A digital library resource sharing method and system based on Bayesian network

Abstract

The present invention discloses a digital library resource sharing method and system based on Bayesian network, including: collecting sharable data and historical reading data of local resources by users; classifying the local resources and the sharable data to obtain classified data, obtaining preference data based on the historical reading data, and obtaining predicted preference data by time series analysis; obtaining a user preference index of the classified data based on the preference data and the predicted preference data, calculating a quality index of the sharable data, and obtaining a sharing index based on the user preference index and the quality index; establishing a resource sharing model based on the sharing index by using a Bayesian network algorithm, and inputting the data to be shared into the resource sharing model to obtain a sharing result. The present invention realizes efficient sharing of digital library resources by intelligently analyzing comprehensive user preferences and resource quality.

Description

Technical Field

[0001] The invention relates to the field of digital library resource management, and in particular to a digital library resource sharing method and system based on Bayesian network. Background Art

[0002] In the field of digital library resource sharing, traditional resource management methods mainly rely on static classification and basic retrieval technology, which often cannot effectively reflect users' personalized reading preferences and are difficult to adapt to the dynamic changes in resource timeliness and quality. In addition, existing systems often face problems of low efficiency and long response time when processing large-scale data and user requests, which limits the breadth and depth of resource sharing. Therefore, there is an urgent need for an intelligent resource sharing method that can integrate user behavior analysis, resource quality and timeliness evaluation to improve resource utilization efficiency and user satisfaction. Summary of the invention

[0003] The purpose of the present invention is to provide a digital library resource sharing method and system based on Bayesian network.

[0004] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0005] The present invention comprises the following steps:

[0006] Collect sharable data and historical reading data of users on local resources;

[0007] Classifying the historical reading data of the local resource and the sharable data to obtain classified data, obtaining preference data based on the historical reading data, and obtaining predicted preference data by using time series analysis;

[0008] Obtaining a user preference index of the classified data based on the preference data and the predicted preference data, calculating a quality index of the shareable data, and obtaining a sharing index based on the user preference index and the quality index;

[0009] A Bayesian network algorithm is used to establish a resource sharing model based on the sharing index, and the data to be shared is input into the resource sharing model to obtain a sharing result.

[0010] Furthermore, the sharable data and the historical reading data include:

[0011] The shareable data includes the shareable digital resources provided by the cooperation platform and their user ratings, number of citations and publication time. The historical reading data includes the existing digital resources on the platform and their reading time, number of downloads and publication time.

[0012] Furthermore, the method of classifying the historical reading data of the local resource and the sharable data to obtain classified data includes:

[0013] Use TF-IDF to convert the historical reading data of local resources and the digital resources in the shareable data into feature vectors, and cluster the feature vectors:

[0014] Calculate the similarity between two feature vectors through semantic similarity:

[0015]

[0016] Among them, S(d i ,d j ) is the measure d i and d j The similarity function of the similarity between i and d j Represent two different eigenvectors, w i,f is d i The weight of feature item f is the TF-IDF value, where f is the keyword in the digital resource, Φ(d i ,d j ) is d i and d j The common feature set, α f is the coefficient used to adjust the importance of feature f based on mutual information, ψ is the Sigmoid function, TF i,f is d i The term frequency of the characteristic term f in , φ is an exponential function, φ(x) = e kx , k is a constant used to control the growth rate of the exponential function, determined by cross-validation,

[0017] Initialization of cluster centers:

[0018]

[0019] Among them, μ k is the cluster center of cluster k, d m Yes D k The eigenvector in , w dm is d m The initial weight, D k is the set of feature vectors initially assigned to cluster k, ν is the regularization constant used to avoid the denominator being zero, ν∈[10 -16 ,10 -6 ],

[0020] Cluster center update:

[0021] C k (t+1)={d|S(d,μk )≥τ}

[0022]

[0023] Among them, C k (t+1) is the set of feature vectors belonging to cluster k at the t+1th iteration, τ is the similarity threshold, and is the average similarity between feature vector pairs. is the gradient of the cluster center, p(C k (t)|μ k ) is the clustering conditional probability, η(t) is the learning rate,

[0024] Neighborhood function and eigenvector weight updates:

[0025]

[0026]

[0027] in is the eigenvector d at the tth iteration m The neighborhood function, σ(t) is the neighborhood size, which decreases with time, δ(t) is the impact factor of the tth iteration, δ(t) = δ0·e -λt , δ0 is the initial impact factor, λ is the decay rate, is the eigenvector d of the t+1th iteration m The weight of

[0028] Clustering quality assessment and optimization:

[0029]

[0030] Among them, Q is the comprehensive clustering quality index, k and k′ represent the indexes of different clusters, and the comprehensive clustering quality index is maximized by adjusting the weight of the feature vector, the cluster center and the neighborhood function.

[0031] Furthermore, the method of obtaining the preference data based on the historical reading data and obtaining the predicted preference data by time series analysis includes:

[0032] Remove the data with reading time less than the preset effective threshold from the historical reading data, and calculate the preference data of local resources based on the number of downloads, published time and historical reading data:

[0033]

[0034] Among them, Re represents the preference data of local resources, w1, w2, w3, w4 are weight coefficients used to adjust the contribution of different factors to the preference data, D is the number of downloads, R represents the historical reading volume of local resources, which is obtained based on historical reading data, and P is the published time. Indicates the average reading time of users for this local resource, e -λP is an exponential decay term, λ is the decay rate, which is used to simulate the gradual decrease in the popularity of local resources over time.

[0035] Use preference data and historical reading data to build a time series analysis model:

[0036] Optimization of multi-step forecast errors of VAR models:

[0037]

[0038] Among them, MSE multi-step is the mean square error of the multi-step forecast, T is the length of the time series, Represents the model coefficients of the time series analysis model, y j,t is the jth time series data point, represents the forecast data, m is the number of variables in the time series, p is the lag order,

[0039] Feature selection and model complexity control of SVR model:

[0040]

[0041] Among them, y i,t is the sample data at time point t and sample index i, n is the number of samples, ∈ is the noise term, w k is the weight corresponding to the kth feature, d is the dimension of the feature, γ feature is the regularization coefficient for feature selection,

[0042] Optimization of timing analysis model:

[0043]

[0044] in, is the prediction result of the time series analysis model, κ is the weight for balancing the predictions of the SVR model and the VAR model, and γ VAR is the regularization coefficient of the VAR model, R VAR is the regularization function of the VAR model, and the prediction results are used as the prediction preference data.

[0045] Furthermore, the method for obtaining the user preference index of the classification data based on the preference data and the predicted preference data includes:

[0046] User Preference Index:

[0047]

[0048] where Y qis the user preference index of the category of the qth classification data, is the preference data of the jth digital resource in the qth category, J is the number of digital resources in category q, is the predicted preference data of the j-th digital resource in the q-th category, s1 is the weight of the preference data, s2 is the weight of the predicted preference data, and s1+s2=1.

[0049] Furthermore, the method for calculating the quality index of the sharable data includes:

[0050] The quality index of shareable data is calculated by user ratings, number of citations, and publication time:

[0051]

[0052] Among them, Z represents the quality index of the sharable data, a1, a2, a3 are weight factors used to adjust the contribution of different influencing factors to the quality index, Y is the number of downloads, t is the current time, τ i H i timestamp, Δt is the time window parameter, α is the parameter for controlling time sensitivity, which is a positive real number, H i represents the i-th user rating of the sharable data, I is the number of user ratings of the sharable data, and e -λP is the exponential decay term, and λ is the decay rate.

[0053] Furthermore, the method for obtaining the sharing index based on the user preference index and the quality index includes:

[0054] Perform logarithmic transformation on user ratings, citation counts, download counts, published time, and historical reading data to establish a data set. The user ratings, citation counts, download counts, and published time in the data set are used as the user preference index set. The user ratings, citation counts, and published time in the data set are used as the quality index set. The weights of the user preference index and the quality index are:

[0055]

[0056] where g v is the contribution value of subset v, S is the data set, x is the data point in S, μ is the mean of S, v is the index of the subset, V is the set of subsets, represents the set of user preference index set and quality index set, ω v represents the weight of subset v,

[0057] Sharing index of shareable data:

[0058] G=ω1Y+ω2Z

[0059] Where G is the sharing index, ω1 is the weight of the user preference index, ω2 is the weight of the quality index, and Y is the user preference index of the classification category where the shareable data is located.

[0060] Furthermore, a method for establishing a resource sharing model based on the sharing index using a Bayesian network algorithm includes:

[0061] Share the sharable data whose sharing index is greater than the preset sharing threshold. Establish the sharing label of the sharable data based on the sharing index and the sharing threshold. 1 means sharing and 0 means not sharing. Normalize and preprocess the user preference index, user rating, number of citations and published time as input features. Use the sharing label as the child node and the input feature as the parent node to establish a directed acyclic graph. Use the uniform prior distribution as the prior condition to train the resource sharing model:

[0062] Predicted probability distribution:

[0063]

[0064] Among them, P represents the predicted probability distribution, Y is the response variable, is the shared label, X is the input feature, E is the training set, Q is the variational distribution of the model parameter θ, λ is the adjustment parameter of the error term, θ prior is the a priori parameter estimate, θ prior A normal distribution centered on

[0065] Goodness of fit:

[0066] G(E, μ, θ) = Σlog[q(v i |θ,∈)·R(v i )]-μ∑log[H(v i )I(v i )]-∑logP(E|v i ,∈)

[0067] Among them, G represents the goodness of fit function, v i is the i-th weight in the neural network, q is the variational distribution, H is the uniform prior distribution, I is the likelihood function, μ is the weight of H and I in the goodness of fit function, ∈ is the uncertainty parameter, R(v i ) is v i The regularization function of

[0068] Probability density function:

[0069]

[0070] Where f is the probability density function, Y pred is the model prediction value, σ is the noise term, α(Y, Ypred ,σ) is the predicted value Y pred The adjustment function of the noise term is the Sigmoid function,

[0071] Goodness of fit for prediction uncertainty:

[0072] U(E, μ, θ) = Σlog[q(v i |θ,e)·R(v i )]-μΣlog[H(v i )I(v i )]∑log[r(Y|Y pred ,σ,α)]

[0073] Wherein, U represents the goodness of fit considering the prediction uncertainty, and the sharable data will be shared when the prediction probability distribution of Y=1 is greater than or equal to 50%.

[0074] A digital library resource sharing system based on Bayesian network, characterized by comprising:

[0075] The collection module collects sharable data and historical reading data of users on local resources;

[0076] A calculation module, which obtains preference data based on the historical reading data, obtains predicted preference data by time series analysis, obtains a user preference index of the classified data based on the preference data and the predicted preference data, calculates a quality index of the shareable data, obtains a sharing index based on the user preference index and the quality index, and establishes a resource sharing model based on the sharing index by using a Bayesian network algorithm;

[0077] A classification module, classifying the historical reading data of the local resource and the sharable data to obtain classified data;

[0078] The output module inputs the data to be shared into the resource sharing model to obtain the sharing result.

[0079] The beneficial effects of the present invention are:

[0080] The present invention can dynamically adjust resource sharing strategies according to users' reading behaviors and preferences, and quantitatively evaluate the quality of resources to ensure that shared resources meet user needs and have high sharing value. Through this method, digital libraries can manage and utilize their resources more efficiently and provide users with more accurate and personalized services. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 The present invention is a flow chart of the steps of a digital library resource sharing method and system based on Bayesian network. DETAILED DESCRIPTION

[0082] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0083] A digital library resource sharing method and system based on Bayesian network of the present invention comprises the following steps:

[0084] like Figure 1 As shown, in this embodiment, the following steps are included:

[0085] Collect sharable data and historical reading data of users on local resources;

[0086] In the actual evaluation, the collected data include:

[0087] Historical reading data: User 1 read digital resource R001, the reading time was 120 minutes, the number of downloads was 5, R001 was released 30 days ago, User 1 read digital resource R002, the reading time was 45 minutes, the number of downloads was 2, R002 was released 45 days ago, User 2 read digital resource R003, the reading time was 90 minutes, the number of downloads was 3, R003 was released 30 days ago,

[0088] Shareable data: Digital resource G001 has an average user rating of 4.5 points, has been cited 15 times, and has been published for 60 days. Digital resource G002 has an average user rating of 3.8 points, has been cited 10 times, and has been published for 45 days. Digital resource G003 has an average user rating of 3.2 points, has not been cited, and has been published for 90 days.

[0089] Classifying the historical reading data of the local resource and the sharable data to obtain classified data, obtaining preference data based on the historical reading data, and obtaining predicted preference data by using time series analysis;

[0090] In the actual evaluation, classification data: first category, R001 and G001, second category, R002 and G002, third category, R003 and G003, preference data: R001, 87, R002, 37.5, R003, 63, predicted preference data: R001, 95.7, R002, 41.3, R003, 69.3;

[0091] Obtaining a user preference index of the classified data based on the preference data and the predicted preference data, calculating a quality index of the shareable data, and obtaining a sharing index based on the user preference index and the quality index;

[0092] In the actual evaluation, user preference index: first category, 91.4, second category, 39.4, third category, 66.2, quality index: G001, 6.79, G002, 5.21, G003, 1.91, sharing index: G001, 66, G002, 29.1, G003, 46.9;

[0093] A Bayesian network algorithm is used to establish a resource sharing model based on the sharing index, and the data to be shared is input into the resource sharing model to obtain a sharing result.

[0094] In this embodiment, the sharable data and the historical reading data include:

[0095] The shareable data includes the shareable digital resources provided by the cooperation platform and their user ratings, number of citations and publication time. The historical reading data includes the existing digital resources on the platform and their reading time, number of downloads and publication time. The reading time is in minutes and the publication time is in days.

[0096] In this embodiment, the method for classifying the historical reading data of the local resource and the sharable data to obtain classified data includes:

[0097] Use TF-IDF to convert the historical reading data of local resources and the digital resources in the shareable data into feature vectors, and cluster the feature vectors:

[0098] Calculate the similarity between two feature vectors through semantic similarity:

[0099]

[0100] Among them, S(d i ,d j ) is the measure d i and d j The similarity function of the similarity between i and d j Represent two different eigenvectors, w i,f is d i The weight of feature item f is the TF-IDF value, where f is the keyword in the digital resource, Φ(d i ,d j ) is d i and d j The common feature set, α f is the coefficient used to adjust the importance of feature f based on mutual information, ψ is the Sigmoid function, TF i,f is d i The term frequency of the characteristic term f in , φ is an exponential function, φ(x) = e kx, k is a constant used to control the growth rate of the exponential function, determined by cross-validation,

[0101] Initialization of cluster centers:

[0102]

[0103] Among them, μ k is the cluster center of cluster k, d m Yes D k The eigenvectors in is d m The initial weight, D k is the set of feature vectors initially assigned to cluster k, ν is the regularization constant used to avoid the denominator being zero, ν∈[10- 16 ,10- 6 ],

[0104] Cluster center update:

[0105] C k (t+1)={d|S(d, μ k )≥τ}

[0106]

[0107] Among them, C k (t+1) is the set of feature vectors belonging to cluster k at the t+1th iteration, τ is the similarity threshold, and is the average similarity between feature vector pairs. is the gradient of the cluster center, p(C k (t)|μ k ) is the clustering conditional probability, η(t) is the learning rate,

[0108] Neighborhood function and eigenvector weight updates:

[0109]

[0110] in is the eigenvector d at the tth iteration m Neighborhood function, σ(t) is the neighborhood size, which decreases with time, δ(t) is the impact factor of the tth iteration, δ(t) = δ0·e -λt , δ0 is the initial impact factor, λ is the decay rate, is the eigenvector d of the t+1th iteration m The weight of

[0111] Clustering quality assessment and optimization:

[0112]

[0113] Among them, Q is the comprehensive clustering quality index, k and k′ represent the indexes of different clusters, and the comprehensive clustering quality index is maximized by adjusting the weight of the feature vector, the cluster center and the neighborhood function.

[0114] In this embodiment, the method of obtaining the preference data based on the historical reading data and obtaining the predicted preference data by time series analysis includes:

[0115] Remove the data with reading time less than the preset effective threshold in the historical reading data. The preset effective threshold is 0.5 minutes. Calculate the preference data of local resources by downloading times, publishing time and historical reading data:

[0116]

[0117] Among them, Re represents the preference data of local resources, w1, w2, w3, w4 are weight coefficients used to adjust the contribution of different factors to the preference data, D is the number of downloads, R represents the historical reading volume of local resources, which is obtained based on historical reading data, and P is the published time. Indicates the average reading time of users for this local resource, e -λP is an exponential decay term, λ is the decay rate, which is used to simulate the gradual decrease in the popularity of local resources over time.

[0118] Use preference data and historical reading data to build a time series analysis model:

[0119] Optimization of multi-step forecast errors of VAR models:

[0120]

[0121] Among them, MSE multi-step is the mean square error of the multi-step forecast, T is the length of the time series, Represents the model coefficients of the time series analysis model, y j,t is the jth time series data point, represents the forecast data, m is the number of variables in the time series, p is the lag order,

[0122] Feature selection and model complexity control of SVR model:

[0123]

[0124] Among them, y i,t is the sample data at time point t and sample index i, n is the number of samples, ∈ is the noise term, w k is the weight corresponding to the kth feature, d is the dimension of the feature, γ feature is the regularization coefficient for feature selection,

[0125] Optimization of timing analysis model:

[0126]

[0127] in, is the prediction result of the time series analysis model, κ is the weight for balancing the predictions of the SVR model and the VAR model, and γ VAR is the regularization coefficient of the VAR model, R VAR is the regularization function of the VAR model, and the prediction results are used as the prediction preference data.

[0128] In this embodiment, the method for obtaining the user preference index of the classification data based on the preference data and the predicted preference data includes:

[0129] User Preference Index:

[0130]

[0131] where Y q is the user preference index of the category of the qth classification data, is the preference data of the jth digital resource in the qth category, J is the number of digital resources in category q, is the predicted preference data of the j-th digital resource in the q-th category, s1 is the weight of the preference data, s2 is the weight of the predicted preference data, and s1+s2=1.

[0132] In this embodiment, the method for calculating the quality index of the sharable data includes:

[0133] The quality index of shareable data is calculated by user ratings, number of citations, and publication time:

[0134]

[0135] Among them, Z represents the quality index of the sharable data, a1, a2, a3 are weight factors used to adjust the contribution of different influencing factors to the quality index, Y is the number of downloads, t is the current time, τ i H i timestamp, Δt is the time window parameter, α is the parameter for controlling time sensitivity, which is a positive real number, H i represents the i-th user rating of the sharable data, I is the number of user ratings of the sharable data, and e -λP is the exponential decay term, and λ is the decay rate.

[0136] In this embodiment, the method for obtaining the sharing index based on the user preference index and the quality index includes:

[0137] Perform logarithmic transformation on user ratings, citation counts, download counts, published time, and historical reading data. The base of the logarithmic transformation is 10. Create a data set. Take the user ratings, citation counts, download counts, and published time in the data set as the user preference index set. Take the user ratings, citation counts, and published time in the data set as the quality index set. The weights of the user preference index and the quality index are:

[0138]

[0139] where g v is the contribution value of subset v, S is the data set, x is the data point in S, μ is the mean of S, v is the index of the subset, V is the set of subsets, represents the set of user preference index set and quality index set, ωv represents the weight of subset v,

[0140] Sharing index of shareable data:

[0141] G=ω1Y+ω2Z

[0142] Where G is the sharing index, ω1 is the weight of the user preference index, ω2 is the weight of the quality index, and Y is the user preference index of the classification category where the shareable data is located.

[0143] In this embodiment, the method of establishing a resource sharing model based on the sharing index using a Bayesian network algorithm includes:

[0144] Share the sharable data whose sharing index is greater than the preset sharing threshold. The preset sharing threshold is 0.5. Establish the sharing label of the sharable data based on the sharing index and the sharing threshold. 1 means sharing and 0 means not sharing. Normalize and preprocess the user preference index, user rating, number of citations and published time as input features. Take the sharing label as the child node and the input feature as the parent node to establish a directed acyclic graph. Use the uniform prior distribution as the prior condition to train the resource sharing model:

[0145] Predicted probability distribution:

[0146]

[0147] Among them, P represents the predicted probability distribution, Y is the response variable, is the shared label, X is the input feature, E is the training set, Q is the variational distribution of the model parameter θ, λ is the adjustment parameter of the error term, θ prior is the a priori parameter estimate, θ prior A normal distribution centered on

[0148] Goodness of fit:

[0149] G(E, μ, θ) = Σlog[q(v i |θ,∈)·R(v i )]-μ∑log[H(v i )I(v i )]-∑logP(E|v i ,∈)

[0150] Among them, G represents the goodness of fit function, v i is the i-th weight in the neural network, q is the variational distribution, H is the uniform prior distribution, I is the likelihood function, μ is the weight of H and I in the goodness of fit function, ∈ is the uncertainty parameter, R(v i ) is v i The regularization function of

[0151] Probability density function:

[0152]

[0153] Where f is the probability density function, Y pred is the model prediction value, σ is the noise term, α(Y, Y pred ,σ) is the predicted value Y pred The adjustment function of the noise term is the Sigmoid function,

[0154] Goodness of fit for prediction uncertainty:

[0155] U(E,μ,θ)=∑log[q(v i |θ,∈)·R(v i )]-μΣlog[H(v i )I(v i )]∑log[f(Y|Y pred ,σ,α)]

[0156] Wherein, U represents the goodness of fit considering the prediction uncertainty, and the sharable data will be shared when the prediction probability distribution of Y=1 is greater than or equal to 50%.

[0157] A digital library resource sharing system based on Bayesian network, characterized by comprising:

[0158] The collection module collects sharable data and historical reading data of users on local resources;

[0159] A calculation module, which obtains preference data based on the historical reading data, obtains predicted preference data by time series analysis, obtains a user preference index of the classified data based on the preference data and the predicted preference data, calculates a quality index of the shareable data, obtains a sharing index based on the user preference index and the quality index, and establishes a resource sharing model based on the sharing index by using a Bayesian network algorithm;

[0160] A classification module, classifying the historical reading data of the local resource and the sharable data to obtain classified data;

[0161] The output module inputs the data to be shared into the resource sharing model to obtain the sharing result.

[0162] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A digital library resource sharing method based on Bayesian network, characterized in that: The following steps are involved: Collect sharable data and historical reading data of users on local resources, Shareable data includes shareable digital resources provided by the cooperation platform and their user ratings, citation times and published time. Historical reading data includes existing digital resources on the platform and their reading time, download times and published time. classifying the historical reading data of the local resource and the sharable data to obtain classified data, Use TF-IDF to convert the historical reading data of local resources and the digital resources in the shareable data into feature vectors, and cluster the feature vectors: Calculate the similarity between two feature vectors through semantic similarity: Among them, S(d i ,d j ) is the measure d i and d j The similarity function of the similarity between i and d j Represent two different eigenvectors, w i,f is d i The weight of feature item f is the TF-IDF value, where f is the keyword in the digital resource, Φ(d i ,d j ) is d i and d j The common feature set, α f is the coefficient used to adjust the importance of feature f based on mutual information, ψ is the Sigmoid function, TF i,f is d i The term frequency of the characteristic term f, Φ is an exponential function, Φ(x) = e kx , k is a constant used to control the growth rate of the exponential function, determined by cross-validation, Initialization of cluster centers: Among them, μ k is the cluster center of cluster k, d m Yes D k The eigenvectors in is d m The initial weight, D k is the set of feature vectors initially assigned to cluster k, ν is the regularization constant used to avoid the denominator being zero, ν∈[10 -16 ,10 -6 ], Cluster center update: C k (t+1)={d|S(d,μ k )≥τ} Among them, C k (t+1) is the set of feature vectors belonging to cluster k at the t+1th iteration, τ is the similarity threshold, and is the average similarity between feature vector pairs. is the gradient of the cluster center, p(C k (t)|μ k ) is the clustering conditional probability, η(t) is the learning rate, Neighborhood function and eigenvector weight updates: in is the eigenvector d at the tth iteration m Neighborhood function, σ(t) is the neighborhood size, which decreases with time, δ(t) is the impact factor of the tth iteration, δ(t) = δ0·e -λt , δ0 is the initial impact factor, λ is the decay rate, is the eigenvector d of the t+1th iteration m The weight of Clustering quality assessment and optimization: Among them, Q is the comprehensive clustering quality index, k and k ′ Represents the index of different clusters, and maximizes the comprehensive clustering quality index by adjusting the weight of the feature vector, cluster center and neighborhood function. Based on the historical reading data, preference data is obtained, and time series analysis is used to obtain predicted preference data. Remove the data with reading time less than the preset effective threshold from the historical reading data, and calculate the preference data of local resources based on the number of downloads, published time and historical reading data: Among them, Re represents the preference data of local resources, w1, w2, w3, w4 are weight coefficients used to adjust the contribution of different factors to the preference data, D is the number of downloads, R represents the historical reading volume of local resources, which is obtained based on historical reading data, and P is the published time. Indicates the average reading time of users for this local resource, e -λP is an exponential decay term, λ is the decay rate, which is used to simulate the gradual decrease in the popularity of local resources over time. Use preference data and historical reading data to build a time series analysis model: Optimization of multi-step forecast errors of VAR models: Among them, MSE multi-step is the mean square error of the multi-step forecast, T is the length of the time series, Represents the model coefficients of the time series analysis model, y j,t is the jth time series data point, represents the forecast data, m is the number of variables in the time series, p is the lag order, Feature selection and model complexity control of SVR model: Among them, y i,t is the sample data at time point t and sample index i, n is the number of samples, ∈ is the noise term, w k is the weight corresponding to the kth feature, d is the dimension of the feature, γ feature is the regularization coefficient for feature selection, Optimization of timing analysis model: in, is the prediction result of the time series analysis model, κ is the weight for balancing the predictions of the SVR model and the VAR model, and γ VAR is the regularization coefficient of the VAR model, R VAR is the regularization function of the VAR model, and the prediction results are used as the prediction preference data; obtaining a user preference index of the classification data based on the preference data and the predicted preference data, User Preference Index: where Y q is the user preference index of the category of the qth classification data, is the preference data of the jth digital resource in the qth category, J is the number of digital resources in category q, is the predicted preference data of the jth digital resource in the qth category, s1 is the weight of the preference data, s2 is the weight of the predicted preference data, s1+s2=1, calculating a quality index of the sharable data, The quality index of shareable data is calculated by user ratings, number of citations, and publication time: Among them, Z represents the quality index of the sharable data, a1, a2, a3 are weight factors used to adjust the contribution of different influencing factors to the quality index, Y is the number of downloads, t is the current time, τ i H i timestamp, Δt is the time window parameter, α is the parameter for controlling time sensitivity, which is a positive real number, H i represents the i-th user rating of the sharable data, I is the number of user ratings of the sharable data, and e -λP is the exponential decay term, λ is the decay rate, Obtaining a sharing index based on the user preference index and the quality index, Perform logarithmic transformation on user ratings, citation counts, download counts, published time, and historical reading data to establish a data set. The user ratings, citation counts, download counts, and published time in the data set are used as the user preference index set. The user ratings, citation counts, and published time in the data set are used as the quality index set. The weights of the user preference index and the quality index are: where g v is the contribution value of subset v, S is the data set, x is the data point in S, μ is the mean of S, v is the index of the subset, V is the set of subsets, represents the set of user preference index set and quality index set, ω v represents the weight of subset v, Sharing index of shareable data: G=ω1Y+ω2Z Where G is the sharing index, ω1 is the weight of the user preference index, ω2 is the weight of the quality index, and Y is the user preference index of the classification category where the shareable data is located; A resource sharing model is established based on the sharing index using a Bayesian network algorithm. Share the sharable data whose sharing index is greater than the preset sharing threshold. Establish the sharing label of the sharable data based on the sharing index and the sharing threshold. 1 means sharing and 0 means not sharing. Normalize and preprocess the user preference index, user rating, number of citations and published time as input features. Use the sharing label as the child node and the input feature as the parent node to establish a directed acyclic graph. Use the uniform prior distribution as the prior condition to train the resource sharing model: Predicted probability distribution: Where P represents the predicted probability distribution, Y is the response variable, is the shared label, X is the input feature, E is the training set, Q is the variational distribution of the model parameter θ, λ is the adjustment parameter of the error term, θprior is the prior parameter estimate, is a normal distribution centered at θprior, Goodness of fit: G(E,μ,θ)=∑log[q(v i |θ,∈)·R(v i )]-μ∑log[H(v i )I(v i )]-ΣlogP(E|v i ,∈) Among them, G represents the goodness of fit function, v i is the i-th weight in the neural network, q is the variational distribution, H is the uniform prior distribution, I is the likelihood function, μ is the weight of H and I in the goodness of fit function, ∈ is the uncertainty parameter, R(v i ) is v i The regularization function of Probability density function: Where f is the probability density function, Y pred is the model prediction value, σ is the noise term, α(Y, Y pred ,σ) is the predicted value Y pred The adjustment function of the noise term is the Sigmoid function, Goodness of fit for prediction uncertainty: U(E,μ,θ)=∑log[q(v i |θ,∈)·R(v i )]-µ∑log[H(v i )I(v i )]Σlog[f(Y|Y pred ,σ,α)] Among them, U represents the goodness of fit considering the prediction uncertainty. When the prediction probability distribution of Y=1 is greater than or equal to 50%, the sharable data will be shared. The data to be shared is input into the resource sharing model to obtain a sharing result.

2. A digital library resource sharing system based on a Bayesian network, used to execute the digital library resource sharing method based on a Bayesian network as claimed in claim 1, characterized in that: include: The collection module collects sharable data and historical reading data of users on local resources; A calculation module, which obtains preference data based on the historical reading data, obtains predicted preference data by time series analysis, obtains a user preference index of the classified data based on the preference data and the predicted preference data, calculates a quality index of the shareable data, obtains a sharing index based on the user preference index and the quality index, and establishes a resource sharing model based on the sharing index by using a Bayesian network algorithm; A classification module, classifying the historical reading data of the local resource and the sharable data to obtain classified data; The output module inputs the data to be shared into the resource sharing model to obtain the sharing result.

Images