Library resource intelligent recommendation method and system based on big data analysis

Abstract

The invention provides a library resource intelligent recommendation method and system based on big data analysis, and relates to the technical field of library resource recommendation. A dynamic user borrowing intention model is constructed by mining behavior association logic, a user borrowing intention label set is generated, the user borrowing intention label set and library resource attribute information are linked to construct a scenarized resource adaptation module, and library collection resources are screened and sorted through the scenarized resource adaptation module to generate a dynamic resource recommendation sequence. User interaction behavior data is collected to iteratively optimize model parameters and mapping rules, and accurate library resource intelligent recommendation is achieved.

Description

Technical Field

[0001] This application relates to the field of library resource recommendation technology, and more specifically, to a method and system for intelligent recommendation of library resources based on big data analysis. Background Technology

[0002] In the traditional field of library resource recommendation, common recommendation methods are often rather crude. In the early days, they mainly relied on manual recommendations, with librarians manually selecting resources based on their experience and understanding of users' general needs. This method was inefficient and had limited coverage, making it difficult to meet the personalized needs of many users.

[0003] With the development of information technology, some libraries have begun to adopt recommendation methods based on simple keyword matching. These methods search and match user-input keywords within the resource database to recommend relevant resources. However, these methods rely too heavily on the accuracy of keywords and fail to deeply understand the user's true intentions and potential needs, often resulting in inaccurate recommendations. Furthermore, existing technologies fail to adequately consider the dynamic changes in user borrowing behavior and the contextual adaptation of resource attributes to user needs, leading to poor recommendation performance and an inability to effectively improve users' utilization efficiency of library resources. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of this application, embodiments of this application provide a method for intelligent recommendation of library resources based on big data analysis, the method comprising: The user borrowing behavior trajectory data is deconstructed and divided into user borrowing behavior unit sequences by time slice. The user borrowing behavior trajectory data includes the start time, return time, resource borrowing channel, resource reading progress and resource evaluation feedback information of the user borrowing library resources. The user borrowing behavior unit sequence is composed of the behavior units corresponding to a single complete borrowing behavior arranged in chronological order. Each behavior unit is associated with a unique user identifier and resource identifier. Based on the user borrowing behavior unit sequence mining behavior association logic and scenario mapping relationship, a dynamic user borrowing intention model is constructed. A user borrowing intention tag set is generated by associating behavior unit features. The user borrowing intention tag set corresponds to the user resource demand tendency under different borrowing scenarios and different reading purposes, and is supplemented and updated in real time with the user's new borrowing behavior. By linking user borrowing intent tags with library resource attribute information, a scenario-based resource adaptation module is constructed. The library resource attribute information includes resource theme, resource content dimension, resource applicable scenario, resource update cycle, resource reading difficulty, and resource author background information. A dynamic mapping rule library between needs and resources is established within the scenario-based resource adaptation module. The library’s collection resources are filtered and sorted through a scenario-based resource adaptation module to generate a dynamic resource recommendation sequence. The dynamic resource recommendation sequence is dynamically adjusted in order and content based on users’ real-time borrowing behavior, resource inventory status and resource update status. Collect user interaction data on dynamic resource recommendation sequences, iteratively optimize the parameters of the dynamic user borrowing intent model and the mapping rules of the scenario-based resource adaptation module based on the interaction data, and generate model and system optimization results.

[0005] In another aspect, embodiments of this application also provide a library resource intelligent recommendation system based on big data analysis, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0006] Based on the above, this invention deconstructs user borrowing behavior trajectory data to deeply explore the underlying logic and scenario mapping relationships of user behavior, constructing a dynamic user borrowing intent model. This model can accurately capture users' resource demand tendencies under different scenarios and reading purposes, and it can be updated in real time with new user borrowing behavior, ensuring that the understanding of user intent is always up-to-date. A scenario-based resource adaptation module, constructed by linking user borrowing intent tag sets with library resource attribute information, establishes a dynamic mapping rule base between needs and resources, achieving scenario-based and precise adaptation of resource recommendations. By collecting user interaction behavior data on recommendation sequences and iteratively optimizing model parameters and mapping rules, the accuracy and adaptability of recommendations are further improved, significantly enhancing the precision and effectiveness of library resource recommendations, and effectively improving users' experience and satisfaction with library resources. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the execution flow of the intelligent library resource recommendation method based on big data analysis provided in the embodiments of this application.

[0008] Figure 2 This is a schematic diagram of the hardware architecture of the intelligent recommendation system for library resources based on big data analysis provided in the embodiments of this application. Detailed Implementation

[0009] The present application will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating an embodiment of the intelligent library resource recommendation method based on big data analysis provided in this application. The following is a detailed description of this intelligent library resource recommendation method based on big data analysis.

[0010] Step S110: Deconstruct the user borrowing behavior trajectory data and split it into user borrowing behavior unit sequences according to time slices. The user borrowing behavior trajectory data includes the start time, return time, resource borrowing channel, resource reading progress, and resource evaluation feedback information of the user borrowing library resources. The user borrowing behavior unit sequence is composed of the behavior units corresponding to a single complete borrowing behavior arranged in chronological order. Each behavior unit is associated with a unique user identifier and resource identifier.

[0011] In this embodiment, the application scenario is a university library's user base, involving a large amount of user borrowing behavior data. When deconstructing the user borrowing behavior trajectory data, the source and scope of the data are first clarified. The user borrowing behavior trajectory data comes from the library's borrowing management system, which records relevant information for each borrowing behavior. The start and return times are accurate to the year, month, day, hour, minute, and second to accurately define the time interval of each borrowing behavior. Resource borrowing channels include online borrowing platforms, library self-service borrowing machines, and manual borrowing windows, and information about these channels is recorded in detail. Resource reading progress is obtained through users' reading records on online reading platforms or reading reports submitted after borrowing. For example, when users read online, the reading time and number of times each page is read are recorded to calculate the approximate reading progress. Resource evaluation feedback information comes from the ratings and text comments submitted by users through the library's evaluation system after borrowing.

[0012] When segmenting user borrowing behavior data into time-based sequences, each complete borrowing transaction is considered a single behavior unit. The start time, return time, borrowing channel, reading progress, and feedback information for each borrowing transaction are extracted from the user's borrowing behavior trajectory data. This information is then integrated into a single behavior unit. All behavior units for the same user are arranged chronologically to form a sequence of user borrowing behavior units. Each behavior unit is associated with a unique user identifier, such as the user's student ID or library card number, and a resource identifier, such as the ISBN of a book or the ISSN of a journal, to ensure accurate identification of the user and the corresponding borrowed resource.

[0013] Step S120: Based on the user borrowing behavior unit sequence, mine the behavior association logic and scenario mapping relationship, construct a dynamic user borrowing intention model, and generate a user borrowing intention tag set through behavior unit feature association. The user borrowing intention tag set corresponds to the user resource demand tendency under different borrowing scenarios and different reading purposes, and is updated in real time with the user's new borrowing behavior.

[0014] Step S121: Extract the core behavioral features of each behavioral unit in the user borrowing behavior unit sequence. The core behavioral features include the borrowing resource category, borrowing duration, reading progress completion ratio, re-borrowing frequency, resource evaluation level, and borrowing time period distribution. Classify and integrate them according to user identifier to form a user behavior unit feature set, with each user corresponding to an independent feature subset.

[0015] In the aforementioned university library application scenario, when extracting the core behavioral characteristics of each behavioral unit, a detailed feature analysis is conducted for each unit. Borrowed resource categories are divided according to the library's classification system, such as literature, science and engineering, history, etc., each with a corresponding classification code obtained by parsing the resource identifier. Borrowing duration is calculated by the time difference between the return time and the start time, accurate to the hour. Reading progress completion rate is calculated based on the user's reading progress information, such as the percentage of total pages read. Re-borrowing frequency is calculated by tracking the number of times a user borrows the same resource within a certain period. Resource rating is converted from user-submitted ratings into specific levels, such as a five-star rating system. Borrowing time distribution divides the day into different time periods, such as morning, afternoon, and evening, and tracks the number of times a user borrows during each time period.

[0016] When classifying and integrating user behavior unit feature sets by user identifier, the core behavioral features of all behavioral units under the same user identifier are extracted and arranged in a predetermined order to form an independent feature subset for that user. The feature subsets of different users are independent of each other and stored separately for subsequent individual analysis and processing of each user's behavioral features.

[0017] Step S122: Perform correlation analysis on the behavioral unit features in the user behavior unit feature set in chronological order, explore the feature inheritance logic and mutation rules between adjacent behavioral units, extract the continuous trend of user borrowing behavior over time, and generate user behavior correlation trend data, which includes feature correlation strength and change node information.

[0018] Step S1221: Reorder the behavioral unit features in the user behavior unit feature set according to time sequence to form a time-series user behavior feature sequence, wherein the time-series user behavior feature sequence contains the timestamp and complete feature information of each behavioral unit.

[0019] In the aforementioned university library scenario, a set of user behavior unit features is obtained. These features may be stored in the order of data collection, not necessarily strictly in chronological order. The features are then reordered chronologically, arranged from earliest to latest based on the start timestamp of each behavior unit.

[0020] In the sorted time-series user behavior feature sequence, each behavior unit contains its corresponding timestamp and complete feature information, such as the type of borrowed resource, borrowing duration, and reading progress completion rate.

[0021] Step S1222: Extract the core behavioral features of two adjacent behavioral units in the time-series user behavior feature sequence, compare the differences and commonalities of the features one by one, calculate the difference values ​​of borrowing resource category, borrowing duration, reading progress completion ratio and borrowing time period, and compare them with the preset feature change threshold to generate an adjacent behavioral feature comparison dataset.

[0022] In this embodiment, two adjacent behavioral units are extracted sequentially from the time-series user behavior feature sequence. For each pair of adjacent behavioral units, their core behavioral features are extracted, such as the type of borrowed resource, borrowing duration, reading progress completion rate, and borrowing time period.

[0023] The differences and commonalities of the aforementioned core behavioral features are compared one by one. For numerical features, such as borrowing duration and reading progress completion rate, their difference values ​​are calculated, which are the feature values ​​of the later behavioral units minus the feature values ​​of the earlier behavioral units. For categorical features, such as borrowing resource categories and borrowing time periods, they are converted into comparable numerical values ​​through encoding before calculating the difference values. The calculated difference values ​​are compared with a preset feature change threshold. If the difference value exceeds the threshold, the feature is considered to have changed significantly. The comparison results, including the difference values, whether the threshold is exceeded, and common features, are compiled to generate an adjacent behavioral feature comparison dataset.

[0024] Step S1223: Statistically analyze the frequency of occurrence of the same resource category in adjacent behavioral units in the time-series user behavior feature sequence, analyze the continuous cycle, interval time and repeated borrowing number of the same type of resource, explore the continuous pattern of borrowing of the same type of resource, and generate a continuous borrowing dataset of the same type of resource.

[0025] In the aforementioned university library scenario, the time-series user behavior feature sequence is traversed to count the frequency of occurrence of the same resource category in adjacent behavior units. For example, the frequency of borrowing computer science books twice consecutively, the frequency of borrowing literature books three times consecutively, etc.

[0026] This analysis examines the continuous borrowing cycle of similar resources, i.e., the duration of consecutive borrowing of the same type of resource; the interval time, i.e., the time interval between two borrowings of the same type of resource; and the number of repeated borrowings, i.e., the number of times a user borrows the same type of resource within a certain period. By analyzing this data, the analysis uncovers continuous patterns in the borrowing of similar resources, such as whether users concentrate their borrowing of a certain type of resource within a certain time period, and whether there are regular patterns in the interval time between two borrowings of the same type of resource. The results of this analysis are then compiled to generate a dataset of continuous borrowing of similar resources.

[0027] Step S1224: Track the conversion of different resource categories in adjacent behavioral units in the time-series user behavior feature sequence, analyze the triggering factors of category conversion, capture the borrowing trend of users switching from one type of resource to another, and generate a resource category conversion dataset, which includes the conversion direction and conversion frequency.

[0028] In this embodiment, the resource categories of adjacent behavioral units in a time-series user behavior feature sequence are tracked. When two adjacent behavioral units have different resource categories, the direction of the category transition is recorded, i.e., from which resource category to which resource category, as well as the time of the transition and the corresponding behavioral unit features.

[0029] Analyze the triggering factors for category switching, considering factors such as the user's borrowing context, reading purpose, and external environment, including whether the user is taking exams or whether relevant courses are offered, to determine the reasons for the category switching. Capture borrowing trends from one type of resource to another; for example, if a user first borrows multiple math books and then starts borrowing physics books, it may indicate a shift in the user's learning focus. Compile information such as the direction of the switch, the frequency of the switch, and the triggering factors to generate a resource category switching dataset.

[0030] Step S1225: Integrate adjacent behavior feature comparison datasets, continuous borrowing datasets of similar resources, and resource category conversion datasets to construct a user behavior association analysis model. The user behavior association analysis model embeds behavior feature association algorithms and trend recognition logic.

[0031] In the aforementioned university library scenario, adjacent behavioral feature comparison datasets, continuous borrowing datasets of similar resources, and resource category conversion datasets are integrated as the data foundation for constructing a user behavior correlation analysis model.

[0032] The user behavior correlation analysis model is built using machine learning algorithms, embedding behavioral feature correlation algorithms and trend recognition logic. The behavioral feature correlation algorithm analyzes the degree of correlation between different behavioral features, such as the correlation between borrowing duration and the reading progress completion rate; the trend recognition logic identifies changing trends in user borrowing behavior, such as resource category conversion trends and borrowing frequency change trends. By integrating multiple datasets and algorithmic logic, a model capable of comprehensively analyzing user behavior correlations is constructed.

[0033] Step S1226: Extract the continuous change features of user borrowing behavior through the user behavior correlation analysis model, identify the key time nodes where the behavior features change significantly, record the differences in behavior features before and after each node, and generate a dataset of behavior change nodes.

[0034] In this embodiment, a time-series user behavior feature sequence is input into the user behavior correlation analysis model. The model extracts continuous change features of user borrowing behavior through behavioral feature correlation algorithms and trend recognition logic, such as the change curve of borrowed resource categories over time and the change trend of borrowing duration.

[0035] Identify key time points where behavioral characteristics change significantly. These key time points typically correspond to major shifts in user borrowing behavior, such as switching from borrowing textbooks to borrowing research reference books. Record the differences in behavioral characteristics before and after each key time point, such as changes in resource type, borrowing duration, and completion rate of reading progress. Organize the above node information and difference data to generate a behavioral change node dataset.

[0036] Step S1227: Analyze the differences in user behavior characteristics before and after the behavior change node, and combine the user borrowing scenario, resource attributes and reading purpose in the corresponding time period to analyze the core factors that lead to the change in behavior characteristics and generate a dataset of behavioral change drivers.

[0037] In this embodiment, for each key time point in the behavior change node dataset, the differences in user behavior characteristics before and after the node are analyzed. This is combined with information such as the user's borrowing scenario for the corresponding time period, such as whether it was the beginning of the semester, exam week, or holiday; resource attributes, such as the topic and difficulty of the borrowed resources; and reading purpose, such as course learning, research projects, or leisure and entertainment.

[0038] Based on the above factors, we analyze the core factors leading to changes in behavioral characteristics. For example, the increased proportion of users borrowing textbooks during exam week may be due to exam preparation needs; the increased proportion of users borrowing literary novels during holidays may be for leisure reading. By compiling these core factors and their analytical basis, we generate a dataset of behavioral change drivers.

[0039] Step S1228: Based on the behavior change node dataset and the behavior change driver dataset, draw a user borrowing behavior trend curve. The user borrowing behavior trend curve shows the change trajectory and key node positions of different feature dimensions, and generates behavior trend curve data.

[0040] In the aforementioned university library scenario, user borrowing behavior trend curves were plotted based on the datasets of behavior change nodes and the datasets of behavior change drivers. Different feature dimensions were selected, such as borrowed resource type, borrowing duration, and reading progress completion rate, and the change trajectory curves for each feature dimension were plotted with time on the horizontal axis and feature values ​​on the vertical axis.

[0041] Mark key nodes on the trend curve, indicating the time points when significant changes in behavioral characteristics occur, and indicate the corresponding driving forces behind these changes at each node. This trend curve visually demonstrates the changes and key nodes in user borrowing behavior across different feature dimensions. Organize the above curve data, including coordinate information, node positions, and driving force annotations, to generate behavioral trend curve data.

[0042] Step S1229: Integrate behavioral trend curve data with various correlation analysis datasets, supplement the logical basis and motivation of behavioral correlation, add correlation type identifiers and confidence parameters to behavioral correlation data, and generate preliminary user behavior correlation trend data.

[0043] In this embodiment, behavioral trend curve data is integrated with various correlation analysis datasets, such as adjacent behavioral feature comparison datasets, continuous borrowing datasets of similar resources, and resource category conversion datasets. Logical basis is added to the behavioral correlation data, such as whether a certain behavioral correlation is based on significant changes in adjacent behavioral features or on continuous borrowing patterns of similar resources; a driving force explanation is added, i.e., the reason for the behavioral correlation, such as changes in the user's learning needs or shifts in interests.

[0044] Add association type identifiers to the behavioral association data, such as feature inheritance association, category conversion association, and continuous borrowing association; add confidence parameters, assigning a certain confidence value to each behavioral association based on the sufficiency of the evidence and the strength of the data support. The higher the confidence value, the more reliable the behavioral association. Through the above integration and supplementation, preliminary user behavior association trend data is generated.

[0045] Step S12210: Classify and organize the preliminary user behavior correlation trend data according to scenario category and behavior change pattern to form the final user behavior correlation trend data.

[0046] In this embodiment, the preliminary user behavior correlation trend data is classified according to scenario categories, such as "course learning scenario," "scientific research scenario," and "leisure reading scenario." Within each scenario category, it is further subdivided according to behavior change patterns, such as continuous borrowing patterns, category conversion patterns, and feature change patterns.

[0047] The categorized and organized data is checked and verified to ensure its accuracy and consistency. Data under the same scenario category and behavioral change patterns are integrated to form the final user behavior correlation trend data, which can then be matched and analyzed with the scenario category classification results.

[0048] Step S123: Based on the time period distribution of user borrowing behavior, the completion rate of reading progress and the resource category, classify user borrowing scenario types, and each user borrowing scenario type has a corresponding behavioral feature identifier.

[0049] In the aforementioned university library scenario, user borrowing scenarios are categorized based on the time period distribution of user borrowing behavior, the completion rate of reading progress, and the type of resources. First, the time period distribution is analyzed, dividing user borrowing times into different combinations such as weekday daytime, weekday evening, weekend daytime, and weekend evening. Then, the completion rate of reading progress is considered, categorized as high completion rate (above 80%), medium completion rate (50% to 80%), and low completion rate (below 50%). Finally, the type of resources is taken into account, such as textbooks, professional reference books, literary novels, and popular science books.

[0050] By analyzing the combination of these three factors, different user borrowing scenario types are identified. For example, borrowing textbooks on weekday evenings with a high completion rate may correspond to a "course learning scenario," while borrowing literary novels on weekend days with a moderate completion rate may correspond to a "leisure reading scenario." Each user borrowing scenario type is assigned a unique behavioral characteristic identifier, which is a combination of a time period distribution code, a reading progress completion rate code, and a resource category code, to quickly identify and distinguish different borrowing scenario types.

[0051] Step S124: Perform bidirectional matching between user behavior correlation trend data and scene category classification results to establish a unique correspondence between behavioral feature combinations and scene categories, and generate a user behavior scene correspondence table. The scene category to which each behavioral feature combination belongs and the basis for association are determined in the user behavior scene correspondence table.

[0052] In this embodiment, user behavior trend data is bidirectionally matched with scenario category classification results. On one hand, different combinations of behavioral features are extracted from the user behavior trend data, such as specific borrowing resource category sequences, borrowing duration trends, and changes in reading progress completion rates. On the other hand, the behavioral feature conditions corresponding to each scenario category in the scenario category classification results are compiled.

[0053] Then, the behavioral feature combinations are compared one by one with the behavioral feature conditions of the scenario categories to find matching combinations and categories, establishing a unique correspondence. For example, a behavioral feature combination might be characterized by continuously borrowing multiple professional reference books, primarily during weekday evenings, with a high completion rate of reading progress; this combination likely corresponds to a "professional research scenario." The generated user behavior scenario correspondence table records in detail the scenario category to which each behavioral feature combination belongs, as well as the basis for the above association, such as which features in the behavioral feature combination meet the behavioral feature conditions of the scenario category.

[0054] Step S125: Based on the user behavior scenario mapping table, map and generate initial user borrowing intent tags to form a set of initial user borrowing intent tags. Each initial user borrowing intent tag corresponds to a core resource demand tendency under a user borrowing scenario type. The tag content of the initial user borrowing intent tag is associated with specific resource attribute features.

[0055] In the aforementioned university library scenario, initial user borrowing intent tags are generated based on a user behavior scenario mapping table. For each scenario category in the user behavior scenario mapping table, the core resource needs of users under that scenario category are analyzed. For example, in the "course learning scenario," the core resource needs of users might be to obtain course-related textbooks and supplementary materials, so the corresponding initial user borrowing intent tag could be "course learning needs."

[0056] Each initial user borrowing intent tag is associated with specific resource attribute characteristics. For example, the resource attribute characteristics associated with the "course learning needs" tag might include textbook version, course name, and related knowledge points. All generated initial user borrowing intent tags are collected to form a user initial borrowing intent tag set.

[0057] Step S126: Integrate the user's initial borrowing intent tag set with the user's historical borrowing behavior data, supplement the behavioral support cases corresponding to each initial user borrowing intent tag, increase the descriptive dimensions and scope of application of the initial user borrowing intent tags, enhance the correlation between the initial user borrowing intent tags and user behavior, and generate a user reinforced borrowing intent tag set.

[0058] In this embodiment, the initial borrowing intent tag set of users is integrated with the user's historical borrowing behavior data. Behavioral cases related to each initial user borrowing intent tag are retrieved from the user's historical borrowing behavior data, such as which specific textbooks and supplementary materials the user borrowed in the "course learning scenario," and information such as the start time and reading progress of these borrowing behaviors.

[0059] The aforementioned behavioral support cases are supplemented into the corresponding initial user borrowing intent tags to increase the descriptive dimensions of the tags. For example, they not only describe the resource demand tendency corresponding to the tag but also provide specific examples of borrowing behavior. Simultaneously, based on the diversity of behavioral support cases, the applicability of the initial user borrowing intent tags is expanded. For instance, the "Course Learning Needs" tag, originally only applicable to a specific course, can be made applicable to more course learning scenarios by supplementing borrowing cases from different courses. Through this integration, the correlation between initial user borrowing intent tags and user behavior is enhanced, generating a set of reinforced user borrowing intent tags.

[0060] Step S127: Remove duplicate tag content and semantically conflicting tag entries from the user's enhanced borrowing intent tag set, and merge semantically similar borrowing intent tags to form a user's standardized borrowing intent tag set.

[0061] In the aforementioned university library scenario, the set of tags representing users' borrowing intentions is processed. First, it checks for duplicate tags—that is, two or more tags with identical text descriptions—and deletes the duplicates, keeping only one. Then, it identifies semantically conflicting tags, such as one tag stating "likes reading science fiction" and another stating "dislikes reading science fiction." For these conflicting tags, further analysis of users' actual behavioral data is needed to determine which tag better reflects the user's true intention. The correct tag is retained, and the conflicting tag is deleted.

[0062] For borrowing intent tags with similar semantics, such as "research material access needs" and "academic research resource needs," they are merged into a more general and accurate tag, such as "scientific research needs." Through the above processing, a standardized set of user borrowing intent tags is formed, ensuring the uniqueness, accuracy, and consistency of the tags.

[0063] Step S128: Using the set of user-defined borrowing intent tags as the training target, input the set of user behavior unit features into the dynamic user borrowing intent model for training, optimize the feature extraction logic and intent mapping algorithm of the dynamic user borrowing intent model, optimize the ability of the dynamic user borrowing intent model to identify potential user borrowing intent, and generate the trained dynamic user borrowing intent model.

[0064] In this embodiment, training data is prepared using a set of user borrowing intent tags as the training target. The data in the user behavior unit feature set is divided into training, validation, and test sets according to a predetermined ratio. The dynamic user borrowing intent model adopts a deep learning architecture, including an input layer, hidden layers, and an output layer. The input layer receives feature data from the user behavior unit feature set; the hidden layer consists of multiple neurons that process the input data through activation functions; and the output layer outputs the predicted results of the user borrowing intent tags.

[0065] During training, user behavior unit features from the training set are input into the model. The model makes predictions based on the current feature extraction logic and intent mapping algorithm, obtaining predicted borrowing intent labels. The predicted results are compared with the actual labels in the user's standardized borrowing intent label set, and the loss function value is calculated. Based on the loss function value, the backpropagation algorithm is used to adjust the model parameters and optimize the feature extraction logic and intent mapping algorithm. Through multiple iterations of training, the model's performance is monitored using a validation set. Training stops when the model's accuracy on the validation set no longer improves. Finally, the model is evaluated using a test set to ensure that its ability to identify potential user borrowing intentions meets expectations, generating a completed dynamic user borrowing intent model.

[0066] Step S129: Mine the hidden borrowing intentions in the user behavior unit sequence through the trained dynamic user borrowing intention model, supplement the potential demand tags not covered by the initial user borrowing intention tags, and form a complete user borrowing intention tag set.

[0067] In the aforementioned university library scenario, user behavior unit sequences are input into a trained dynamic user borrowing intent model. The model analyzes these sequences, extracting features and mapping intent to uncover hidden borrowing intentions within user behavior. For example, a user might frequently borrow different books in a particular field, but the initial user borrowing intent labels might not have covered in-depth research needs in that field; the model can identify this potential need.

[0068] The discovered potential demand tags are added to the user's standardized borrowing intent tag set to form a complete user borrowing intent tag set. These potential demand tags can more comprehensively reflect the user's borrowing intent and improve the accuracy of the recommendation system.

[0069] Step S1210: Sort the complete set of user borrowing intent tags according to the scenario category and the correlation strength between the tags and user behavior to form the final set of user borrowing intent tags.

[0070] In this embodiment, the complete set of user borrowing intent tags is grouped according to scenario categories, such as classifying tags belonging to different scenario categories like "course learning scenario," "scientific research scenario," and "leisure reading scenario." Then, within each scenario category, tags are sorted according to the strength of their association with user behavior. The association strength is calculated by analyzing factors such as the number of supporting behavioral cases for the tag and the significance of behavioral features; tags with higher association strength rank higher in the sorting.

[0071] The ranking results from different scenario categories are integrated according to a set scenario priority to form the final set of user borrowing intent tags. This ranking method allows the recommendation system to prioritize users' high-relevance needs in key scenarios, improving the effectiveness of recommendations.

[0072] Step S130: Link the user's borrowing intent tag set with the library resource attribute information to build a scenario-based resource adaptation module. The library resource attribute information includes resource theme, resource content dimension, resource applicable scenario, resource update cycle, resource reading difficulty and resource author background information. The scenario-based resource adaptation module establishes a dynamic mapping rule library between needs and resources.

[0073] Step S131: Analyze the user's borrowing intent tag set, extract the core demand features and extended demand features corresponding to each borrowing intent tag, analyze the user's resource demand focus and potential demand direction in different scenarios, and generate a user demand feature list. The user demand feature list is organized by scenario category.

[0074] In this embodiment, the user's borrowing intent tag set is parsed. For each borrowing intent tag, its textual description and associated behavioral support cases are analyzed to extract core demand features, that is, the user's main demand for resources directly expressed by the tag. For example, the core demand feature of "course learning demand" is to obtain course-related textbooks. At the same time, extended demand features are mined, that is, potential needs derived from the core demand features, such as the need for related tutoring materials, exercise books, etc., in addition to textbooks.

[0075] The analysis identifies key user resource needs across different scenarios. For example, in a "scientific research and academic scenario," the key needs are cutting-edge research findings and professional data; potential needs include interdisciplinary resources in related fields. These needs are then categorized and organized by scenario type to generate a list of user needs characteristics. Each scenario category in the list includes detailed descriptions of core and extended needs characteristics.

[0076] Step S132: Classify the library resource attribute information by resource theme, resource content dimension, resource applicable scenario, resource update cycle, resource reading difficulty and resource author background information, and supplement the resource collection location, inventory quantity and borrowing record information to form a library resource attribute classification dataset.

[0077] In the aforementioned university library scenario, the resource attribute information is categorized. Based on resource theme, resources are divided into broad categories such as computer science, literature and art, and history and culture, with each category further subdivided. Based on content, resources are categorized into theoretical, applied, and case-based resources. Based on applicable scenarios, resources are categorized into teaching, research, and general reading resources. Based on update frequency, resources are categorized into high-frequency, medium-frequency, and low-frequency updates. Based on reading difficulty, resources are categorized into beginner, intermediate, and expert levels. Based on author background information, resources are categorized into academic authors, industry expert authors, and amateur authors.

[0078] Supplementing this data includes the location information of the resources, such as the specific floor and shelf number in the library; inventory information, i.e., the number of resources currently available for borrowing; and borrowing record information, including the number of borrowings in the past and the characteristics of the borrowers. Integrating this categorized and supplemented information creates a library resource attribute classification dataset, which can then be matched with user needs and characteristics.

[0079] Step S133: Establish a bidirectional mapping relationship between the user demand feature list and the library resource attribute classification dataset, associate and match each demand feature with the corresponding resource attribute, determine the association logic between the demand feature and the resource attribute, and generate a user demand resource association table, which contains association strength information.

[0080] Step S1331: Extract each requirement feature from the user requirement feature list, break it down into multiple specific requirement elements, determine the core and extended requirements of each requirement feature, supplement the behavioral support basis corresponding to the requirement elements, and generate the user requirement element decomposition results.

[0081] In the aforementioned university library scenario, each demand feature in the user demand feature list is extracted. For each demand feature, multiple specific demand elements are extracted from its description and associated user borrowing intent tags. For example, the "course learning demand" feature can be broken down into demand elements such as "obtaining the course textbook," "obtaining after-class tutoring materials," and "obtaining relevant exercise sets."

[0082] Identify the core demand for each demand feature, i.e., the most important and direct demand element; and extended demands, i.e., other demand elements derived from the core demand. Supplement the behavioral supporting evidence for each demand element, such as whether there are relevant borrowing records in the user's borrowing history, or whether there are relevant demand expressions in the user's evaluation feedback, etc. Organize the above decomposition results to generate the user demand element decomposition results.

[0083] Step S1332: Refine and break down each library resource attribute in the library resource attribute classification dataset to form multiple refined resource attribute elements, supplement the specific description and feature identifier of each attribute element, and generate a list of library resource attribute elements, which is organized according to resource category.

[0084] In this embodiment, each library resource attribute in the library resource attribute classification dataset is further subdivided. For example, the "resource theme" attribute can be further subdivided into attribute elements such as "primary theme", "secondary theme", and "keywords"; the "resource content dimension" can be further subdivided into attribute elements such as "theoretical knowledge", "practical cases", and "experimental guidance".

[0085] Supplement each attribute element with a detailed description, such as the specific name of the "primary theme" and the specific words for "keywords"; add feature identifiers to each attribute element, such as using specific codes to represent different attribute elements. Organize the refined resource attribute elements according to resource categories to generate a list of library resource attribute elements, facilitating subsequent matching with user needs.

[0086] Step S1333: Compare the user demand element breakdown results with the library resource attribute element list one by one, match elements with similar semantics, logical connections and feature fits, establish a two-way correspondence relationship at the element level, record the association logic of each correspondence relationship, and generate an element-level association dataset.

[0087] In the aforementioned university library scenario, each demand element in the user demand element breakdown is compared with each resource attribute element in the library resource attribute element list. Natural language processing technology and semantic analysis algorithms are used to determine whether the demand elements and resource attribute elements are semantically similar; the logical relationships between them are analyzed, such as whether causal relationships or inclusion relationships exist; and feature matching is checked, such as whether the feature values ​​of the demand elements match the feature values ​​of the resource attribute elements.

[0088] For elements that are semantically similar, logically related, and feature-matched, establish a two-way correspondence at the element level, meaning that demand elements correspond to resource attribute elements, and vice versa. Record the association logic for each correspondence, such as whether it is based on semantic similarity, logical inclusion, or feature value matching. Organize the above correspondences and association logic to generate an element-level association dataset.

[0089] Step S1334: Based on the feature-level association dataset, calculate the association strength between each demand feature and its corresponding resource attribute. The association strength is determined by the number of matching features, semantic similarity, and logical correlation, and generate a demand-resource association strength dataset.

[0090] In this embodiment, based on the feature-level association dataset, the association strength between each demand feature and its corresponding resource attribute is calculated. The calculation of association strength considers the number of element matches, i.e., the number of successful matches between demand elements decomposed from demand features and resource attribute elements refined from resource attributes; semantic similarity, i.e., the degree of semantic similarity between successfully matched elements; and logical relevance, i.e., the degree of logical relevance between successfully matched elements.

[0091] Weights are assigned to each of these three factors. Then, the values ​​of element matching quantity, semantic similarity, and logical relevance are multiplied by their respective weights and summed to obtain a comprehensive value of association strength. The comprehensive values ​​of association strength between each demand feature and resource attribute are then compiled to generate a demand-resource association strength dataset.

[0092] Step S1335: Sort the matching relationship between demand features and resource attributes according to the correlation strength value to generate a priority matching dataset.

[0093] In this embodiment, the matching relationships between demand features and resource attributes are sorted according to their correlation strength values. The matching relationships are arranged from highest to lowest correlation strength, with higher correlation strength values ​​appearing earlier in the ranking. For matching relationships with the same correlation strength value, they are sorted by the number of matching elements, with those having more matching elements appearing first; if the number of matching elements is also the same, they are sorted by semantic similarity. The sorted matching relationships are then compiled to generate a priority matching dataset, so that matching relationships with high correlation strength are given priority during resource matching.

[0094] Step S1336: Integrate the priority matching dataset and the feature-level association dataset, supplement the logical support description for each matching relationship, describe the association basis and adaptation scenarios of demand features and resource attributes, and generate a preliminary user demand resource association table.

[0095] In the aforementioned university library scenario, a priority matching dataset and a feature-level association dataset are integrated. Matching relationships in the priority matching dataset are associated with corresponding feature-level association relationships in the feature-level association dataset, and logical supporting explanations are provided for each matching relationship.

[0096] The logical support description explains the basis for the association between demand characteristics and resource attributes, such as which elements are used for matching and the calculation process of the association strength; the applicable scenario, that is, which user borrowing scenario this matching relationship applies to. Organize the above information into a table to generate a preliminary user demand-resource association table, which includes columns for demand characteristics, resource attributes, association strength, association basis, and applicable scenario.

[0097] Step S1337: Remove duplicate matching relationships and invalid association entries from the preliminary user requirement resource association table, and standardize the data format and expression specifications to generate a standardized user requirement resource association table.

[0098] In this embodiment, the initial user requirement resource association table is processed. The table is checked for duplicate matching relationships, i.e., entries with the same requirement characteristics and resource attributes. If duplicate entries are found, they are deleted, leaving only one. Invalid association entries are identified, such as those with zero association strength, no logical supporting explanation, or unclear applicable scenarios, and are removed.

[0099] Standardize data formats, such as retaining the same number of decimal places for association strength values ​​and using consistent date formats; standardize expression standards, such as using consistent names for requirement features and resource attributes, and maintaining a consistent language style for logical support descriptions. After these processes, a standardized user requirement resource association table is generated.

[0100] Step S1338: Supplement the historical adaptation data of library resources and user borrowing feedback data. Using the historical borrowing data and user rating data of library resources, calculate the historical borrowing frequency and average rating corresponding to each matching relationship in the standardized user demand resource association table. Mark the matching relationships with frequencies and ratings below the corresponding thresholds as relationships to be verified, adjust the association strength values, correct unreasonable matching relationships, and generate a verified user demand resource association table.

[0101] In this embodiment, supplementary historical adaptation data of library resources are included, namely, the number of times the resource has been recommended to users and the number of times it has been accepted by users in the past; user borrowing feedback data includes users' borrowing records and rating data for the resource.

[0102] Historical borrowing data from library resources is used to calculate the historical borrowing frequency corresponding to each matching relationship in the user demand resource association table, i.e., the number of times the resource under that matching relationship has been borrowed by the user; average rating is calculated using user rating data. The historical borrowing frequency is compared with a preset frequency threshold, and the average rating is compared with a preset rating threshold. Matching relationships where both the frequency and rating are lower than the corresponding thresholds are marked as relationships to be verified.

[0103] For relationships to be verified, their association basis and logical support are re-examined, and the association strength value is adjusted, such as by reducing the association strength. For obviously unreasonable matching relationships, such as those with extremely low historical borrowing frequency and ratings, they are directly corrected or deleted. Through the above processing, a resource association table for verifying user needs is generated.

[0104] Step S1339: Classify and organize the verification user demand resource association table according to user demand feature categories and library resource attribute categories, adjust the layout structure and query logic of the user demand resource association table, and generate a categorized user demand resource association table.

[0105] In the aforementioned university library scenario, the resource association table for verifying user needs is categorized according to user need characteristics, such as "course learning needs," "scientific research needs," and "leisure reading needs." Within each need characteristic category, it is further subdivided according to library resource attribute categories, such as "textbook resources," "professional reference book resources," and "literary fiction resources."

[0106] The layout of the association tables was adjusted to group matching relationships of the same requirement feature category and resource attribute category together for easier viewing and management. The query logic was optimized by creating indexes for requirement feature categories and resource attribute categories, enabling quick retrieval of matching relationships under specific categories. These adjustments generated a categorized user requirement resource association table.

[0107] Step S13310: Organize and categorize the user demand resource association table, supplement the dynamic adjustment description of the association relationship, determine the update conditions, update cycle and update process of the association relationship, and form the final user demand resource association table.

[0108] In this embodiment, a user demand resource association table is compiled and categorized, and the data in the table is checked for completeness, accuracy, and uniformity of format. A supplementary explanation of the dynamic adjustment of association relationships is provided, specifying under what circumstances adjustments are needed, such as changes in user demand characteristics, updates to resource attributes, or the accumulation of historical borrowing data to a certain extent.

[0109] Define the update conditions for the association relationships, such as triggering an update when the historical borrowing frequency of a certain matching relationship remains below a threshold for a period of time; the update cycle, such as monthly or quarterly updates; and the update process, such as whether the system automatically adjusts based on data changes or requires manual review and confirmation. Add the above information to the categorized user demand resource association table to form the final user demand resource association table.

[0110] Step S134: Divide the resource adaptation levels according to the correlation strength between user demand characteristics and resource attributes. Each adaptation level corresponds to the number of associated features and matching logic, forming a standardized adaptation level division standard.

[0111] In the aforementioned university library scenario, resource adaptation levels are defined based on the strength of the correlation between user demand characteristics and resource attributes. Higher correlation strength corresponds to higher adaptation levels, and lower correlation strength corresponds to lower adaptation levels. Each adaptation level has a defined range for the number of associated features; for example, a first-level adaptation level requires at least five associated features, and a second-level adaptation level requires three to five associated features. Simultaneously, the matching logic for each adaptation level is clearly defined; for example, a first-level adaptation level requires complete matching of core demand features and most matching of extended demand features, while a second-level adaptation level requires basic matching of core demand features and partial matching of extended demand features. This standardized adaptation level division standard ensures consistency and operability in the resource adaptation level division.

[0112] Step S135: Based on the adaptation level classification standard, the library resources are labeled with adaptation levels. Each library resource corresponds to multiple adaptation levels in various scenarios. At the same time, the corresponding user borrowing intent tags are associated to generate a library resource adaptation level table. The library resource adaptation level table is stored according to resource category.

[0113] In this embodiment, library resources are labeled with adaptation levels based on the adaptation level classification standard. For each library resource, its adaptation level is determined by combining its resource attribute information and considering the strength and number of association features with user needs in different user borrowing scenarios. For example, a computer science textbook might be labeled as a level 1 adaptation level in a "course learning scenario" and as a level 2 adaptation level in a "scientific research scenario".

[0114] Simultaneously, the resource is associated with corresponding user borrowing intent tags, such as "course learning needs" or "research material needs." Library resources labeled with appropriate levels and associated tags are then categorized and stored according to resource type, generating a library resource appropriate level table to facilitate quick searching and filtering of resources in different scenarios.

[0115] Step S136: Construct a scenario-based resource adaptation module framework, dividing it into a requirement parsing layer, a resource matching layer, an adaptation sorting layer, and a dynamic adjustment layer. Set the functional positioning, data flow path, and inter-linkage relationship of each layer, and set the data transmission process and interaction method.

[0116] In the aforementioned university library scenario, a scenario-based resource adaptation module framework is constructed. The function of the demand parsing layer is to receive the user's borrowing intention tag set and parse out the user's real-time demand characteristics; the function of the resource matching layer is to match suitable resources from the library resource attribute classification dataset according to the user's demand characteristics; the function of the adaptation ranking layer is to rank the matched resources according to factors such as adaptation level; and the function of the dynamic adjustment layer is to dynamically adjust the resource matching and ranking results based on user interaction behavior and changes in resource status.

[0117] The data flow path is defined as follows: the requirement parsing layer passes the parsed real-time user requirement feature data to the resource matching layer; the resource matching layer passes the matched resource data to the adaptation and sorting layer; the adaptation and sorting layer passes the sorting results to the dynamic adjustment layer; and the dynamic adjustment layer outputs the final result. Each layer interacts with the other through interfaces, using specific data formats to ensure data accuracy and integrity.

[0118] Step S137: Embed a real-time intent recognition module in the demand parsing layer, connect to the dynamic user borrowing intent model, and parse the intent tags and demand features corresponding to the user's current borrowing behavior in real time to generate real-time user demand feature data.

[0119] In this embodiment, a real-time intent recognition module is embedded in the demand parsing layer. This module interfaces with a dynamic user borrowing intent model. When a user initiates a new borrowing activity, the real-time intent recognition module sends the user's current borrowing behavior data to the dynamic user borrowing intent model. The model analyzes the behavior data, identifying the corresponding intent tags and demand characteristics. The real-time intent recognition module then integrates this information to generate real-time user demand characteristic data.

[0120] Step S138: Integrate the user demand resource association table and the library resource adaptation hierarchy table in the resource matching layer, filter the corresponding library resources according to the user's real-time demand feature data, and at the same time associate the library resource inventory status information to generate a preliminary matching resource set.

[0121] In the aforementioned university library scenario, the resource matching layer integrates a user demand resource association table and a library resource adaptation hierarchy table. Upon receiving real-time user demand characteristic data, the system searches for corresponding resource attributes in the user demand resource association table based on these characteristics, and then combines this with the library resource adaptation hierarchy table to filter out library resources that meet the adaptation hierarchy requirements.

[0122] Simultaneously, the system correlates with the library's resource inventory status to ensure that the selected resources are currently available for borrowing. These resources are then integrated to generate a preliminary set of matched resources.

[0123] Step S139: Set up a sorting module in the adaptation sorting layer, and adjust the resource arrangement order by combining the adaptation level, resource update cycle, user historical preferences and resource borrowing popularity to generate adaptation sorting results.

[0124] In this embodiment, a sorting module is set up in the adaptation sorting layer. The sorting module first performs a preliminary sorting based on the resource's adaptation level, with resources of higher adaptation levels ranked first. Then, considering the resource update cycle, resources with more recent updates are appropriately moved forward. Next, user historical preferences are taken into account, prioritizing resources of types or themes that users have previously liked. Finally, resource borrowing popularity is considered, with resources with higher borrowing popularity also gaining an advantage in the sorting. By comprehensively considering the above factors, the order of resources is adjusted to generate the adaptation sorting result.

[0125] Step S1310: Construct adaptive adjustment rules in the dynamic adjustment layer, connect with user interaction behavior data and resource status change data, update the requirement resource association table and adaptation level division standard in real time, integrate functional modules and data association logic of each level, and form a scenario-based resource adaptation module.

[0126] In this embodiment, adaptive adjustment rules are constructed in the dynamic adjustment layer. These rules update the demand resource association table and the adaptation level classification criteria in real time based on user interaction data with recommended resources, such as clicks, borrowing, and reviews, as well as data on changes in library resource inventory status and the addition of new resources. For example, if a user increases the borrowing frequency of a certain type of resource, the adaptive adjustment rules will increase the association strength between that type of resource and the corresponding demand feature, adjusting its position in the demand resource association table; if the inventory of a resource becomes zero, it will be excluded during resource matching.

[0127] By integrating the functional modules and data association logic of the demand analysis layer, resource matching layer, adaptation sorting layer, and dynamic adjustment layer, a complete scenario-based resource adaptation module is formed, enabling dynamic adaptation and recommendation of library resources.

[0128] Step S140: The library's collection resources are filtered and sorted through the scenario-based resource adaptation module to generate a dynamic resource recommendation sequence. The dynamic resource recommendation sequence dynamically adjusts the order and content of the resources based on the user's real-time borrowing behavior, resource inventory status, and resource update status.

[0129] Step S141: Start the requirement parsing layer of the scenario-based resource adaptation module, obtain the user's current borrowing behavior data and historical borrowing behavior unit sequence in real time, call the dynamic user borrowing intent model to parse and generate the user's current borrowing intent tag and corresponding real-time requirement feature data, and identify the user's current key needs.

[0130] In this embodiment, the requirement parsing layer of the scenario-based resource adaptation module is activated. The requirement parsing layer obtains the user's current borrowing behavior data in real time through the interface with the library borrowing management system, such as the resource identifier, start time, borrowing channel, etc., as well as the user's historical borrowing behavior unit sequence.

[0131] The acquired data is sent to a dynamic user borrowing intent model. The model analyzes and processes this data, parsing and generating current borrowing intent tags for the user, such as "course learning needs" or "research material needs," and extracts corresponding real-time demand feature data, such as the resource topic, applicable scenario, and reading difficulty. By analyzing the real-time demand feature data, the model identifies the user's current priority needs, such as the user's most pressing need being the latest textbook for a specific course.

[0132] Step S142: Input the real-time user demand feature data into the resource matching layer of the scenario-based resource adaptation module, call the user demand resource association table and the library resource adaptation level table, filter out library resources whose adaptation level meets the set requirements, filter out library resources whose current inventory quantity is higher than the preset inventory threshold and whose borrowing time does not exceed the preset long-term borrowing threshold, and form a candidate resource pool.

[0133] In the aforementioned university library scenario, real-time user demand characteristics are input into the resource matching layer. The resource matching layer then calls the user demand resource association table to find the corresponding resource attributes based on the user's real-time demand characteristics. Next, it calls the library resource adaptation level table to filter out library resources whose adaptation level meets set requirements, such as filtering only resources at the first and second level adaptation levels.

[0134] Simultaneously, resources with current inventory levels exceeding a preset inventory threshold are selected to ensure a sufficient number of borrowable resources; resources with borrowing durations not exceeding a preset long-term borrowing threshold are also selected to avoid recommending resources that have been borrowed for extended periods and cannot be accessed promptly. Resources meeting these criteria are then integrated to form a candidate resource pool.

[0135] Step S143: Extract library resource attribute information from the candidate resource pool, supplement the library resources with recent borrowing records, resource update time, reading difficulty level and author background details, supplement the library resource description content, and generate a candidate resource details dataset, with each library resource corresponding to a complete attribute description.

[0136] In this embodiment, library resource attribute information from the candidate resource pool is extracted from the library resource attribute classification dataset, including resource theme, resource content dimension, and applicable scenarios. Then, the recent borrowing records of the resource are supplemented, such as the number of times borrowed and the borrowers in the past month; the resource update time, i.e., the publication time of the latest version; the reading difficulty level, such as beginner, intermediate, and advanced; and author background details, such as the author's academic achievements and research direction.

[0137] Simultaneously, supplementary descriptive content for library resources is provided, such as book summaries and table of contents. This information is then integrated to generate a complete attribute description for each library resource, forming a candidate resource details dataset to help users better understand the resource content.

[0138] Step S144: Input the candidate resource details dataset into the adaptation sorting layer of the scenario-based resource adaptation module, and perform preliminary sorting of the candidate resource details dataset according to the adaptation level to form a preliminary sorting result.

[0139] In this embodiment, the candidate resource details dataset is input into the adaptation sorting layer. The adaptation sorting layer sorts the resources in the candidate resource details dataset according to their adaptation level information. Resources with higher adaptation levels are ranked first, and resources with lower adaptation levels are ranked last. For resources with the same adaptation level, they are sorted according to their entry time, with those entered more recently ranked first. Through the above sorting method, a preliminary sorting result is formed.

[0140] Step S145: Adjust the preliminary sorting results based on the resource update cycle information, move the library resources whose update time is closer to the current time forward, and at the same time take into account the adaptation level. The adjustment weights are allocated according to the number of related features of the adaptation level and the length of time since the resource update time, and the time-adaptation sorting results are generated.

[0141] Step S1451: Extract the resource update cycle information of each library resource in the preliminary sorting results, record the first publication time, latest update time and update frequency of the resource, supplement the resource update content summary, and generate a library resource update information dataset.

[0142] In the aforementioned university library scenario, the resource update cycle information for each library resource in the preliminary ranking results is extracted from the library resource attribute information. This includes recording the resource's first publication date, latest update date, and update frequency (the number of times the resource has been updated within a given period).

[0143] Supplement the resource update summary, briefly describing the main content of each update, such as which chapters were added, which errors were corrected, and what new data was added. Compile this information to create a record for each library resource, generating a library resource update information dataset.

[0144] Step S1452: Sort the library resources in the library resource update information dataset independently according to the latest update time, distinguish the updated resources with different update time spans, record the timeliness sorting position of each library resource, and generate independent sorting results of resource timeliness.

[0145] In this embodiment, the library resources in the library resource update information dataset are independently sorted according to their latest update time. The resources are arranged from most recent to oldest in terms of their latest update time, with the most recently updated resource at the top.

[0146] Distinguish between updated resources with different update time spans. For example, classify resources updated within one month as "latest updated resources," those updated between one and three months as "relatively newest updated resources," and those updated more than three months as "early updated resources." Record the position of each library resource in the timeliness ranking, i.e., its ranking number. Combine the ranking results and position information to generate an independent timeliness ranking result for each resource.

[0147] Step S1453: Analyze the adaptation level information in the preliminary sorting results, count the number, distribution location and proportion of library resources corresponding to each adaptation level, determine the sorting distribution characteristics of resources at different adaptation levels, and generate an adaptation level distribution dataset.

[0148] In this embodiment, the adaptation level information in the preliminary sorting results is parsed to obtain the adaptation level of each library resource. The number of library resources corresponding to each adaptation level is counted, such as how many books are in the first-level adaptation level, how many are in the second-level adaptation level, etc.; the distribution location, i.e., the specific location index of these resources in the preliminary sorting results; and the percentage, i.e., the percentage of resources in each adaptation level out of the total number of resources.

[0149] Analyze the ranking and distribution characteristics of resources at different adaptation levels, such as whether resources at the first-level adaptation level are mainly concentrated at the beginning of the ranking results, and whether resources at the second-level adaptation level are relatively evenly distributed. Compile the above statistical and analytical results to generate an adaptation level distribution dataset.

[0150] Step S1454: Establish a balance rule between adaptation level and resource timeliness. Standardize the number of associated features corresponding to the adaptation level to generate standardized associated feature quantity values. Set the adaptation weight to be positively correlated with the standardized associated feature quantity values. At the same time, standardize the length of time between the resource update time and the current time to generate standardized time length values. Set the timeliness weight to be negatively correlated with the standardized time length values ​​and generate a weight allocation scheme.

[0151] In the aforementioned university library scenario, a balance rule is established between adaptation level and resource timeliness. The number of associated features corresponding to each adaptation level is standardized, mapping the number of associated features to a fixed numerical range, such as between zero and one, generating standardized associated feature quantity values. An adaptation weight is set to be positively correlated with the standardized associated feature quantity values; that is, the larger the standardized associated feature quantity values, the larger the adaptation weight.

[0152] Simultaneously, the time elapsed since the resource update is applied is standardized, mapping it to a value between zero and one to generate a standardized time length value. A timeliness weight is set that is negatively correlated with the standardized time length value; that is, the smaller the standardized time length value (indicating the update time is closer to the current time), the greater the timeliness weight. Based on the settings of the adaptation weight and timeliness weight, a weight allocation scheme is generated, clarifying how to calculate the adaptation weight and timeliness weight based on the standardized number of related features and the time length value.

[0153] Step S1455: Calculate the comprehensive score of each library resource according to the weight allocation scheme. The comprehensive score is obtained by summing the values ​​corresponding to the adaptation weight and the timeliness weight, and generating a comprehensive score dataset of library resources.

[0154] In this embodiment, a comprehensive score for each library resource is calculated according to a weighting scheme. For each library resource, an adaptation weight value is calculated based on the standardized correlation feature quantity value corresponding to its adaptation level, and a timeliness weight value is calculated based on the standardized time length value corresponding to the length of time between its resource update time and the current time.

[0155] The adaptation weight and timeliness weight are added together to obtain the overall score of the library resource. For example, if the adaptation weight of a resource is 0.6 and the timeliness weight is 0.3, the overall score is 0.9. The overall scores of each library resource are then compiled to generate a library resource overall score dataset.

[0156] Step S1456: Re-sort the library resources in the preliminary sorting results according to the comprehensive score value from high to low, and sort the resources with the same comprehensive score according to the relevance of the resource theme to generate the initial timeliness-adaptive sorting results.

[0157] In this embodiment, the library resources in the initial ranking are reordered according to their comprehensive scores, from highest to lowest. Resources with higher comprehensive scores are ranked first, and those with lower scores are ranked last. For library resources with the same comprehensive score, they are ranked according to the relevance of the resource's theme to the user's real-time needs, with resources of higher relevance ranked first. Through this ranking, an initial timeliness-adapted ranking result is generated.

[0158] Step S1457: Analyze the resource distribution of the initial timeliness adaptation sorting results, and count the number of resources appearing at a preset proportion position before the sorting sequence for different adaptation levels and timeliness intervals; set a maximum occurrence threshold for each level and interval, and for resource types that exceed the maximum occurrence threshold, adjust some of their resources to the end of the initial timeliness adaptation sorting results so that the distribution of each type of resource at the beginning of the initial timeliness adaptation sorting results meets the preset diversity ratio requirements, and generate distribution adjustment results.

[0159] In the aforementioned university library scenario, the resource distribution based on the initial timeliness adaptation sorting results is analyzed. The number of resources appearing at a predetermined percentage position (e.g., the first 30%) before the sorting sequence for different adaptation levels (e.g., Level 1, Level 2, Level 3) and different timeliness intervals (e.g., "latest updated resources," "relatively newest updated resources," "earlier updated resources") is statistically analyzed.

[0160] A maximum occurrence threshold is set for each adaptation level and time interval, representing the maximum number of resources of that type that can appear in the first preset proportion. If the occurrence number of a certain type of resource exceeds the maximum occurrence threshold, some of its resources are adjusted to the later part of the initial time-adaptation ranking result. During the adjustment, resources with higher overall scores are retained first, while resources with relatively lower scores are adjusted. Through the above adjustments, the distribution of each type of resource in the first part of the initial time-adaptation ranking result meets the preset diversity ratio requirements, avoiding excessive concentration of any one type of resource, and generating a distribution adjustment result.

[0161] Step S1458: Adjust the order of resources in the distribution adjustment results. For library resources of the same adaptation level and the same time interval, sort them according to the degree of fit between the reading difficulty of the resources and the user's historical reading difficulty habits, so that the reading difficulty matches the user's preferences and generate difficulty adaptation sorting results.

[0162] In this embodiment, the order of resources in the distribution adjustment results is adjusted. For library resources of the same adaptation level and the same time interval, their reading difficulty is analyzed, such as beginner, intermediate, and advanced. At the same time, the user's historical reading difficulty habits are analyzed, that is, the distribution of reading difficulty of resources borrowed by the user in the past, to summarize the user's preferred reading difficulty.

[0163] The algorithm calculates the degree to which the reading difficulty of a resource matches the user's historical reading difficulty habits. Resources with a high degree of match, such as those whose reading difficulty aligns with the user's preferred level, are ranked higher. Through these adjustments, the reading difficulty of recommended resources is made more closely aligned with user preferences, generating a difficulty-adapted ranking result.

[0164] Step S1459: Supplement the author background similarity information of the resources, make secondary fine adjustments to the library resources in the difficulty matching ranking results, and move the resources whose author background matches the user's historical preference authors forward to generate the final timeliness matching ranking results.

[0165] In the aforementioned university library scenario, supplement the difficulty-adaptive ranking results with author background similarity information for library resources. Analyze the background of resource authors, such as academic fields, research directions, published papers, and awards received; simultaneously analyze the background information of authors with historical user preferences.

[0166] The algorithm calculates the relevance between the author's background and the user's historical preferred author background. Resources with high relevance, such as those whose authors belong to the same academic field or have similar research directions, are ranked higher in the difficulty matching results. Through this secondary fine-tuning, the final timeliness matching ranking results are generated, further improving the personalization of recommendations.

[0167] Step S14510: Format the final timeliness adaptation sorting result, verify whether the sorting logic conforms to the preset priority rules, whether the weight allocation conforms to the weight allocation scheme, whether the resource distribution meets the preset diversity ratio requirements, and output the timeliness adaptation sorting result based on the verification results.

[0168] In this embodiment, the final timeliness adaptation sorting result is formatted to unify the display format of resource information, such as the order and font style of book title, author, publisher, update time, etc.

[0169] The validation process verifies whether the sorting logic conforms to the preset priority rules, i.e., whether it sorts first by overall score, then by topic relevance, reading difficulty fit, and author background fit; whether the weight allocation conforms to the weight allocation scheme, i.e., whether the calculation of adaptation weight and timeliness weight is correct; and whether the resource distribution meets the preset diversity ratio requirements, i.e., whether the distribution of resources at different adaptation levels and timeliness intervals meets the requirements in the first preset ratio positions. If any issues are found during validation, the process returns to the corresponding steps for adjustment; if the validation passes, the timeliness adaptation sorting results are output.

[0170] Step S146: Extract the preferred resource features from the user's historical borrowing behavior unit sequence, compare them with the library resource attributes in the timeliness adaptation ranking results, further adjust the resources that match the user's historical preference features, improve the display priority of the user's preferred resources, and generate the preference adaptation ranking results.

[0171] In this embodiment, preferred resource features are extracted from the user's historical borrowing behavior unit sequence. The attributes of resources borrowed in the past are analyzed, such as frequently borrowed topics, favorite authors, and preferred reading difficulty, to summarize the user's preferred resource features.

[0172] The library resource attributes in the timeliness-adapted ranking results are compared with the user's preferred resource characteristics to calculate the fit. Resources with high fit, such as those whose themes align with the user's preferences or whose authors are favorites, are further prioritized in the ranking results. Through these adjustments, preference-adapted ranking results are generated, making recommendations more aligned with the user's personal preferences.

[0173] Step S147: Collect data on changes in library resource inventory status and new resource entry data in real time, update the resource inventory information in the preference matching sorting results, remove newly added borrowed resources, add new resources that meet the requirements, and generate inventory update sorting results.

[0174] In this embodiment, by connecting with the library inventory management system in real time, data on changes in the status of library resource inventory, such as data on resources being borrowed, returned, and new resources being added to the inventory, are collected in real time.

[0175] Based on this data, update the resource inventory information in the preference-adaptation ranking results. For newly added resources with a borrowing status and zero inventory, remove them from the ranking results. For newly added resources that meet the user's real-time needs and adaptation level requirements, add them to the corresponding positions in the ranking results. Through these updates, generate an inventory-updated ranking result, ensuring that all recommended resources are currently available for borrowing.

[0176] Step S148: Select library resources from the candidate resource pool that meet the adaptation level requirements and match the user's needs. If the number of resources in the inventory update ranking result is lower than the preset recommended sequence length threshold, select library resources from the candidate resource pool that meet the preset adaptation level requirements and match the user's needs to supplement the resource pool until the number of resources reaches the recommended sequence length threshold. When supplementing resources, select resources whose difference in subject classification between the supplemented resources and existing resources is higher than the preset diversity threshold, and generate a supplementary ranking result.

[0177] In the aforementioned university library scenario, the system checks whether the number of resources in the inventory update ranking results reaches a preset recommended sequence length threshold. If it is lower than this threshold, library resources that meet the preset adaptation level (such as level two or above) and match the user's needs are selected from the candidate resource pool to supplement the resource pool.

[0178] When supplementing resources, the degree of difference between the supplemented resources and existing resources in terms of topic classification is calculated. Resources with a degree of difference higher than a preset diversity threshold are selected to ensure that the recommended sequence has a certain degree of topic diversity and avoid recommending overly singular resources. The supplemented resources are then sorted by comprehensive score and integrated into the inventory update ranking results to generate the supplementary ranking results.

[0179] Step S149: Set dynamic adjustment trigger conditions. When a user generates new borrowing or collection behavior, or when the resource inventory status changes or new resources are added to the inventory, the sorting result adjustment rules are triggered, and a dynamic adjustment signal is generated.

[0180] In this embodiment, dynamic adjustment trigger conditions are set. The dynamic adjustment trigger conditions are met when a user performs a new borrowing behavior, such as borrowing a new resource; or performs a collection behavior, such as collecting a resource from the recommended sequence; or the inventory status of library resources changes, such as a resource being checked out and causing a decrease in inventory; or when new resources are added to the library.

[0181] At this point, the system automatically triggers the sorting result adjustment rules, generates a dynamic adjustment signal, and notifies the dynamic adjustment layer to adjust the recommended sequence in real time.

[0182] Step S1410: Based on the dynamic adjustment signal, update and supplement the sorting results in real time, adjust the resource arrangement order and resource content, and form a dynamic resource recommendation sequence that is adjusted in real time according to changes in user behavior and resource status.

[0183] In this embodiment, after receiving a dynamic adjustment signal, the dynamic adjustment layer updates and supplements the ranking results in real time based on the user behavior data or resource status change data contained in the signal. For example, if a user borrows a new book about artificial intelligence, the dynamic adjustment layer analyzes this behavior, updates the user's real-time demand characteristics, and then adjusts the order of resources, increasing the recommendation weight of relevant artificial intelligence resources. If the inventory of a resource becomes zero, the dynamic adjustment layer removes it from the recommendation sequence and supplements it with other suitable resources. Through the above real-time updates, a dynamic resource recommendation sequence is formed.

[0184] Step S150: Collect user interaction behavior data on dynamic resource recommendation sequences, iteratively optimize the parameters of the dynamic user borrowing intent model and the mapping rules of the scenario-based resource adaptation module based on the interaction behavior data, and generate model and system optimization results.

[0185] Step S151: Construct a full-dimensional interactive behavior collection module to collect all interactive behavior data of users to the dynamic resource recommendation sequence, classify and store them according to user identifier and timestamp to form an original interactive dataset. The interactive behavior data includes detailed data corresponding to user click behavior, user borrowing behavior, user collection behavior, user skipping behavior, user dwell time and user evaluation feedback.

[0186] In this embodiment, a full-dimensional interactive behavior collection module is constructed. This module collects all interactive behavior data of users on dynamic resource recommendation sequences by embedding data collection code into user interaction interfaces such as the library's online recommendation platform and self-service borrowing equipment.

[0187] For user click behavior, record detailed data such as the clicked resource identifier, click time, and click location; for user borrowing behavior, record the borrowed resource identifier, borrowing time, and borrowing channel; for user collection behavior, record the collected resource identifier and collection time; for user skipping behavior, record the skipped resource sequence position and skip time; for user dwell time, record the dwell time on each resource details page; and for user evaluation feedback, record the rating, comment content, and evaluation time.

[0188] The collected interaction data is categorized and stored according to user identifier and timestamp. The data under each user identifier is arranged in chronological order according to the timestamp, forming the original interaction dataset.

[0189] Step S152: Clean the original interaction dataset to generate a valid interaction dataset. Parse the valid interaction dataset and classify and statistically analyze the frequency, duration, and type of interaction behavior corresponding to each library resource according to user identifier and borrowing intent tag. Generate a user interaction behavior statistical data set, which contains the association information between interaction behavior and intent tag.

[0190] In the aforementioned university library scenario, the original interactive dataset underwent data cleaning. This included removing duplicate data records, such as duplicate parts in records of multiple clicks by the same user on the same resource; handling missing values ​​by deleting data records lacking key information such as user or resource identifiers; and correcting outlier data, such as data with negative dwell times. After these processes, a valid interactive dataset was generated.

[0191] The effective interaction dataset is analyzed and categorized by user identifier and borrowing intent tag. For each user identifier and corresponding borrowing intent tag, the frequency of interaction behavior for each library resource is statistically analyzed, such as the number of clicks and borrowings; the interaction duration, such as total dwell time; and the interaction type, such as click, borrowing, and adding to favorites. This statistical information is then compiled to generate a user interaction behavior statistical dataset, which includes the association information between interaction behavior and intent tag, such as which borrowing intent tag corresponds to a particular interaction behavior.

[0192] Step S153: Perform correlation analysis between the user interaction behavior statistics set and the user borrowing intent tag set, calculate the interaction conversion rate of the resource corresponding to each intent tag, and adjust the weight parameters of the corresponding tags in the user borrowing intent tag set proportionally according to the interaction conversion rate value. The adjustment range of the tag weight parameters is positively correlated with the interaction conversion rate value, and generate the tag weight adjustment result.

[0193] In this embodiment, the user interaction behavior statistics set is correlated with the user borrowing intent tag set for analysis. For each intent tag, the number of times the corresponding resource is interacted with by the user (e.g., clicks, borrowing), and the total number of recommended resources under that tag are counted. The interaction conversion rate, i.e., the ratio of the number of interactions to the total number of recommendations, is then calculated.

[0194] Based on the interaction conversion rate, the weight parameters of the corresponding tags in the user borrowing intent tag set are adjusted proportionally. Tags with high interaction conversion rates indicate a strong user demand for resources under that tag, and their weight parameters are adjusted more significantly; tags with low interaction conversion rates have their weight parameters adjusted less significantly or even less significantly. The adjustment range of tag weight parameters is positively correlated with the interaction conversion rate. Through the above adjustments, the tag weight adjustment results are generated.

[0195] Step S154: Input the label weight adjustment result into the dynamic user borrowing intention model, optimize the feature extraction logic, intention mapping algorithm and parameter configuration of the dynamic user borrowing intention model, improve the sensitivity and accuracy of the dynamic user borrowing intention model in recognizing the user's potential borrowing intention, and generate the model parameter optimization result.

[0196] In this embodiment, the adjusted label weights are input into the dynamic user borrowing intent model. Based on the new label weight parameters, the model adjusts its feature extraction logic, such as assigning higher extraction priority to behavioral features associated with high-weight labels. The intent mapping algorithm is optimized so that the model considers high-weight labels more when mapping user behavior to borrowing intent labels. Simultaneously, other model parameter configurations are adjusted, such as the connection weights between neurons.

[0197] Through the above optimization measures, the sensitivity and accuracy of the dynamic user borrowing intent model in identifying users' potential borrowing intentions are improved, and the optimized model parameters are generated.

[0198] Step S155: Analyze the library resources that users have not interacted with in the effective interaction dataset, and combine the resource adaptation level, user demand characteristics and resource attribute information to investigate the causes of adaptation deviations and generate a deviation analysis report. The causes of adaptation deviations include demand feature mapping deviations, adaptation level labeling deviations and sorting position deviations.

[0199] Step S1551: From the effective interaction dataset, filter out library resources with zero user clicks, borrowing status of "not borrowed", collection status of "not collected", and page dwell time of less than a preset time threshold to form a list of resources that users have not interacted with. Record the adaptation level, complete attribute information, sorting position and corresponding user borrowing intent tag of each library resource.

[0200] In the aforementioned university library scenario, library resources that users have not interacted with are filtered from the effective interaction dataset. Filtering criteria are set as follows: zero clicks (meaning the user has never clicked on the resource); not borrowed (meaning the user has not borrowed the resource); not saved (meaning the user has not saved the resource); and page dwell time is less than a preset threshold, such as less than 30 seconds (meaning the user spends very little time on the resource's details page).

[0201] Library resources that simultaneously meet these conditions are filtered out to form a list of resources that the user has not interacted with. Each library resource in the list is recorded with its appropriate level (e.g., Level 1, Level 2); complete attribute information such as resource subject, author, update time, and reading difficulty; its ranking position in the recommendation sequence; and the corresponding user borrowing intent tag, such as "course learning needs" or "research material needs".

[0202] Step S1552: Statistically analyze the distribution of library resources in the list of resources not interacted by users across different adaptation levels, calculate the proportion of resources not interacted at each adaptation level, compare the differences in the non-interaction rate of different adaptation levels, and generate a statistical data set of non-interaction data for each adaptation level.

[0203] In this embodiment, the distribution of library resources across adaptation levels in the resource list that the user has not interacted with is statistically analyzed. The number of resources at each adaptation level (Level 1, Level 2, Level 3, etc.) is counted separately.

[0204] Calculate the non-interaction ratio for resources at each adaptation level, which is the ratio of the number of non-interactive resources at that level to the total number of resources at that level in the recommendation sequence. Compare the differences in non-interaction rates across different adaptation levels to analyze whether higher adaptation levels generally correspond to lower non-interaction rates, or whether there are any abnormally high non-interaction rates for certain adaptation levels. Organize the above statistical data and comparison results to generate a statistical dataset of non-interaction data for each adaptation level.

[0205] Step S1553: Compare the attribute information of the uninterrupted resources with the real-time user demand feature data one by one, calculate the matching degree between the attribute values ​​of resource theme, content dimension, reading difficulty and applicable scenario and the corresponding demand features in the real-time user demand feature data, mark the dimensions with matching degree lower than the preset matching threshold as mismatched dimensions, record the specific deviation points, and generate an attribute demand deviation dataset.

[0206] In this embodiment, the attribute information of each uninterrupted resource in the list of uninterrupted resources is compared with the user's real-time demand feature data. For the resource topic, the matching degree with the user's real-time demand topic is calculated; for the content dimension, the matching degree with the user's demand content dimension is calculated; for the reading difficulty, the matching degree with the user's demand reading difficulty is calculated; and for the applicable scenario, the matching degree with the user's current borrowing scenario is calculated.

[0207] The matching degree of each dimension is compared with a preset matching threshold. If the matching degree is lower than the threshold, the dimension is marked as a mismatched dimension. Specific deviations of the mismatched dimensions are recorded, such as the resource topic being irrelevant to the requirement topic, or the reading difficulty exceeding the user's requirement difficulty. This information is then compiled to generate an attribute requirement deviation dataset.

[0208] Step S1554: Analyze the arrangement position of non-interactive resources in the recommendation sequence, count the number and proportion of non-interactive resources in different sorting positions, analyze the impact of sorting position on user interaction behavior, calculate the ratio of the sorting position of non-interactive resources to the total length of the recommendation sequence, and analyze the correlation between this ratio and the non-interaction rate, and generate a sorting position impact dataset.

[0209] In this embodiment, the arrangement of uninterrupted resources in the user's uninterrupted resource list within the recommendation sequence is analyzed. The recommendation sequence is divided into different position intervals, such as the first 20%, 20% to 40%, 40% to 60%, 60% to 80%, and the last 20%.

[0210] The analysis involves calculating the number and percentage of uninterrupted resources within each position interval, i.e., the ratio of uninterrupted resources in that interval to the total number of uninterrupted resources. It then analyzes the impact of ranking position on user interaction behavior, such as whether resources ranked higher have a lower uninterrupted rate, or vice versa. The analysis also calculates the ratio of the ranking position of an uninterrupted resource to the total length of the recommendation sequence, i.e., the ranking position divided by the total length, yielding a relative position value. Finally, it analyzes the correlation between this relative position value and the uninterrupted rate, such as whether a larger relative position value corresponds to a higher uninterrupted rate. These results are then compiled to generate a dataset on the impact of ranking position.

[0211] Step S1555: Combine the user's historical borrowing behavior unit sequence, compare the attribute differences, adaptation level differences, and intent tag fit differences between the uninterrupted resources and the user's historical preference characteristics, analyze the main behavioral motivations for the user's non-interaction, and generate a behavioral motivation analysis dataset.

[0212] In this embodiment, the user's historical borrowing behavior sequence is combined to extract the user's historical preference features, such as preferred resource themes, authors, reading difficulty, and adaptation levels. The differences in attributes between uninterrupted resources and the user's historical preference features are compared, such as whether the theme of the uninterrupted resource differs from the user's preferred theme; differences in adaptation levels, such as whether the adaptation level of the uninterrupted resource is lower than the adaptation level of resources the user usually borrows; and differences in intent tag fit, such as whether the fit between the intent tags corresponding to the uninterrupted resource and the user's main intent tags is low.

[0213] Through the above comparisons, we analyzed the main behavioral motivations for user non-interaction, such as resources not matching user's historical preferences, insufficient adaptation level, and mismatched intent tags. The results of this motivation analysis were then compiled to generate a behavioral motivation analysis dataset.

[0214] Step S1556: Compare the mapping relationship, adaptation level division criteria and sorting algorithm output in the user demand resource association table with the actual user interaction records in the effective interaction dataset, calculate the degree of consistency between prediction and reality, mark the part with a consistency lower than a preset threshold as a system-level matching deviation, and generate a system matching deviation dataset.

[0215] In the aforementioned university library scenario, the mapping relationship in the user demand resource association table (i.e., the association relationship between demand features and resource attributes), the adaptation level division standard (i.e., the basis for dividing resource adaptation levels), and the result of the ranking algorithm (i.e., the recommendation sequence) are compared with the actual user interaction records in the effective interaction dataset.

[0216] The calculation determines the degree of agreement between predictions and reality, such as how many recommended resources were actually interacted with by users, and how many of the interacted resources were from the recommended sequence. Parts with an agreement degree below a preset threshold, such as low recommendation accuracy for resources corresponding to a specific need feature, actual interaction rates for resources at a certain adaptation level being far lower than expected, or ranking algorithms placing resources of user interest at the bottom, are identified as system-level matching deviations. These deviation information are then compiled to generate a system matching deviation dataset.

[0217] Step S1557: Integrate the matching hierarchy non-interactive statistical data set, attribute requirement deviation dataset, sorting position influence dataset, behavioral motivation analysis dataset, and system matching deviation dataset, and sort out the specific manifestations, scope of influence, and core reasons of various deviations.

[0218] In this embodiment, various deviation datasets are integrated. For the non-interaction statistics dataset at the adaptation level, the specific manifestations of non-interaction rates at different adaptation levels, the number of resources involved (scope of impact), and the core reasons are analyzed, such as inaccurate labeling of adaptation levels. For the attribute requirement deviation dataset, the specific manifestations of mismatched dimensions, the resources and requirement features involved (scope of impact), and the core reasons are analyzed, such as deviation in requirement feature mapping. For the ranking position impact dataset, the specific manifestations of non-interaction rates at different position intervals, the impact on the overall interaction rate (scope of impact), and the core reasons are analyzed, such as ranking position deviation. For the behavioral motivation analysis dataset, the specific manifestations of the main behavioral motivations for user non-interaction, the user groups and resource types involved (scope of impact), and the core reasons are analyzed, such as not conforming to user historical preferences. For the system matching deviation dataset, the specific manifestations of system-level matching deviations, the impact on the overall performance of the recommendation system (scope of impact), and the core reasons are analyzed, such as unreasonable mapping relationships and improper setting of ranking algorithm parameters.

[0219] Step S1558: Classify and statistically analyze the various deviation causes, calculate the number and proportion of non-interactive resources corresponding to each deviation cause, distinguish between primary and secondary deviation causes, and generate statistical analysis results of deviation causes.

[0220] In this embodiment, the various deviation causes identified after integration are categorized and statistically analyzed. The deviation causes are classified into categories such as requirement feature mapping deviation, adaptation level annotation deviation, sorting position deviation, non-compliance with user historical preferences, unreasonable system mapping relationships, and improper sorting algorithm parameter settings.

[0221] Calculate the number of non-interactive resources corresponding to each cause of deviation, i.e., how many non-interactive resources were caused by that cause; and the percentage, i.e., the ratio of the number of non-interactive resources caused by that cause to the total number of non-interactive resources. Based on the quantity and percentage, distinguish between primary and secondary causes of deviation; those with a higher percentage are primary causes of deviation, and those with a lower percentage are secondary causes of deviation. Compile the above statistical results to generate the statistical analysis results of deviation causes.

[0222] Step S1559: For different causes of deviation, analyze the degree of impact, scope of impact and subsequent improvement direction of the deviation on the resource recommendation effect, and generate a deviation impact assessment report.

[0223] In this embodiment, the impact of each cause of deviation on the resource recommendation effect is analyzed. The degree of impact is measured by indicators such as the number of uninteracted resources, the non-interaction rate, and the impact on user satisfaction. The scope of impact is analyzed, including the range of affected user groups, resource types, and recommendation scenarios.

[0224] Based on the analysis of the causes and impacts of the deviations, directions for future improvements are proposed. For example, for deviations in demand feature mapping, the improvement direction is to optimize the mapping relationship between demand features and resource attributes; for deviations in adaptation level labeling, the improvement direction is to improve the adaptation level division criteria; for deviations in sorting position, the improvement direction is to adjust the weights of each factor in the sorting algorithm. The above analysis is then compiled to generate a deviation impact assessment report.

[0225] Step S15510: Compile the statistical analysis results of the causes of deviations and the deviation impact assessment report, and write them according to the type of deviation, degree of impact and correction ideas to form a deviation analysis report.

[0226] In this embodiment, the statistical analysis results of the causes of deviations and the deviation impact assessment report are compiled. Deviations are categorized by type, such as requirement feature mapping deviation, adaptation level labeling deviation, and sorting position deviation. Under each deviation type, they are arranged from highest to lowest impact, with detailed descriptions of the specific manifestations, statistical data, degree of impact, scope of impact, and correction strategies.

[0227] The corrective approach is based on the improvement directions outlined in the deviation impact assessment report, proposing specific corrective measures and steps. The above content is then organized systematically to form a deviation analysis report.

[0228] Step S156: Based on the deviation analysis report, adjust the user demand resource association table and adaptation level division standard in the scenario-based resource adaptation module, correct the mapping relationship between demand characteristics and resource attributes, optimize the adaptation level annotation logic, and generate a system rule adjustment plan.

[0229] In this embodiment, the user demand resource association table is adjusted based on the adaptation deviation reasons in the deviation analysis report. For deviations in demand feature mapping, the mapping relationship between demand features and resource attributes is corrected to ensure that resource attributes accurately reflect user demand features. For deviations in adaptation level labeling, the adaptation level division criteria are optimized to make the labeling of adaptation levels more accurate. For deviations in sorting position, the weights of each factor in the sorting algorithm are adjusted so that important resources can be ranked higher. Through these adjustments, a system rule adjustment scheme is generated.

[0230] Step S157: Apply the system rule adjustment scheme to the scenario-based resource adaptation module, update the running parameters and mapping rules of the requirement parsing layer, resource matching layer, adaptation sorting layer and dynamic adjustment layer, and generate system optimization results.

[0231] In this embodiment, the system rule adjustment scheme is applied to each level of the scenario-based resource adaptation module. The requirement parsing layer updates the operating parameters related to user requirement feature parsing; the resource matching layer updates the user requirement resource association table and the adaptation level division criteria; the adaptation sorting layer updates the sorting algorithm parameters and mapping rules; and the dynamic adjustment layer updates the adaptive adjustment rules. Through these updates, system optimization results are generated, improving the performance of the scenario-based resource adaptation module.

[0232] Step S158: Integrate the model parameter optimization results and system optimization results to construct a verification dataset. Using the verification dataset, calculate the intent recognition accuracy and recall rate of the optimized dynamic user borrowing intent model, as well as the recommendation click-through rate and borrowing conversion rate of the scenario-based resource adaptation module, and generate an effect verification report containing the above quantitative indicators.

[0233] In the aforementioned university library scenario, the model parameter optimization results and system optimization results were integrated to construct a validation dataset. The validation dataset includes a portion of user borrowing behavior data and interaction behavior data that were not involved in model training and optimization.

[0234] Using a validation dataset, calculate the intent recognition accuracy (the proportion of times the model correctly identifies a user's borrowing intent) and recall (the proportion of times the model identifies a user's genuine borrowing intent) of the optimized dynamic user borrowing intent model. Simultaneously, calculate the recommendation click-through rate (CTR) of the contextual resource adaptation module (the ratio of user clicks on recommended resources to the total number of recommendations) and the borrowing conversion rate (the ratio of user borrowing recommended resources to clicks). Compile these quantitative metrics to generate a performance validation report and evaluate the performance of the optimized model and module.

[0235] Step S159: Based on the effect verification report, fine-tune the model parameters and system rules, correct the deviations that occurred during the optimization process, determine the final optimization scheme, and form the final model and system optimization results.

[0236] In this embodiment, the quantitative indicators in the analysis effect verification report are analyzed. If the intent recognition accuracy, recall rate, recommendation click-through rate, or borrowing conversion rate do not meet the expected targets, the parameters of the dynamic user borrowing intent model are fine-tuned based on the specific data in the report, such as further adjusting feature weights or algorithm parameters; at the same time, the system rules of the scenario-based resource adaptation module are fine-tuned, such as readjusting the mapping relationship or ranking weights.

[0237] Correcting deviations that occurred during the optimization process, such as inaccurate identification of certain types of user intent and poor matching of resources for specific topics. After multiple fine-tunings, the final optimization scheme was determined, resulting in the final model and system optimization results, ensuring that the intelligent recommendation system achieves optimal performance.

[0238] Figure 2 The illustration shows a schematic diagram of the hardware structure of a big data analysis-based intelligent library resource recommendation system 100, provided in an embodiment of this application, for implementing the above-described big data analysis-based intelligent library resource recommendation method. Figure 2 As shown, the intelligent recommendation system for library resources based on big data analysis 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0239] Machine-readable storage medium 120 may store data and / or instructions. In some embodiments, machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, machine-readable storage medium 120 may store data and / or instructions used by the big data analytics-based intelligent library resource recommendation system 100 to perform or use in order to complete the exemplary methods described in this application. One or more processors 110 execute the computer-executable instructions stored in machine-readable storage medium 120, causing processor 110 to perform the big data analytics-based intelligent library resource recommendation method as described in the above method embodiments. Processor 110, machine-readable storage medium 120, and communication unit 140 are connected via bus 130, and processor 110 may be used to control the transmission and reception operations of communication unit 140.

[0240] The specific implementation process of processor 110 can be found in the various method embodiments executed by the above-mentioned intelligent recommendation system for library resources based on big data analysis. The implementation principle and technical effect are similar, and will not be repeated here.

[0241] Furthermore, this application embodiment also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned intelligent recommendation method for library resources based on big data analysis is implemented.

[0242] It should be noted that, in order to simplify the description disclosed in this application and thus help to understand one or more embodiments of the invention, the foregoing description of the embodiments of this application may sometimes combine multiple features into one embodiment, drawing or description thereof.

Claims

1. A method for intelligent recommendation of library resources based on big data analysis, characterized in that, The method includes: The user borrowing behavior trajectory data is deconstructed and divided into user borrowing behavior unit sequences by time slice. The user borrowing behavior trajectory data includes the start time, return time, resource borrowing channel, resource reading progress and resource evaluation feedback information of the user borrowing library resources. The user borrowing behavior unit sequence is composed of the behavior units corresponding to a single complete borrowing behavior arranged in chronological order. Each behavior unit is associated with a unique user identifier and resource identifier. Based on the user borrowing behavior unit sequence mining behavior association logic and scenario mapping relationship, a dynamic user borrowing intention model is constructed. A user borrowing intention tag set is generated by associating behavior unit features. The user borrowing intention tag set corresponds to the user resource demand tendency under different borrowing scenarios and different reading purposes, and is supplemented and updated in real time with the user's new borrowing behavior. By linking user borrowing intent tags with library resource attribute information, a scenario-based resource adaptation module is constructed. The library resource attribute information includes resource theme, resource content dimension, resource applicable scenario, resource update cycle, resource reading difficulty, and resource author background information. A dynamic mapping rule library between needs and resources is established within the scenario-based resource adaptation module. The library’s collection resources are filtered and sorted through a scenario-based resource adaptation module to generate a dynamic resource recommendation sequence. The dynamic resource recommendation sequence is dynamically adjusted in order and content based on users’ real-time borrowing behavior, resource inventory status and resource update status. Collect user interaction data on dynamic resource recommendation sequences, iteratively optimize the parameters of the dynamic user borrowing intent model and the mapping rules of the scenario-based resource adaptation module based on the interaction data, and generate model and system optimization results.

2. The intelligent library resource recommendation method based on big data analysis according to claim 1, characterized in that, The process involves mining the behavioral association logic and scenario mapping relationship based on user borrowing behavior unit sequences to construct a dynamic user borrowing intent model. This model generates a set of user borrowing intent tags through behavioral unit feature association, including: Extract the core behavioral features of each behavioral unit in the user borrowing behavior unit sequence. The core behavioral features include the type of borrowed resource, borrowing duration, reading progress completion rate, frequency of re-borrowing, resource evaluation level and borrowing time period distribution. Classify and integrate them according to user identifier to form a user behavior unit feature set, with each user corresponding to an independent feature subset. The behavioral unit features in the user behavior unit feature set are analyzed in chronological order to explore the feature inheritance logic and mutation rules between adjacent behavioral units, extract the continuous trend of user borrowing behavior over time, and generate user behavior correlation trend data, which includes feature correlation strength and change node information. Based on the time period distribution of user borrowing behavior, the reading progress completion rate and resource category, user borrowing scenario types are divided, and each user borrowing scenario type is associated with a corresponding behavioral feature identifier; The user behavior trend data is bidirectionally matched with the scene category classification results to establish a unique correspondence between the combination of behavioral features and the scene category, and a user behavior scene correspondence table is generated. The scene category to which each combination of behavioral features belongs and the basis for association are determined in the user behavior scene correspondence table. Based on the user behavior scenario mapping table, initial user borrowing intent tags are generated and a set of initial user borrowing intent tags is formed. Each initial user borrowing intent tag corresponds to a core resource demand tendency under a user borrowing scenario type. The tag content of the initial user borrowing intent tag is associated with specific resource attribute features. Integrate the user's initial borrowing intent tag set with the user's historical borrowing behavior data, supplement the behavioral support cases corresponding to each initial user borrowing intent tag, increase the descriptive dimensions and scope of application of the initial user borrowing intent tags, enhance the correlation between the initial user borrowing intent tags and user behavior, and generate a user-enhanced borrowing intent tag set. Remove duplicate tags and semantically conflicting tags from the user's enhanced borrowing intent tag set, and merge semantically similar borrowing intent tags to form a user-standardized borrowing intent tag set; Using the set of user-defined borrowing intent tags as the training target, the set of user behavior unit features is input into the dynamic user borrowing intent model for training. The feature extraction logic and intent mapping algorithm of the dynamic user borrowing intent model are optimized, and the ability of the dynamic user borrowing intent model to identify potential user borrowing intent is optimized, thus generating the trained dynamic user borrowing intent model. By training a dynamic user borrowing intent model, the borrowing intent hidden in the user behavior unit sequence is mined, and potential demand tags not covered by the initial user borrowing intent tags are supplemented to form a complete set of user borrowing intent tags. The user's complete borrowing intent tag set is sorted according to the scenario category and the strength of the association between the tag and user behavior to form the final user borrowing intent tag set.

3. The intelligent library resource recommendation method based on big data analysis according to claim 1, characterized in that, The linked user borrowing intent tag set and library resource attribute information are used to construct a scenario-based resource adaptation module, including: The user borrowing intent tag set is analyzed, the core and extended demand features corresponding to each borrowing intent tag are extracted, the user's resource demand focus and potential demand direction in different scenarios are analyzed, and a user demand feature list is generated. The user demand feature list is organized by scenario category. The library resource attribute information is classified and processed according to resource theme, resource content dimension, resource applicable scenario, resource update cycle, resource reading difficulty and resource author background information. The library collection location, inventory quantity and borrowing record information are supplemented to form a library resource attribute classification dataset. A bidirectional mapping relationship is established between the user demand feature list and the library resource attribute classification dataset. Each demand feature is associated and matched with the corresponding resource attribute. The association logic between the demand feature and the resource attribute is determined, and a user demand resource association table is generated. The user demand resource association table contains association strength information. Resource adaptation levels are divided according to the strength of the correlation between user demand characteristics and resource attributes. Each adaptation level corresponds to a number of related features and matching logic, forming a standardized adaptation level division standard. Based on the adaptation level classification standard, library resources are labeled with adaptation levels. Each library resource corresponds to multiple adaptation levels in various scenarios and is associated with corresponding user borrowing intent tags to generate a library resource adaptation level table. The library resource adaptation level table is stored according to resource category. Construct a scenario-based resource adaptation module framework, dividing it into a requirement parsing layer, a resource matching layer, an adaptation sorting layer, and a dynamic adjustment layer. Define the functional positioning, data flow path, and inter-linkage relationship of each layer, and define the data transmission process and interaction method. A real-time intent recognition module is embedded in the demand parsing layer, which connects to the dynamic user borrowing intent model to parse the intent tags and demand features corresponding to the user's current borrowing behavior in real time and generate real-time user demand feature data. The resource matching layer integrates the user demand resource association table and the library resource adaptation hierarchy table, filters the corresponding library resources according to the user's real-time demand characteristic data, and associates the library resource inventory status information to generate a preliminary matching resource set. A sorting module is set up in the adaptation sorting layer. The resource arrangement order is adjusted based on the adaptation level, resource update cycle, user historical preferences and resource borrowing popularity to generate adaptation sorting results. In the dynamic adjustment layer, adaptive adjustment rules are built, which connect to user interaction behavior data and resource status change data, update the requirement resource association table and adaptation level division standard in real time, and integrate the functional modules and data association logic of each layer to form a scenario-based resource adaptation module.

4. The intelligent library resource recommendation method based on big data analysis according to claim 1, characterized in that, The process of filtering and sorting library collection resources through a scenario-based resource adaptation module to generate a dynamic resource recommendation sequence includes: The requirement parsing layer of the scenario-based resource adaptation module is activated to obtain the user's current borrowing behavior data and historical borrowing behavior unit sequence in real time. The dynamic user borrowing intent model is called to parse and generate the user's current borrowing intent tag and corresponding real-time requirement feature data to identify the user's current key needs. Input real-time user demand feature data into the resource matching layer of the scenario-based resource adaptation module, call the user demand resource association table and the library resource adaptation level table, filter out library resources whose adaptation level meets the set requirements, filter out library resources whose current inventory quantity is higher than the preset inventory threshold and whose borrowing time does not exceed the preset long-term borrowing threshold, and form a candidate resource pool. Extract library resource attribute information from the candidate resource pool, supplement the library resources with recent borrowing records, resource update time, reading difficulty level and author background details, supplement the library resource description content, and generate a candidate resource detail dataset, with each library resource corresponding to a complete attribute description; The candidate resource details dataset is input into the adaptation sorting layer of the scenario-based resource adaptation module, and the candidate resource details dataset is initially sorted according to the adaptation level to form a preliminary sorting result; The preliminary sorting results are adjusted by combining the resource update cycle information. Library resources with update times closer to the current time are moved forward in position. At the same time, the adaptation level is taken into account. The adjustment weights are allocated according to the number of related features of the adaptation level and the length of time from the current update time to generate the time-adaptation sorting results. Extract the preferred resource features from the user's historical borrowing behavior unit sequence, compare them with the library resource attributes in the timeliness adaptation ranking results, further adjust the resources that match the user's historical preference features to improve the display priority of the user's preferred resources, and generate preference adaptation ranking results. The system collects real-time data on changes in library resource inventory status and new resource entry data, updates resource inventory information in the preference-adapted sorting results, removes newly added borrowed resources, adds new resources that meet the requirements, and generates updated inventory sorting results. Library resources that meet the adaptation level requirements and match the user's needs are selected from the candidate resource pool. If the number of resources in the inventory update ranking result is lower than the preset recommended sequence length threshold, library resources that meet the adaptation level requirements and match the user's needs are selected from the candidate resource pool to supplement the resource pool until the number of resources reaches the recommended sequence length threshold. When supplementing resources, resources whose difference in subject classification between the supplemented resources and existing resources is higher than the preset diversity threshold are selected to generate supplementary ranking results. Set dynamic adjustment trigger conditions. When a user generates new borrowing or collection behavior, or when the resource inventory status changes or new resources are added to the inventory, the sorting result adjustment rules are triggered, and a dynamic adjustment signal is generated. Based on the real-time updates and supplementation of sorting results using dynamic adjustment signals, the resource arrangement order and resource content are adjusted to form a dynamic resource recommendation sequence that is adjusted in real time according to changes in user behavior and resource status.

5. The intelligent library resource recommendation method based on big data analysis according to claim 1, characterized in that, The process of collecting user interaction data on dynamic resource recommendation sequences, and iteratively optimizing the parameters of the dynamic user borrowing intent model and the mapping rules of the contextualized resource adaptation module based on the interaction data, includes: A full-dimensional interactive behavior collection module is constructed to collect all interactive behavior data of users to the dynamic resource recommendation sequence. The data is classified and stored according to user identifier and timestamp to form the original interactive dataset. The interactive behavior data includes detailed data corresponding to user click behavior, user borrowing behavior, user collection behavior, user skipping behavior, user dwell time and user evaluation feedback. The original interaction dataset is cleaned to generate a valid interaction dataset. The valid interaction dataset is parsed, and the frequency, duration, and type of interaction behavior corresponding to each library resource are statistically analyzed according to user identifier and borrowing intent tag. A user interaction behavior statistical data set is generated, which contains the association information between interaction behavior and intent tag. The system performs correlation analysis between the user interaction behavior statistics set and the user borrowing intent tag set, calculates the interaction conversion rate of the resource corresponding to each intent tag, and adjusts the weight parameters of the corresponding tags in the user borrowing intent tag set proportionally based on the interaction conversion rate value. The adjustment range of the tag weight parameters is positively correlated with the interaction conversion rate value, and generates the tag weight adjustment result. The label weight adjustment results are input into the dynamic user borrowing intention model to optimize the feature extraction logic, intention mapping algorithm and parameter configuration of the dynamic user borrowing intention model, thereby improving the sensitivity and accuracy of the dynamic user borrowing intention model in recognizing users' potential borrowing intentions and generating model parameter optimization results. Analyze library resources that users have not interacted with in the effective interaction dataset, and combine resource adaptation level, user demand characteristics and resource attribute information to investigate the causes of adaptation deviation and generate a deviation analysis report. The causes of adaptation deviation include demand feature mapping deviation, adaptation level labeling deviation and sorting position deviation. Based on the deviation analysis report, the user demand resource association table and adaptation level division standard in the scenario-based resource adaptation module were adjusted, the mapping relationship between demand characteristics and resource attributes was corrected, the adaptation level annotation logic was optimized, and a system rule adjustment plan was generated. The system rule adjustment scheme is applied to the scenario-based resource adaptation module to update the running parameters and mapping rules of the requirement parsing layer, resource matching layer, adaptation sorting layer and dynamic adjustment layer, and generate system optimization results. Integrate the model parameter optimization results and system optimization results to construct a validation dataset. Using the validation dataset, calculate the intent recognition accuracy and recall rate of the optimized dynamic user borrowing intent model, as well as the recommendation click-through rate and borrowing conversion rate of the scenario-based resource adaptation module, and generate an effect validation report containing the above quantitative indicators. Based on the effect verification report, the model parameters and system rules are fine-tuned, deviations that occur during the optimization process are corrected, the final optimization scheme is determined, and the final model and system optimization results are formed.

6. The intelligent library resource recommendation method based on big data analysis according to claim 2, characterized in that, The process involves performing correlation analysis on the behavioral unit features in the user behavior unit feature set in chronological order, mining the feature inheritance logic and mutation patterns between adjacent behavioral units, extracting the continuous trend of user borrowing behavior over time, and generating user behavior correlation trend data, including: The behavioral units in the user behavior unit feature set are reordered in chronological order to form a temporal user behavior feature sequence, which includes the timestamp and complete feature information of each behavioral unit. Extract the core features of two adjacent behavioral units in the time-series user behavior feature sequence, compare the differences and commonalities of features one by one, calculate the differences in borrowing resource category, borrowing duration, reading progress completion ratio and borrowing time period, and compare them with the preset feature change threshold to generate a dataset of adjacent behavioral features comparison. The frequency of occurrence of the same resource category in adjacent behavioral units in the time-series user behavior feature sequence is statistically analyzed. The continuous cycle, interval time and number of repeated borrowing of the same type of resource are analyzed to explore the continuous pattern of borrowing of the same type of resource and generate a continuous borrowing dataset of the same type of resource. Track the conversion of different resource categories in adjacent behavioral units in a time-series user behavior feature sequence, analyze the triggering factors of category conversion, capture the borrowing trend of users switching from one type of resource to another, and generate a resource category conversion dataset, which includes the conversion direction and conversion frequency. By integrating adjacent behavioral feature comparison datasets, continuous borrowing datasets of similar resources, and resource category conversion datasets, a user behavior correlation analysis model is constructed. The user behavior correlation analysis model embeds behavioral feature correlation algorithms and trend recognition logic. By extracting the continuous change characteristics of user borrowing behavior through a user behavior correlation analysis model, identifying key time nodes where the behavior characteristics change significantly, recording the differences in behavior characteristics before and after each node, and generating a dataset of behavior change nodes; Analyze the differences in user behavior characteristics before and after the behavior change nodes, and combine the user borrowing scenarios, resource attributes and reading purposes in the corresponding time period to analyze the core factors that lead to the change in behavior characteristics and generate a dataset of behavioral change drivers. Based on the dataset of behavioral change nodes and the dataset of behavioral change drivers, a user borrowing behavior trend curve is plotted. The user borrowing behavior trend curve shows the change trajectory and key node positions of different feature dimensions, and generates behavioral trend curve data. Integrate behavioral trend curve data with various correlation analysis datasets, supplement the logical basis and motivation of behavioral correlation, add correlation type identifiers and confidence parameters to behavioral correlation data, and generate preliminary user behavior correlation trend data; Preliminary user behavior correlation trend data is categorized and organized according to scenario type and behavioral change patterns to form final user behavior correlation trend data.

7. The intelligent library resource recommendation method based on big data analysis according to claim 3, characterized in that, The process of establishing a bidirectional mapping relationship between the user demand feature list and the library resource attribute classification dataset, associating and matching each demand feature with its corresponding resource attribute, determining the association logic between demand features and resource attributes, and generating a user demand resource association table includes: Extract each requirement feature from the user requirement feature list, break it down into multiple specific requirement elements, determine the core and extended requirements of each requirement feature, supplement the behavioral support basis corresponding to the requirement elements, and generate the user requirement element decomposition results. Each library resource attribute in the library resource attribute classification dataset is further broken down into multiple detailed resource attribute elements. Each attribute element is supplemented with a specific description and feature identifier to generate a list of library resource attribute elements, which is then categorized by resource type. Compare the user demand element breakdown results with the library resource attribute element list one by one, match elements with similar semantics, logical connections and feature fit, establish a two-way correspondence relationship at the element level, record the association logic of each correspondence relationship, and generate an element-level association dataset. Based on the feature-level association dataset, the association strength between each demand feature and its corresponding resource attribute is calculated. The association strength is determined by the number of matching features, semantic similarity, and logical correlation, thus generating a demand-resource association strength dataset. The matching relationships between demand features and resource attributes are sorted according to the correlation strength value to generate a priority matching dataset; Integrate the priority matching dataset and the feature-level association dataset, supplement the logical support description for each matching relationship, describe the association basis and adaptation scenarios of demand characteristics and resource attributes, and generate a preliminary user demand resource association table. Remove duplicate matching relationships and invalid association entries from the initial user requirement resource association table, and standardize the data format and expression specifications to generate a standardized user requirement resource association table; Supplement the library resources with historical adaptation data and user borrowing feedback data. Use the library resources’ historical borrowing data and user rating data to calculate the historical borrowing frequency and average rating corresponding to each matching relationship in the standardized user demand resource association table. Mark the matching relationships with frequencies and ratings below the corresponding thresholds as relationships to be verified, adjust the association strength values, correct unreasonable matching relationships, and generate a verified user demand resource association table. The user demand resource association table is classified and organized according to the user demand characteristic category and the library resource attribute category. The layout structure and query logic of the user demand resource association table are adjusted to generate a categorized user demand resource association table. Organize and categorize the user demand resource association table, supplement the dynamic adjustment instructions for the association relationship, determine the update conditions, update cycle and update process of the association relationship, and form the final user demand resource association table.

8. The intelligent recommendation method for library resources based on big data analysis according to claim 4, characterized in that, The preliminary sorting results are adjusted by incorporating resource update cycle information. Library resources with updates closer to the current time are moved to the next higher position. Simultaneously, the adaptation level is considered, and adjustment weights are allocated based on the number of features associated with the adaptation level and the length of time since the resource update. This generates a timeliness-adaptive sorting result, including: Extract the resource update cycle information of each library resource from the preliminary sorting results, record the first publication time, latest update time and update frequency of the resource, supplement the resource update content summary, and generate a library resource update information dataset; The library resources in the library resource update information dataset are independently sorted according to the latest update time, and updated resources with different update time spans are distinguished. The timeliness sorting position of each library resource is recorded, and the independent timeliness sorting results of resources are generated. The adaptation level information in the preliminary sorting results is analyzed, and the number, distribution location and proportion of library resources corresponding to each adaptation level are counted. The sorting distribution characteristics of resources at different adaptation levels are determined, and an adaptation level distribution dataset is generated. Establish a balance rule between adaptation level and resource timeliness. Standardize the number of associated features corresponding to the adaptation level to generate standardized associated feature quantity values. Set the adaptation weight to be positively correlated with the standardized associated feature quantity values. At the same time, standardize the length of time between the resource update time and the current time to generate standardized time length values. Set the timeliness weight to be negatively correlated with the standardized time length values ​​and generate a weight allocation scheme. According to the weight allocation scheme, the comprehensive score of each library resource is calculated. The comprehensive score is obtained by summing the values ​​corresponding to the adaptation weight and the timeliness weight, and a comprehensive score dataset of library resources is generated. The library resources in the preliminary ranking results are re-ranked according to their comprehensive scores from high to low. At the same time, resources with the same comprehensive score are ranked according to the relevance of the resource theme to generate an initial time-appropriate ranking result. Analyze the resource distribution of the initial timeliness adaptation sorting results, and count the number of resources of different adaptation levels and timeliness intervals appearing at a preset proportion position before the sorting sequence; set a maximum occurrence threshold for each level and interval, and for resource types that exceed the maximum occurrence threshold, adjust some of their resources to the end of the initial timeliness adaptation sorting results so that the distribution of each type of resource at the beginning of the initial timeliness adaptation sorting results meets the preset diversity ratio requirement, and generate distribution adjustment results; Adjust the order of resources in the distribution adjustment results. For library resources of the same adaptation level and the same time interval, sort them according to the degree of fit between the reading difficulty of the resources and the user's historical reading difficulty habits, so that the reading difficulty matches the user's preferences and generates difficulty adaptation sorting results. Supplementing the author background similarity information of the resources, the library resources in the difficulty-adapted ranking results are fine-tuned again, and the resources whose author background matches the user's historical preference for authors are moved forward to generate the final timeliness-adapted ranking results. The final timeliness adaptation sorting result is formatted, and the sorting logic is verified to meet the preset priority rules, the weight allocation to meet the weight allocation scheme, and the resource distribution to meet the preset diversity ratio requirements. The timeliness adaptation sorting result is then output based on the verification results.

9. The intelligent recommendation method for library resources based on big data analysis according to claim 5, characterized in that, The analysis identifies library resources in the effective interaction dataset that users have not interacted with. Combining resource adaptation levels, user needs, and resource attribute information, the analysis identifies the causes of adaptation discrepancies and generates a discrepancy analysis report, including: From the effective interaction dataset, library resources with zero user clicks, borrowing status of "not borrowed", collection status of "not collected", and page dwell time of less than a preset time threshold are selected to form a list of resources that users have not interacted with. The adaptation level, complete attribute information, sorting position and corresponding user borrowing intent tags of each library resource are recorded. The distribution of library resources across adaptation levels in the list of resources not interacted with by users is statistically analyzed. The proportion of resources not interacted with at each adaptation level is calculated. The differences in the non-interaction rates of different adaptation levels are compared, and a statistical dataset of non-interaction data at each adaptation level is generated. Compare the attribute information of non-interactive resources with the real-time user demand feature data one by one, calculate the matching degree between the attribute values ​​of resource theme, content dimension, reading difficulty and applicable scenario and the corresponding demand features in the real-time user demand feature data, mark the dimensions with matching degree lower than the preset matching threshold as mismatched dimensions, record the specific deviation points, and generate attribute demand deviation dataset. Analyze the position of uninterrupted resources in the recommendation sequence, count the number and proportion of uninterrupted resources in different ranking positions, analyze the impact of ranking position on user interaction behavior, calculate the ratio of the ranking position of uninterrupted resources to the total length of the recommendation sequence, and analyze the correlation between this ratio and the non-interaction rate, generating a dataset on the impact of ranking position. By combining the user's historical borrowing behavior unit sequence, comparing the differences in attributes, adaptation levels, and intent tag fit between the uninterrupted resources and the user's historical borrowed resources, we can analyze the main behavioral motivations for the user's lack of interaction and generate a behavioral motivation analysis dataset. The mapping relationship, adaptation level division criteria and sorting algorithm output in the user demand resource association table are compared with the actual user interaction records in the effective interaction dataset. The degree of match between the prediction and the actual result is calculated. The part with the degree of match below the preset threshold is marked as the system-level matching deviation, and a system matching deviation dataset is generated. Integrate the non-interactive statistical data set of the matching hierarchy, the attribute requirement deviation dataset, the ranking position influence dataset, the behavioral motivation analysis dataset, and the system matching deviation dataset, and sort out the specific manifestations, scope of influence, and core causes of various deviations; Classify and statistically analyze the causes of various deviations, calculate the number and proportion of non-interactive resources corresponding to each cause of deviation, distinguish between primary and secondary causes of deviation, and generate statistical analysis results of deviation causes. For different causes of deviation, analyze the degree and scope of the deviation's impact on the resource recommendation effect and the direction for subsequent improvement, generate a deviation correction plan, and generate a deviation impact assessment report; Compile the statistical analysis results of the causes of deviations and the deviation impact assessment report, and write them according to the type of deviation, degree of impact and correction ideas to form a deviation analysis report.

10. A library resource intelligent recommendation system based on big data analysis, characterized in that, The system includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the memory to implement the intelligent library resource recommendation method based on big data analysis as described in any one of claims 1-9.

Images