A prediction-based shelving method based on discipline aging heterogeneity driving

Abstract

The application relates to the field of library resource management, and specifically discloses a prediction type shelving method based on discipline aging heterogeneity driving, which comprises the following steps: S1, data acquisition and preprocessing; S2, aging rate modeling and verification; S3, construction of a literature attenuation law prediction model; S4, selection of parameter combinations; S5, four-quadrant discipline classification; S6, fine sorting of discipline resource shelving priorities; and S7, generation of a layered shelving strategy. The application is based on the Bernan literature aging negative exponential model, parameters of each discipline book are calculated, the future attenuation parameter prediction method based on the GBRT algorithm is used to calculate the optimal parameters of each discipline, the disciplines are divided into four quadrants by taking the global mean value as a threshold, and a layered weighted evaluation model is constructed based on the discipline characteristics, so that the fine sorting of the discipline resource shelving priorities is realized through a weight mechanism. The application breaks through the limitations of traditional static management and realizes the efficient allocation of library resources.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of library resource management, in particular to a predictive shelving method based on discipline aging heterogeneity driving. BACKGROUND

[0002] Under the background of digital transformation, paper books are still an important part of the resource system of university libraries. However, the lack of physical space and aging problems pose a serious challenge to the service efficiency of the library.

[0003] In terms of management strategy, traditional literature management strategies are mostly based on fixed periods or experience, ignoring the differences in knowledge update rates of different disciplines, which leads to the misremoval of high-value literature or the long-term occupation of space by low-efficiency resources.

[0004] In terms of evaluation indicators, existing technologies focus on a single indicator (such as borrowing frequency or publication age), do not integrate the dynamic characteristics of literature aging, ignore the impact of literature aging rate on long-term utilization value, and the single indicator strategy cannot quantify the dynamic characteristics of literature aging, resulting in low space allocation efficiency (such as popular new books occupying core areas but quickly becoming obsolete).

[0005] In terms of model construction, existing research mainly constructs aging models based on journal citation data, such as Price index and half-life model, but there are significant differences in time dimension and behavior motivation between borrowing behavior of paper literature and journal behavior. For example, classic theory books in the humanities and social sciences may be borrowed for a long time, while in the engineering and technology field, they quickly age due to rapid technological iteration. Existing research has not established a quantitative priority model, making it difficult to dynamically respond to the development trend of disciplines.

[0006] The existing technology has the following key problems: static model has poor adaptability, cannot accurately capture the differences in aging rates between disciplines, leading to space waste or misremoval of resources; data utilization is one-sided, relying on a single indicator, lacking a multi-parameter dynamic optimization mechanism, making it difficult to balance immediate needs and long-term value; lack of cross-disciplinary collaboration, no quantitative correlation framework between discipline priority and space allocation, making it difficult to achieve precise resource scheduling.

[0007] Therefore, there is an urgent need for a paper literature predictive classification shelving method based on discipline aging heterogeneity driving that can accurately capture the differences in aging rates between disciplines and integrate multi-dimensional indicators to build a dynamic optimization mechanism, in order to improve the space allocation efficiency and service efficiency of the library. SUMMARY

[0008] The present application provides a predictive shelving method based on discipline aging heterogeneity driving to solve the problems of lack of space and literature aging in academic libraries, effectively improving the precision of space allocation, the conversion rate of long-term utilization value of literature, and the dynamic adaptation efficiency of discipline services.

[0009] To achieve the above object, the application provides a prediction-based shelving method based on discipline aging heterogeneity driving, comprising:

[0010] S1, data acquisition and preprocessing: extracting the collection year and borrowing records of multi-disciplinary paper books, constructing time series data set according to discipline classification, and eliminating abnormal point data;

[0011] S2, aging rate modeling and verification: based on the Bernoulli document aging negative exponential model, constructing a book borrowing rate decay analysis framework and calculating the half-life, using the residual standard error RMSE to evaluate the goodness of model fitting, and verifying the statistical significance of the initial borrowing rate k and the decay coefficient b through F test;

[0012] S3, constructing a document decay law prediction model: using time series gradient boosting regression tree GBRT algorithm to predict the document borrowing rate k0 and decay coefficient b0 in the next three years;

[0013] S4, selecting parameter combination: using equal weight average method to calculate the optimal borrowing rate k1 and decay coefficient b1 of each discipline, and taking the mean value of the prediction results in the next three years as the shelving parameters of each discipline;

[0014] S5, four-quadrant discipline classification: taking the global mean and as the threshold, the disciplines are divided into four quadrants;

[0015] S6, fine sorting of discipline resource shelving priority: based on the characteristics of the disciplines, a hierarchical weighted evaluation model is constructed, and the fine sorting of the discipline resource shelving priority is realized through the weight mechanism;

[0016] S7, generating hierarchical shelving strategy: according to the priority score, the disciplines are divided into first-line, second-line, and third-line, and the resource allocation is dynamically adjusted.

[0017] Preferably, in S2, based on the Bernoulli document aging negative exponential model, the mathematical expression for constructing the book borrowing rate decay analysis framework is:

[0018] ;

[0019] In the formula, C(t) is the book borrowing rate, i.e. the number of borrowed copies per number of books; t is the book age, i.e. the borrowing year minus the collection year; k is the initial borrowing rate, i.e. the reference value when t=0; b is the decay coefficient, indicating the speed of the book borrowing rate decreasing with time.

[0020] Preferably, in S2, the calculation formula of the half-life T1 / 2 is:

[0021] .

[0022] Preferably, in S5, the four quadrants for dividing the discipline are high k-value - high b-value, low k-value - high b-value, low k-value - low b-value, and high k-value - low b-value.

[0023] Preferably, in S6, the disciplinary characteristics include resource utilization efficiency, knowledge obsolescence rate, and disciplinary factors.

[0024] Preferably, in S6, the specific steps for refining the priority ranking of subject resources through a weighting mechanism include processing the initial borrowing rate k value through maximum-minimum normalization and numerical standardization of the inverse half-life, wherein the initial borrowing rate normalization value k... norm The calculation formula is:

[0025] ;

[0026] Standardized value of reciprocal half-life h lnorm The calculation formula is:

[0027] ;

[0028] In the formula, It is the minimum value among the optimal borrowing rates for all subjects. It is the maximum value among the optimal borrowing rates for all subjects. It is the minimum value of the set of reciprocals of the optimal half-life for all disciplines. It is the maximum value of the set of reciprocals of the optimal half-life for all disciplines.

[0029] Preferably, in S6, the calculation formula for constructing the hierarchical weighted evaluation model based on subject characteristics is as follows:

[0030] ;

[0031] In the formula, Δ wdiscipline This is a subject weighting adjustment item.

[0032] Preferably, the subject weight adjustment term Δ wdiscipline The weighting ratio is set according to the actual needs of the institution.

[0033] Preferably, in S7, first-tier subjects are open bookshelves, second-tier subjects are thematic display areas, and third-tier subjects are dense storage areas.

[0034] Preferably, in S7, first-tier subject resources are deployed in high-traffic areas and equipped with self-service borrowing and returning equipment; second-tier subject resources are located using RFID tags and allocated on demand; and third-tier subject resources are allocated by reservation system and by region.

[0035] Therefore, this invention proposes a predictive shelving method based on the heterogeneity of discipline aging, which has the following advantages:

[0036] (1) The application predicts and dynamically extracts the initial borrowing rate and decay coefficient of the subject literature by the time series gradient boosting regression tree algorithm, effectively solves the problem that the traditional fixed cycle or empirical rule cannot adapt to the subject heterogeneity, and significantly improves the flexibility and adaptability of resource management.

[0037] (2) The application constructs a four-quadrant classification system combining the initial borrowing rate and decay coefficient, realizes the quantitative balance of resource immediate demand and long-term value, accurately identifies high-value literature and inefficient resources, and provides strong support for fine management and space optimization.

[0038] (3) The application uses the priority score recalculation and half-life trend analysis module to support resource level dynamic adjustment, can optimize the storage location in real time according to the literature borrowing trend and aging rate, ensure efficient use of resources, and has strong practicality and operability.

[0039] The technical solutions of the application will be further described in detail below with the help of the drawings and examples. DETAILED DESCRIPTION

[0040] Figure 1 is a four-quadrant distribution of subject book aging of the prediction-based shelving method based on subject aging heterogeneity driving. DETAILED DESCRIPTION

[0041] In order to make the technical solutions, advantages and purposes of the application clearer, the technical solutions of the embodiments of the application will be clearly and completely described below. The described embodiments are part of the embodiments of the application, not all the embodiments. Based on the described embodiments of the application, all other embodiments obtained by those of ordinary skill in the art without creative labor belong to the protection scope of the present application.

[0042] Unless otherwise defined, the technical terms or scientific terms used in the application should be understood as the usual meaning understood by those skilled in the art in the field of the application.

[0043] As shown in Figure 1 The application provides a prediction-based shelving method based on subject aging heterogeneity driving, which comprises:

[0044] S1, data acquisition and preprocessing: extracting the collection year and borrowing record of multi-subject paper books, constructing a time series data set according to subject classification, and eliminating abnormal point data;

[0045] S2, aging rate modeling and verification: based on Bernan literature aging negative exponential model, the construction of book borrowing rate decay analysis framework and calculation of half-life, using residual standard error RMSE to evaluate the goodness of fit of the model, through F test to verify the statistical significance of the initial borrowing rate k and decay coefficient b;

[0046] In S2, based on Bernan literature aging negative exponential model, the mathematical expression of the construction of book borrowing rate decay analysis framework is:

[0047] ;

[0048] In the formula, C(t) is the book borrowing rate, i.e. the number of borrowed books per number of books; t is the book age, i.e. the borrowing year minus the collection year; k is the initial borrowing rate, i.e. the benchmark value when t=0; b is the decay coefficient, which represents the speed of decline of book borrowing rate over time;

[0049] The formula for calculating the half-life T1 / 2 is:

[0050] ;

[0051] S3, construction of literature decay law prediction model: using time series gradient boosting regression tree GBRT algorithm to predict the literature borrowing rate k0 and decay coefficient b0 in the next three years;

[0052] S4, selection of parameter combination: using equal weight average method to calculate the optimal borrowing rate k1 and decay coefficient b1 of each discipline, and taking the mean value of the prediction results in the next three years as the shelving parameters of each discipline;

[0053] S5, four quadrant discipline classification: taking the global mean and as the threshold, the disciplines are divided into four quadrants, among which the four quadrants of discipline division are high k value-high b value, low k value-high b value, low k value-low b value, and high k value-low b value;

[0054] S6, fine sorting of discipline resource shelving priority: based on the characteristics of disciplines, a hierarchical weighted evaluation model is constructed, and the fine sorting of discipline resource shelving priority is realized through the weight mechanism; the characteristics of disciplines include resource utilization efficiency, knowledge aging speed and discipline factors;

[0055] In S6, the specific steps of fine sorting of discipline resource shelving priority through the weight mechanism include maximum-minimum normalization of initial borrowing rate k value and numerical standardization of half-life reciprocal, wherein the calculation formula of initial borrowing rate normalized value k norm is:

[0056] ;

[0057] Half-life reciprocal normalized value h lnorm The calculation formula is:

[0058] ;

[0059] In the formula, is the minimum value in the optimal borrowing rate set of all disciplines, is the maximum value in the optimal borrowing rate set of all disciplines, is the minimum value in the optimal half-life reciprocal set of all disciplines, is the maximum value in the optimal half-life reciprocal set of all disciplines.

[0060] The calculation formula of the weighted score model is:

[0061] ;

[0062] In the formula, Δ wdiscipline is the discipline weight adjustment term, and the weight proportion of the discipline weight adjustment term Δ wdiscipline is set according to the actual needs of the institution;

[0063] S7, generate a hierarchical shelving strategy: divide the disciplines into the first line, the second line, and the third line according to the priority score, and dynamically adjust the resource allocation.

[0064] Among them, the first-line discipline is the open shelf, the second-line discipline is the theme display area, and the third-line discipline is the intensive storage area; the first-line discipline resource is deployed in the high-flow area and is equipped with self-service borrowing and returning equipment, the second-line discipline resource adopts the RFID tag positioning and on-demand allocation mechanism, and the third-line discipline resource adopts the reservation system and resource allocation by area.

[0065] Embodiment one:

[0066] (1) Data collection and preprocessing:

[0067] The data source is the paper book statistical data extracted from the Library Huanyan Literature Information Service System Libsys5.6, and the knowledge organization is strictly followed according to the “Chinese Library Classification (Fifth Edition)” for 22 discipline categories.

[0068] The data collection range covers the complete collection period from 2001 to 2023, including: (1) The collection year is set as January 1, 2001 as the starting point of the time series to exclude systematic bias caused by early information missing, and the number of books collected in each collection year is derived; (2) Borrowing data is taken to December 31, 2023 to ensure the integrity of the observation window, borrowing records are counted by collection year, and borrowing records from 2013 to 2020 and 2022 to 2023 are used in the study, with 2021 data excluded due to data interruption. For example, to count the borrowing records in 2013, the borrowing times of the collections in 2001-2013 in 2013 need to be derived, and so on. The Python 3.12.4 programming environment (Anaconda) is used to complete the algorithm implementation.

[0069] (2) Aging rate modeling and verification:

[0070] Based on the Berner literature aging negative exponential model, the borrowing rate decay law of Chinese paper books in 22 major disciplines from 2013 to 2023 (excluding 2021) is nonlinearly fitted. The parameter estimation is completed by the scipy.optimize.curve_fit function of Python 3.12.4, and the model convergence is verified by the residual standard error RMSE to ensure the reliability of the fitting results.

[0071] The F test statistics of all disciplines are significantly higher than the F critical value F Critical =4.32, and the "Significant Model" index is TRUE (model significance index p<0.05), indicating that the negative exponential model has statistical significance for the borrowing decay law of all disciplines.

[0072] The t values of the initial borrowing rate k and the decay rate b of all disciplines are significantly higher than the t critical value (t critical =2.08).

[0073] The R² mean of the 22 major disciplines is 0.75, and the model can explain more than 75% of the borrowing rate variation. The minimum RMSE is 0.005, the maximum is 0.18, and the mean is 0.05.

[0074] (3) Constructing literature decay law prediction model:

[0075] The time series gradient boosting regression tree GBRT algorithm is used to predict the literature borrowing rate k0 and decay coefficient b0 in 2024, 2025 and 2026.

[0076] Social sciences are set with more flexible parameters (max_depth=4, subsample=0.9), while natural sciences are set with stronger regularization (max_depth=3, subsample=0.8) to control overfitting. Predictions are limited within 20% of the historical 15-85th percentile range to avoid extreme extrapolation.

[0077] RMSE (Root Mean Square Error) and MAPE (Mean Absolute Percentage Error) are used to compare the 2024 predictions with the actual values of the same year, and the evaluation is divided into two levels: discipline and discipline category (A-K are classified as social sciences and N-Z are classified as natural sciences according to the Chinese Discipline Classification Standard).

[0078] The overall prediction accuracy of k0 values is excellent: the MAPE of all disciplines is 11.9%, among which the social sciences is 10.8% and the natural sciences is 12.9%, both lower than the usual 20% good threshold set in the field of information and emotion. The half-life prediction shows obvious differences between disciplines: the MAPE of natural sciences is 13.9%, still within the "good" range; the MAPE of social sciences is as high as 40.4%, exceeding the acceptable range. This difference is mainly due to the large-scale adjustment (deletion and relocation) of paper collections in social sciences during the library reconstruction in 2024, resulting in structural mutations in borrowing base and half-life calculation; while the decay characteristics of newly acquired literature are not affected, so the prediction of k0 values remains stable. The RMSE of k0 and b0 also shows the same direction: the RMSE of b0 value in social sciences (0.058) is about 1.6 times that of natural sciences (0.037), indicating that the fluctuation of borrowing base is the main cause of error amplification.

[0079] (4) Selection of parameter combinations and four-quadrant discipline classification:

[0080] For each discipline, the parameter combination is obtained by directly taking the average of the three-year prediction results of 2024, 2025, and 2026 for k0 and b0.

[0081] Based on the optimal parameters of 22 disciplines, see Table 1. The global mean ( ) is used as the boundary to construct a four-quadrant classification system, as shown in Figure 1 .

[0082] Table 1 Optimal parameters of disciplines and quadrant classification:

[0083] ;

[0084] First quadrant (I): high k value, high b value (k1≥0.3, b1≥0.13);

[0085] Characteristics: initial demand is strong but quickly declines, requires high-frequency updates, and prioritizes digital resource access rights;

[0086] Disciplines: C (General Social Science), H (Language), I (Literature), N (General Natural Science), T (Industry Technology), U (Transportation). For example, N (General Natural Science), k = 0.43, b = 0.15, the interdisciplinary knowledge reorganization accelerates, and the emerging cross-field (such as AI + biology) literature is rapidly replaced.

[0087] Second quadrant (II): low k value, high b value (k1<0.3, b1≥0.13);

[0088] Features: limited demand and rapid obsolescence, need for accurate procurement;

[0089] Disciplines: F (Economics), P (Astronomy / Earth), S (Agricultural Science), X (Environmental Science). For example, F (Economics), the demand for classic works of economic theory is enduring, and the emerging economic field (such as digital economy) literature is updated faster but the core theory is stable.

[0090] Third quadrant (III): low k value, low b value (k1<0.3, b1<0.13);

[0091] Features: stable demand and long-term impact, need to extend the preservation period and build a special knowledge base;

[0092] Disciplines: A (Marxist Theory), D (Politics and Law), E (Military), G (Culture and Sports), J (Arts), K (History and Geography), R (Medicine and Health), V (Aerospace), Z (Comprehensive Books), among them, A (Marxist Theory), classic theory works demand stable.

[0093] Fourth quadrant (IV): high k value, low b value (k1≥0.3, b1<0.13);

[0094] Features: strong and enduring demand, ensure paper resource reserves, optimize the distribution of copies and the layout of reading rooms;

[0095] Disciplines: B (Philosophy), O (Mathematical Science), Q (Biological Science), among them, B (Philosophy), the works of this discipline are classic and have a long-lasting impact.

[0096] (5) Fine sorting of subject resource shelving priority:

[0097] In the index system and weight design, the weight of the first index resource utilization efficiency and knowledge aging rate is 50% respectively, among them, the initial borrowing rate (k value) in the Berner model represents the resource utilization efficiency, and the standardized value of the inverse of half-life (1 / T1 / 2) represents the knowledge aging speed (the smaller the half-life, the faster the aging speed).

[0098] The secondary indicators increase the weight of A-class disciplines by 30%, taking into full consideration the 12 A-class disciplines of Tongji University, including civil engineering, environmental science and engineering, and urban planning. The 12 disciplines are classified into 6 major disciplines (T, X, F, J, O, and U), and a 30% weight increase is implemented for the 6 major disciplines containing A-class disciplines.

[0099] The initial borrowing rate k is subjected to maximum-minimum normalization and numerical standardization of the half-life reciprocal, where the normalized value of the initial borrowing rate k norm is calculated by the formula:

[0100] ;

[0101] The normalized value of the half-life reciprocal h lnorm is calculated by the formula:

[0102] ;

[0103] wherein is the minimum value in the optimal borrowing rate set of all disciplines, is the maximum value in the optimal borrowing rate set of all disciplines, is the minimum value in the optimal half-life reciprocal set of all disciplines, is the maximum value in the optimal half-life reciprocal set of all disciplines.

[0104] The calculation formula of the weighted scoring model is:

[0105] ;

[0106] wherein Δ wdiscipline is the discipline weight adjustment term. The data is subjected to standardization processing and the comprehensive scores of each discipline are obtained by using the above formula. The resource shelving priority of the 22 disciplines is calculated by the above method, as shown in Table 2.

[0107] Table 2 Resource shelving priority of 22 disciplines:

[0108] ;

[0109] (6) Generation of hierarchical shelving strategy:

[0110] The hierarchical preservation and management strategy is designed in combination with data-driven decision-making and spatial optimization principles, as shown in Table 3 and Table 4.

[0111] Table 3 Preservation hierarchical standards:

[0112] ;

[0113] Table 4 Management strategy of each preservation level:

[0114] ;

[0115] Therefore, the application provides a prediction-based shelving method based on subject aging heterogeneity driving, which dynamically extracts subject literature characteristics, constructs a multi-dimensional evaluation system, and designs a dynamic shelving mechanism, accurately adapts to subject differences, optimizes space allocation, improves resource utilization efficiency, realizes the leap of literature storage from "experience driving" to "precise data driving", solves the core pain points such as space waste, high misretirement rate and lagging response in traditional management, and has significant academic value and industrial application potential.

[0116] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand: its still can be modified or equivalent to replace the technical scheme of the present application, and these modifications or equivalent replacements also cannot make the modified technical scheme deviate from the spirit and scope of the technical scheme of the present application.

Claims

1. A discipline-aging heterogeneity driven predictive shelving method, comprising: Comprising: S1, data collection and preprocessing: extract the collection year and borrowing records of multidisciplinary paper books, construct time series data set according to subject classification, and eliminate abnormal point data; S2, aging rate modeling and verification: based on Bernoulli document aging negative exponential model, construct book borrowing rate decay analysis framework and calculate half-life, use residual standard error RMSE to evaluate model fitting goodness, and verify statistical significance of initial borrowing rate k and decay coefficient b through F test; S3, construct document decay law prediction model: use time series gradient boosting regression tree GBRT algorithm to predict future three years of document borrowing rate k0 and decay coefficient b0; S4, select parameter combination: calculate the optimal borrowing rate k1 and decay coefficient b1 of each subject using equal weight average method, and take the mean value of the prediction results in the next three years as the subject cataloging parameters; S5. Four-quadrant subject classification: divide subjects into four quadrants with global mean as threshold and as threshold. S6, fine sorting of subject resource cataloging priority: based on subject characteristics, construct a hierarchical weighted evaluation model, and realize fine sorting of subject resource cataloging priority through weight mechanism; S7, generate hierarchical cataloging strategy: divide the subjects into first-line, second-line and third-line according to priority score, and dynamically adjust resource allocation.

2. The subject-based aging heterogeneity driven predictive shelving method of claim 1, wherein, In S2, based on Bernoulli document aging negative exponential model, the mathematical expression of constructing book borrowing rate decay analysis framework is: ； In the formula, C(t) is the book borrowing rate, i.e. the number of borrowed copies per number of books; t is the book age, i.e. the borrowing year minus the collection year; k is the initial borrowing rate, i.e. the benchmark value when t=0; b is the decay coefficient, which represents the speed of book borrowing rate decline over time.

3. The subject-based aging heterogeneity driven predictive shelving method of claim 1, wherein, In S2, the calculation formula of half-life T1 / 2 is: 。 4. The subject-based aging heterogeneity driven predictive shelving method of claim 1, wherein, In S5, the four quadrants of subject division are high k value-high b value, low k value-high b value, low k value-low b value, and high k value-low b value.

5. The subject-based aging heterogeneity driven predictive shelving method of claim 1, wherein, In S6, the subject characteristics include resource utilization efficiency, knowledge aging speed, and subject factors.

6. The subject-based aging heterogeneity driven predictive shelving method of claim 1, wherein, In S6, the specific steps of realizing the fine sequencing of the subject resource cataloging priority through the weight mechanism include the processing of the maximum-minimum normalization of the initial borrowing rate k value and the numerical standardization of the half-life reciprocal, wherein the normalized value k of the initial borrowing rate norm The calculation formula is: ； Half-life reciprocal normalized value h lnorm The calculation formula is: ； wherein is the minimum of the set of optimal borrowing rates for all subjects, is the maximum of the set of optimal borrowing rates for all subjects, is the minimum of the set of optimal inverse half-lives for all subjects, is the maximum of the set of optimal inverse half-lives for all subjects.

7. The subject-based aging heterogeneity driven predictive shelving method of claim 6, wherein, In S6, the calculation formula of constructing hierarchical weighted evaluation model based on subject characteristics is: ； In the formula, is a subject weight adjustment term.

8. The subject-based aging heterogeneity driven predictive shelving method of claim 1, wherein, In S7, the first-line subject is open shelf, the second-line subject is theme display area, and the third-line subject is intensive storage area.

9. The subject-based aging heterogeneity driven predictive shelving method of claim 1, wherein, In S7, the first-line subject resource is deployed in high-flow area and equipped with self-service borrowing and returning equipment, the second-line subject resource uses RFID tag positioning and on-demand allocation mechanism, and the third-line subject resource uses reservation system and resource allocation by area.

Images