complex relationships between academic resources, thereby improving the accuracy of academic resource recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1It is a flowchart of an academic resource recommendation method based on knowledge graph and big model provided by one embodiment of the present invention;

[0058] Figure 2 It is a flowchart of another academic resource recommendation method based on knowledge graph and big model provided by one embodiment of the present invention;

[0059] Figure 3 It is a schematic diagram of an academic resource recommendation system based on a knowledge graph and a large model provided by an embodiment of the present invention;

[0060] Figure 4 It is a schematic diagram of another academic resource recommendation system based on knowledge graph and big model provided by one embodiment of the present invention;

[0061] Figure 5 It is a schematic diagram of another academic resource recommendation system based on knowledge graph and big model provided by one embodiment of the present invention;

[0062] Figure 6 This is a schematic diagram of another academic resource recommendation system based on knowledge graph and big model provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0064] like Figure 1 As shown, an embodiment of the present invention provides an academic resource recommendation method based on a knowledge graph and a large model, and the method may include the following steps:

[0065] Step 101: using web crawler technology and at least one API interface, based on at least one of an academic database, an open source platform, and academic conference resources, to obtain academic resources to be recommended;

[0066] Step 102: using entity recognition and relationship extraction technology, determine the key entities and entity relationships in the academic resources to be recommended, and construct a knowledge graph based on the key entities and entity relationships;

[0067] Step 103: Determine a target large model based on the language comprehension ability, generalization performance, and training efficiency of the large model, wherein the target large model includes a multi-level attention mechanism and a fusion layer;

[0068] Step 104: Determine a training target of a target large model by using at least one of a cross entropy loss function and a BPR loss function and at least one evaluation index, wherein the target large model includes a user feedback mechanism, and the user feedback mechanism is used to adjust parameters of the target large model;

[0069] Step 105: By integrating the key entity embedding vectors and entity relationship embedding vectors in the knowledge graph with the text vector representation, the knowledge graph is fused with the target large model to recommend target academic resources.

[0070] In an embodiment of the present invention, a method for recommending academic resources based on knowledge graph and large model is provided. The method first obtains academic papers, data sets and scholar information in batches from multiple authoritative academic data sources (e.g., academic databases, open source platforms and academic conference resources) through advanced data collection methods (e.g., web crawler technology and at least one API interface) to wait for recommended academic resources, laying a solid foundation for subsequent steps. The knowledge graph construction technology is then used to identify key entities and entity relationships in the academic resources to be recommended, forming a knowledge graph of academic fields with clear structure and rich content. On this basis, the present invention selects and customizes a deep learning large model, i.e., a target large model, based on the language comprehension ability, generalization performance and training efficiency of the large model. At the same time, a multi-level attention mechanism is innovatively introduced, which not only focuses on the key information inside the text, but also focuses on the correlation information between cross-text and cross-resources, so as to realize the deep mining of complex relationships between academic resources. In addition, an efficient fusion layer is also designed to realize the seamless connection and complementary enhancement of text information and structured knowledge. Then the loss function and multi-dimensional evaluation indicators are used to design the training objectives of the target large model. Through strategies such as task customization, training objective design, and knowledge graph fusion, we have achieved a deep integration of text information and structured knowledge, enhancing the large model's ability to understand the complex relationships between academic resources, thereby improving the accuracy of academic resource recommendations.

[0071] In order to recommend high-quality academic resource data, in one embodiment of the present invention, the above embodiment further includes after step 101 and before step 102:

[0072] Performing data preprocessing operations on the acquired academic resources to be recommended to generate preprocessed academic resources, wherein the data preprocessing operations are used to perform data cleaning on the academic resources to be recommended, and the data cleaning includes: intelligent deduplication algorithm, standardized data format, missing value filling and noise removal operations based on statistics and rules;

[0073] Determine the text semantics and citation relationships in the preprocessed academic resources using natural language processing technology, wherein the natural language processing technology includes: BERT variants and TextRank optimization algorithms;

[0074] The A2 includes:

[0075] Utilizing entity recognition and relationship extraction techniques, key entities and entity relationships in the preprocessed academic resources are determined, and a knowledge graph is constructed based on the key entities and the entity relationships.

[0076] In an embodiment of the present invention, in order to recommend high-quality academic resources, it is necessary to perform data preprocessing operations on the collected academic resources to be recommended. In the data preprocessing stage, a strict data cleaning process is implemented, including intelligent deduplication algorithms, standardized data formats, missing value filling based on statistics and rules, and the use of machine learning technology to identify and remove noise data to ensure the purity and integrity of the data. In addition, by integrating advanced natural language processing technologies (such as BERT variants and TextRank optimization algorithms), the text semantics is deeply excavated to achieve accurate extraction of keywords and detailed classification and annotation of academic resources, laying a solid data foundation for subsequent steps. Based on the high-quality data after preprocessing, advanced entity recognition and relationship extraction technologies can be used to accurately identify key entities (such as authors, papers, journals and institutions) and the complex relationships between them (such as writing, citations and collaborations) from academic papers, data sets and scholar information. On this basis, the powerful capabilities of the graph database are used to build a knowledge map. At the same time, in order to ensure the timeliness and accuracy of the knowledge map, an automated data update mechanism can be designed to track newly published papers and updated citation relationships in real time, so as to achieve dynamic maintenance and continuous optimization of the knowledge map.

[0077] In order to improve the performance of the large model in the recommendation task, in one embodiment of the present invention, after step A3 and before step A4 in the above embodiment, the following is further included:

[0078] Based on the specific context of academic resources and user needs, the parameters of the target large model are optimized, wherein the parameter optimization includes: adjusting the model architecture and optimizing the pre-training tasks.

[0079] In an embodiment of the present invention, for the special needs of academic resource recommendation, the target large model is a basic large model (e.g., BERT series and GPT series) carefully selected from the current top deep learning large models by comprehensively considering the model's language comprehension ability, generalization performance and training efficiency. This model evaluates the performance of different models in the task of academic resource recommendation through comparative experiments, and determines the final large model to be used based on the experimental results. On this basis, in order to ensure that the large model can fully adapt to the specific context and needs of academic resources, it can be adjusted and optimized in a targeted manner (e.g., adjusting the model architecture, optimizing the pre-training tasks) to improve the performance of the large model in the recommendation task.

[0080] In order to further improve the performance of the large model in the recommendation task, in one embodiment of the present invention, after step A4 and before step A5 in the above embodiment, the following is further included:

[0081] Using at least one of a Bayesian optimization algorithm and a hyperparameter search algorithm, adjusting key hyperparameters of the target large model;

[0082] The academic resource recommendation performance of the target large model is evaluated by utilizing the user feedback mechanism and evaluation indicators, wherein the evaluation indicators include at least one of accuracy, recall, F1 score and ranking evaluation indicator NDCG.

[0083] In an embodiment of the present invention, in order to further improve the performance of the large model in the recommendation task, in the model training and optimization stage, first, the key hyperparameters of the large model can be fine-tuned using advanced algorithms such as Bayesian optimization and hyperparameter search to find the optimal parameter combination. Secondly, large-scale parallel training is performed on a high-performance computing platform to accelerate the model convergence process and improve training efficiency. At the same time, strategies such as early stopping and learning rate decay can be used to prevent model overfitting and improve generalization ability. Finally, special optimization and adjustment are performed for the fusion layer and the recommendation algorithm, and the impact of different strategies on model performance is evaluated through comparative experiments, and the optimal solution is selected to achieve continuous performance improvement and functional improvement of the recommendation system. In order to ensure the long-term effectiveness and stability of the recommendation system, a complete system evaluation mechanism can be established (for example, regularly collecting user feedback, comparing and testing the recommendation effects of different versions, and analyzing system logs, etc.), and evaluation indicators (for example, accuracy, recall, F1 score, and ranking evaluation indicator NDCG, etc.) are used to comprehensively evaluate the performance of the recommendation system. Based on the evaluation results, we continuously adjust the optimization strategy, including optimizing model parameters, improving recommendation algorithms, and updating knowledge graphs, to ensure that the recommendation system can continue to adapt to the rapid changes in academic resources and the diversified development of user needs. In addition, this solution focuses on tracking the latest progress in natural language processing technology and knowledge graph technology, so as to apply new technologies to the recommendation system in a timely manner and promote the continuous improvement of system performance.

[0084] In order to improve the accuracy of academic resource recommendation, in one embodiment of the present invention, step A5 in the above embodiment includes:

[0085] The key entity embedding vector and the entity relationship embedding vector in the knowledge graph are represented by a text vector to construct a target knowledge graph;

[0086] The target knowledge graph is fused with the target large model based on a preset fusion strategy to recommend target academic resources, wherein the fusion strategy includes at least one of splicing, weighted summation and attention mechanism.

[0087] In an embodiment of the present invention, in the process of model training, in order to achieve deep fusion of knowledge graph and text information, and seamless connection of text information and structured information, the entity embedding vector and relationship embedding vector in the knowledge graph are organically combined with the text vector representation, and multiple methods are tried in the training process to determine the optimal fusion strategy (for example, splicing, weighted summation or attention mechanism, etc.). This deep fusion method enables the large model to understand the connotation and extension of academic resources more comprehensively, and further improve the accuracy and relevance of recommendations.

[0088] like Figure 2 As shown, in order to more clearly illustrate the technical solution and advantages of the present invention, the following is a detailed description of an academic resource recommendation method based on a knowledge graph and a large model provided by an embodiment of the present invention, which may specifically include the following steps:

[0089] Step 201: using web crawler technology and at least one API interface, based on at least one of an academic database, an open source platform, and academic conference resources, to obtain academic resources to be recommended;

[0090] Specifically, the scope of data collection is not limited to the basic information of the paper (such as name, abstract, keywords and journal, etc.), but also deeply explores deep-level information such as citation relationships and author cooperation networks.

[0091] Step 202: performing data preprocessing operations on the acquired academic resources to be recommended to generate preprocessed academic resources, wherein the data preprocessing operations are used to perform data cleaning on the academic resources to be recommended, and the data cleaning includes: intelligent deduplication algorithm, standardized data format, missing value filling and noise removal operations based on statistics and rules;

[0092] Step 203: using natural language processing technology to determine text semantics and citation relationships in the preprocessed academic resources, wherein the natural language processing technology includes: BERT variants and TextRank optimization algorithms;

[0093] Step 204: using entity recognition and relationship extraction technology, determining key entities and entity relationships in the preprocessed academic resources, and constructing a knowledge graph based on the key entities and entity relationships;

[0094] Step 205: Determine a target large model based on the language comprehension ability, generalization performance, and training efficiency of the large model, wherein the target large model includes a multi-level attention mechanism and a fusion layer;

[0095] Specifically, the fusion layer can adopt advanced fusion algorithms (for example, weighted fusion based on attention mechanism, fusion assisted by graph neural network, etc.) to deeply integrate the structured information in the knowledge graph with the text vector representation, so as to achieve seamless connection and complementary enhancement of text information and structured knowledge.

[0096] Step 206: Optimizing parameters of the target large model based on the specific context of the academic resources and user needs, wherein the parameter optimization includes: adjusting the model architecture and optimizing the pre-training tasks;

[0097] Step 207: Determine a training target of a target large model by using at least one of a cross entropy loss function and a BPR loss function and at least one evaluation index, wherein the target large model includes a user feedback mechanism, and the user feedback mechanism is used to adjust parameters of the target large model;

[0098] Specifically, in order to accurately measure the accuracy of the big model's recommendations and guide the training direction of the big model, scientific and reasonable training objectives can be designed, and a combination of multiple loss functions such as the cross entropy loss function and the BPR loss function can be used to comprehensively reflect the difference between the recommendation results of the big model and the actual situation.

[0099] Step 208: Using at least one of a Bayesian optimization algorithm and a hyperparameter search algorithm, adjust key hyperparameters of the target large model;

[0100] Step 209: using the user feedback mechanism and evaluation indicators to evaluate the academic resource recommendation performance of the target large model, wherein the evaluation indicators include: at least one of accuracy, recall, F1 score, and ranking evaluation indicator NDCG;

[0101] Step 210: The key entity embedding vectors and entity relationship embedding vectors in the knowledge graph are represented by the text vector to construct a target knowledge graph;

[0102] Step 211: Based on a pre-set fusion strategy, the target knowledge graph is fused with the target large model to recommend target academic resources, wherein the fusion strategy includes at least one of splicing, weighted summation and attention mechanism.

[0103] Specifically, knowledge graphs, with their powerful entity relationship representation capabilities, can clearly display the complex relationship network between academic resources and provide rich structured knowledge support for recommendation systems. Deep learning large models, especially pre-trained language models, with their excellent text understanding and generation capabilities, can deeply explore the semantic features of academic resources and improve the accuracy and personalization of recommendations.

[0104] like Figure 3As shown, an embodiment of the present invention provides an academic resource recommendation system based on a knowledge graph and a large model, the system comprising:

[0105] Acquisition module 301: used to acquire academic resources to be recommended based on at least one of academic databases, open source platforms and academic conference resources by using web crawler technology and at least one API interface;

[0106] The knowledge graph construction module 302 is used to determine the key entities and entity relationships in the academic resources to be recommended by using entity recognition and relationship extraction technology, and to construct a knowledge graph based on the key entities and the entity relationships;

[0107] A large model selection module 303 is used to determine a target large model based on the language understanding ability, generalization performance and training efficiency of the large model, wherein the target large model includes a multi-level attention mechanism and a fusion layer;

[0108] A training target determination module 304 is used to determine the training target of the target large model by using at least one of the cross entropy loss function and the BPR loss function and at least one evaluation index, wherein the target large model includes a user feedback mechanism, and the user feedback mechanism is used to adjust the parameters of the target large model;

[0109] Fusion module 305: used to fuse the knowledge graph with the target big model by integrating the key entity embedding vector and the entity relationship embedding vector in the knowledge graph with the text vector representation to recommend target academic resources.

[0110] based on Figure 3 An academic resource recommendation system based on knowledge graph and large model is shown in Figure 4 As shown, in the embodiment of the present invention, after the acquisition module 301 and before the knowledge graph construction module 302, it further includes: a data processing module 306 and a semantic determination module 307;

[0111] The data processing module 306 is used to perform data preprocessing operations on the acquired academic resources to be recommended to generate preprocessed academic resources, wherein the data preprocessing operations are used to perform data cleaning on the academic resources to be recommended, and the data cleaning includes: intelligent deduplication algorithm, standardized data format, missing value filling and noise removal operations based on statistics and rules;

[0112] The semantic determination module 307 is used to determine the text semantics and citation relationships in the preprocessed academic resources using natural language processing technology, wherein the natural language processing technology includes: BERT variants and TextRank optimization algorithms;

[0113] The knowledge graph construction module 302 is also used to use entity recognition and relationship extraction technology to determine the key entities and entity relationships in the preprocessed academic resources, and to construct a knowledge graph based on the key entities and the entity relationships.

[0114] based on Figure 4 An academic resource recommendation system based on knowledge graph and large model is shown in Figure 5 As shown, in this embodiment of the present invention, after the large model selection module 303 and before the training target determination module 304, it further includes:

[0115] The model optimization module 308 is used to optimize the parameters of the target large model based on the specific context of the academic resources and user needs, wherein the parameter optimization includes: adjusting the model architecture and optimizing the pre-training tasks.

[0116] based on Figure 5 An academic resource recommendation system based on knowledge graph and large model is shown in Figure 6 As shown, in the embodiment of the present invention, after the training target determination module 304 and before the fusion module 305, it further includes: a model evaluation module 309;

[0117] The model evaluation module 309 is used to perform:

[0118] Using at least one of a Bayesian optimization algorithm and a hyperparameter search algorithm, adjusting key hyperparameters of the target large model;

[0119] The academic resource recommendation performance of the target large model is evaluated by utilizing the user feedback mechanism and evaluation indicators, wherein the evaluation indicators include at least one of accuracy, recall, F1 score and ranking evaluation indicator NDCG.

[0120] like Figure 6 As shown, in this embodiment of the present invention, the fusion module 305 is further used to execute:

[0121] The key entity embedding vector and the entity relationship embedding vector in the knowledge graph are represented by a text vector to construct a target knowledge graph;

[0122] The target knowledge graph is fused with the target large model based on a preset fusion strategy to recommend target academic resources, wherein the fusion strategy includes at least one of splicing, weighted summation and attention mechanism.

[0123] It is understandable that the structure illustrated in the embodiment of the present invention does not constitute a specific limitation on an academic resource recommendation system based on a knowledge graph and a large model. In other embodiments of the present invention, an academic resource recommendation system based on a knowledge graph and a large model may include more or fewer components than shown in the figure, or combine certain components, or split certain components, or arrange the components differently. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0124] The information interaction, execution process and other contents between the units in the above-mentioned device are based on the same concept as the embodiment of the method of the present invention. For specific contents, please refer to the description in the embodiment of the method of the present invention, and no further description is given here.

[0125] The embodiment of the present invention also provides an academic resource recommendation system based on knowledge graph and large model, including: at least one memory and at least one processor;

[0126] at least one memory for storing a machine-readable program;

[0127] At least one processor is used to call a machine-readable program to execute an academic resource recommendation method based on a knowledge graph and a large model in any embodiment of the present invention.

[0128] An embodiment of the present invention also provides a computer-readable medium on which computer instructions are stored. When the computer instructions are executed by a processor, the processor executes an academic resource recommendation method based on a knowledge graph and a large model in any embodiment of the present invention.

[0129] Specifically, a system or device equipped with a storage medium can be provided, on which software program code that implements the functions of any of the above-mentioned embodiments is stored, and a computer (or CPU or MPU) of the system or device can be enabled to read and execute the program code stored in the storage medium.

[0130] In this case, the program code itself read from the storage medium can realize the function of any one of the above-mentioned embodiments, and thus the program code and the storage medium storing the program code constitute a part of the present invention.

[0131] The storage medium embodiments for providing the program code include a floppy disk, a hard disk, a magneto-optical disk, an optical disk (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), a magnetic tape, a non-volatile memory card, and a ROM. Alternatively, the program code can be downloaded from a server computer by a communication network.

[0132] In addition, it should be clear that the functions of any of the above embodiments can be implemented not only by executing the program code read by the computer, but also by enabling an operating system operating on the computer to complete part or all of the actual operations based on instructions from the program code.

[0133] In addition, it can be understood that the program code read from the storage medium is written to a memory provided in an expansion board inserted into the computer or written to a memory provided in an expansion unit connected to the computer, and then based on the instructions of the program code, a CPU installed on the expansion board or the expansion unit is enabled to perform part or all of the actual operations, thereby realizing the functions of any of the above-mentioned embodiments.

[0134] Each embodiment of the present invention has at least the following beneficial effects:

[0135] 1. In an embodiment of the present invention, a method for recommending academic resources based on knowledge graph and large model is provided. The method first obtains academic papers, data sets and scholar information in batches from multiple authoritative academic data sources (e.g., academic databases, open source platforms and academic conference resources) through advanced data collection methods (e.g., web crawler technology and at least one API interface) to wait for recommended academic resources, laying a solid foundation for subsequent steps. Then, the knowledge graph construction technology is used to identify the key entities and entity relationships in the academic resources to be recommended, and form a knowledge graph of the academic field with clear structure and rich content. On this basis, the present invention selects and customizes the deep learning large model, i.e., the target large model, based on the language comprehension ability, generalization performance and training efficiency of the large model. At the same time, a multi-level attention mechanism is innovatively introduced, which not only focuses on the key information inside the text, but also focuses on the correlation information between cross-text and cross-resources, so as to realize the deep mining of the complex relationship between academic resources. In addition, an efficient fusion layer is designed to realize the seamless connection and complementary enhancement of text information and structured knowledge. Then, the training target of the target large model is designed by using the loss function and multi-dimensional evaluation indicators. Through strategies such as task customization, training target design, and knowledge graph fusion, we have achieved a deep fusion of text information and structured knowledge, enhancing the ability of the big model to understand the complex relationships between academic resources, thereby improving the accuracy of academic resource recommendations;

[0136] 2. In an embodiment of the present invention, in order to recommend high-quality academic resources, it is necessary to perform data preprocessing operations on the collected academic resources to be recommended. In the data preprocessing stage, a strict data cleaning process is implemented, including intelligent deduplication algorithms, standardized data formats, missing value filling based on statistics and rules, and the use of machine learning technology to identify and eliminate noise data to ensure the purity and integrity of the data. In addition, by integrating advanced natural language processing technologies (such as BERT variants and TextRank optimization algorithms), the text semantics is deeply mined to achieve accurate extraction of keywords and detailed classification and annotation of academic resources, laying a solid data foundation for subsequent steps. Based on the high-quality data after preprocessing, advanced entity recognition and relationship extraction technologies can be used to accurately identify key entities (such as authors, papers, journals and institutions) and the complex relationships between them (such as writing, citations and collaboration) from academic papers, data sets and scholar information. On this basis, the powerful capabilities of the graph database are used to build a knowledge graph. At the same time, in order to ensure the timeliness and accuracy of the knowledge graph, an automated data update mechanism can be designed to track newly published papers and updated citation relationships in real time, etc., to achieve dynamic maintenance and continuous optimization of the knowledge graph;

[0137] 3. In an embodiment of the present invention, in view of the special needs of academic resource recommendation, the target large model is a basic large model (e.g., BERT series and GPT series) carefully selected from the current top deep learning large models by comprehensively considering the model's language comprehension ability, generalization performance and training efficiency. This model evaluates the performance of different models in the task of academic resource recommendation through comparative experiments, and determines the final large model to be used based on the experimental results. On this basis, in order to ensure that the large model can fully adapt to the specific context and needs of academic resources, it can be adjusted and optimized in a targeted manner (e.g., adjusting the model architecture, optimizing the pre-training tasks) to improve the performance of the large model in the recommendation task.

[0138] It should be noted that not all steps and modules in the above-mentioned processes and system structure diagrams are necessary, and some steps or modules can be ignored according to actual needs. The execution order of each step is not fixed and can be adjusted as needed. The system structure described in the above-mentioned embodiments can be a physical structure or a logical structure, that is, some modules may be implemented by the same physical entity, or some modules may be implemented by multiple physical entities, or some components in multiple independent devices may be implemented together.

[0139] In the above embodiments, the hardware unit can be realized by mechanical means or electrical means. For example, a hardware unit can include permanent dedicated circuits or logic (such as special processors, FPGA or ASIC) to complete the corresponding operation. The hardware unit can also include programmable logic or circuits (such as general-purpose processors or other programmable processors), which can be temporarily set by software to complete the corresponding operation. Concrete implementation (mechanical means or dedicated permanent circuits or temporarily set circuits) can be determined based on cost and time considerations.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The academic resource recommendation method based on knowledge graph and big model is characterized by: The method includes: A1: Using web crawler technology and at least one API interface, obtain academic resources to be recommended based on at least one of academic databases, open source platforms and academic conference resources; A2: Using entity recognition and relationship extraction technology, determine the key entities and entity relationships in the academic resources to be recommended, and construct a knowledge graph based on the key entities and the entity relationships; A3: Determine the target large model based on the language understanding ability, generalization performance and training efficiency of the large model, wherein the target large model includes a multi-level attention mechanism and a fusion layer; A4: Determine a training target of the target large model by using at least one of a cross entropy loss function and a BPR loss function and at least one evaluation index, wherein the target large model includes a user feedback mechanism, and the user feedback mechanism is used to adjust parameters of the target large model; A5: By integrating the key entity embedding vectors and the entity relationship embedding vectors in the knowledge graph with text vector representations, the knowledge graph is fused with the target big model to recommend target academic resources.

2. The method according to claim 1, characterized in that After A1 and before A2, further comprising: Performing data preprocessing operations on the acquired academic resources to be recommended to generate preprocessed academic resources, wherein the data preprocessing operations are used to perform data cleaning on the academic resources to be recommended, and the data cleaning includes: intelligent deduplication algorithm, standardized data format, missing value filling and noise removal operations based on statistics and rules; Determine the text semantics and citation relationships in the preprocessed academic resources using natural language processing technology, wherein the natural language processing technology includes: BERT variants and TextRank optimization algorithms; The A2 includes: Utilizing entity recognition and relationship extraction techniques, key entities and entity relationships in the preprocessed academic resources are determined, and a knowledge graph is constructed based on the key entities and the entity relationships.

3. The method according to claim 1, characterized in that After A3 and before A4, further comprising: Based on the specific context of academic resources and user needs, the parameters of the target large model are optimized, wherein the parameter optimization includes: adjusting the model architecture and optimizing the pre-training tasks.

4. The method according to claim 1, characterized in that After A4 and before A5, further comprising: Using at least one of a Bayesian optimization algorithm and a hyperparameter search algorithm, adjusting key hyperparameters of the target large model; Using the user feedback mechanism and evaluation indicators, the academic resource recommendation performance of the target large model is evaluated, wherein the evaluation indicators include: at least one of accuracy, recall, F1 score and ranking evaluation indicator NDCG; and / or, The A5 includes: The key entity embedding vector and the entity relationship embedding vector in the knowledge graph are represented by a text vector to construct a target knowledge graph; The target knowledge graph is fused with the target large model based on a preset fusion strategy to recommend target academic resources, wherein the fusion strategy includes at least one of splicing, weighted summation and attention mechanism.

5. The academic resource recommendation system based on knowledge graph and big model is characterized by: The system includes: Acquisition module: used to acquire academic resources to be recommended based on at least one of academic databases, open source platforms and academic conference resources by using web crawler technology and at least one API interface; A knowledge graph construction module is used to determine the key entities and entity relationships in the academic resources to be recommended by using entity recognition and relationship extraction technology, and to construct a knowledge graph based on the key entities and the entity relationships; Large model selection module: used to determine the target large model based on the language understanding ability, generalization performance and training efficiency of the large model, wherein the target large model includes a multi-level attention mechanism and a fusion layer; A training target determination module: used to determine the training target of the target large model by using at least one of the cross entropy loss function and the BPR loss function and at least one evaluation index, wherein the target large model includes a user feedback mechanism, and the user feedback mechanism is used to adjust the parameters of the target large model; Fusion module: used to fuse the knowledge graph with the target big model by integrating the key entity embedding vector and the entity relationship embedding vector in the knowledge graph with the text vector representation, so as to recommend target academic resources.

6. The system according to claim 5, characterized in that After the acquisition module and before the knowledge graph construction module, it further includes: a data processing module and a semantic determination module; The data processing module is used to perform data preprocessing operations on the acquired academic resources to be recommended to generate preprocessed academic resources, wherein the data preprocessing operations are used to perform data cleaning on the academic resources to be recommended, and the data cleaning includes: intelligent deduplication algorithm, standardized data format, missing value filling and noise removal operations based on statistics and rules; The semantic determination module is used to determine the text semantics and citation relationships in the preprocessed academic resources using natural language processing technology, wherein the natural language processing technology includes: BERT variants and TextRank optimization algorithms; The knowledge graph construction module is also used to use entity recognition and relationship extraction technology to determine the key entities and entity relationships in the preprocessed academic resources, and to construct a knowledge graph based on the key entities and the entity relationships.

7. The system according to claim 5, characterized in that After the large model selection module and before the training target determination module, it further includes: The model optimization module is used to optimize the parameters of the target large model based on the specific context of academic resources and user needs, wherein the parameter optimization includes: adjusting the model architecture and optimizing the pre-training tasks.

8. The system according to claim 5, characterized in that After the training target determination module and before the fusion module, it further includes: a model evaluation module; The model evaluation module is used to perform: Using at least one of a Bayesian optimization algorithm and a hyperparameter search algorithm, adjusting key hyperparameters of the target large model; Using the user feedback mechanism and evaluation indicators, the academic resource recommendation performance of the target large model is evaluated, wherein the evaluation indicators include: at least one of accuracy, recall, F1 score and ranking evaluation indicator NDCG; and / or, The fusion module is further used to execute: The key entity embedding vector and the entity relationship embedding vector in the knowledge graph are represented by a text vector to construct a target knowledge graph; The target knowledge graph is fused with the target large model based on a preset fusion strategy to recommend target academic resources, wherein the fusion strategy includes at least one of splicing, weighted summation and attention mechanism.

9. The academic resource recommendation system based on knowledge graph and big model is characterized by: include: at least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is configured to call the machine-readable program to execute the method according to any one of claims 1 to 4.

10. A computer readable medium, characterized in that The computer readable medium stores computer instructions, which, when executed by a processor, cause the processor to perform any one of the methods of claims 1 to 4.

Images