Internet-based physical book intelligent management and control system

Abstract

The invention belongs to the technical field of information processing, and discloses a physical book intelligent management and control system based on the Internet. Comprising a book information acquisition module used for acquiring book information data; the reading preference analysis module is used for acquiring reading preference data of the reader according to the book information data; the inventory management module is used for adjusting the book inventory according to the reading preference data of the reader; the method comprises the steps of judging whether book inventory early warning information is generated or not by monitoring the book inventory; the book stock is adjusted according to the reading preference data of the readers, reading interests and hobbies of the readers can be known, the books are reasonably distributed according to borrowing or purchasing habits of the readers, whether the book stock is sufficient or not is judged by monitoring the book stock, the book stock can be supplemented in time, and it is ensured that the readers can obtain the needed books in time.

Description

Technical Field

[0001] The present invention relates to the technical field of data management and analysis. More specifically, the present invention relates to an intelligent control system for physical books based on the Internet. Background Art

[0002] In recent years, with the popularization of e-reading devices and the explosive growth of network information, the sales of physical books have faced unprecedented challenges. The traditional paper book management and warehousing models have been difficult to meet the diverse and personalized needs of current consumers. At the same time, due to the lack of effective digital management means, publishers and bookstores also face many difficulties in book procurement, inventory, marketing and other aspects.

[0003] Therefore, it is urgent to establish an intelligent control system for physical books based on the Internet, make full use of emerging technologies such as big data, Internet of Things, cloud computing, etc., and realize the intelligent management of the entire life cycle of physical books.

[0004] The patent with the application publication number of CN107169545 discloses an intelligent bookshelf control system and method. The system includes: a scanning information acquisition unit for acquiring the basic information of each book within its reading range by using a reader / writer; an analysis and processing unit for analyzing and processing the basic information of each book acquired by the scanning information acquisition unit to obtain the label content of each book; a matching unit for matching the obtained label content of the book with the book labels that have been classified and stored before, finding the book with the highest matching degree, and obtaining the storage location information of the book with the highest matching degree; a book shelving control unit for controlling the robotic arm to store the book at the corresponding position of the intelligent bookshelf according to the storage location information of the book with the highest matching degree obtained by the matching unit. Through the present invention, the automation of book management can be realized.

[0005] It does not consider adjusting the book inventory according to the reader's reading preference data, which is not conducive to understanding the reader's reading interests and hobbies, and cannot reasonably allocate books according to the reader's borrowing or purchasing habits. It does not consider judging whether the book inventory is sufficient by monitoring the book inventory, which is not conducive to the timely replenishment of the book inventory and cannot ensure that readers can obtain the required books in time;

[0006] In view of this, the present invention proposes an intelligent control system for physical books based on the Internet to solve the above problems. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art and to achieve the above object, the present invention provides the following technical solution: An intelligent control system for physical books based on the Internet, comprising:

[0008] A behavior data acquisition module for acquiring reader behavior data;

[0009] A reading preference analysis module for obtaining readers' reading preference data according to readers' behavior data;

[0010] An inventory management module for adjusting the inventory levels of different books according to readers' reading preference data; judging whether to generate a warning message for book inventory levels by monitoring the inventory levels of different books;

[0011] The various modules are connected by wired and / or wireless means to achieve data transmission between the modules.

[0012] Furthermore, the readers' behavior data includes readers' borrowing activity, readers' evaluation data, and collection data.

[0013] Furthermore, the method for obtaining the readers' behavior data includes:

[0014] The readers' borrowing activity includes the borrowing duration of books by readers within a week, the number of books of the same type borrowed, and the persistence of borrowing books; when readers borrow books, they scan the RFID tags on the books through the self-service book borrowing and returning machines set in bookstores or libraries, and the self-service book borrowing and returning machines record the borrowing duration of books by readers within a week and the number of books of the same type borrowed;

[0015] The readers' borrowing activity is: where ql is the borrowing duration of books by readers within a week; yk is the number of books of the same type borrowed within a week; zm is the persistence of borrowing books within a week;

[0016] Dividing the borrowing duration of books by readers within a week by the number of books of the same type borrowed within a week to obtain the average borrowing duration of each book, and reflecting the persistence of borrowing books through the average borrowing duration; combining the borrowing duration of books by readers within a week ql and the number of books of the same type borrowed within a week yk to calculate the persistence of borrowing books within a week

[0017] The readers' evaluation data includes the scoring scores, comment durations, and comment numbers of readers on books within a week;

[0018] A readers' feedback box is set at the service desk of the bookstore or library. Readers leave their post-view comments and scoring scores on books in the feedback box, and the staff record the scoring scores and comment numbers of readers on books within a week; the comment durations of readers within a week are recorded through the cameras in the bookstore or library;

[0019] The collected data includes the time when readers collect books within a week, the return rate of the collected books, and the ways of collecting books; readers select their preferred books on the browsing shelves in the bookstore or library and notify the front desk staff to collect the preferred books, and the staff records the time when readers collect books within a week; by recording the number bx of books that readers cancel the collection within a week and the total number jv of books collected within a week by the staff, the return rate of the books collected within a week is calculated. The ways of collecting books include collecting the books to be collected in a classified form.

[0020] Further, the method of collecting the books to be collected in a classified form includes:

[0021] Classify the ways of collecting books through the K-means clustering algorithm to obtain q different types of clusters, and each cluster represents a way of collecting books with different types of characteristics. The ways of collecting books with different types of characteristics include collecting books according to the theme or field, collecting books according to the author or editor, collecting books according to the publication time, collecting books according to the reading status, and collecting books according to the decoration style.

[0022] Further, the method of obtaining the reading preference data of readers according to the reader behavior data includes:

[0023] Perform standard deviation normalization processing on the obtained reader behavior data, convert it into a standard normal distribution with a mean of 0 and a standard deviation of 1, eliminate the dimension influence between the data, and finally organize and obtain the normalized feature data set.

[0024] Construct a reading preference prediction model. The input of the reading preference prediction model is historical feature data; the output label is the reading preference data of readers; combine machine learning and deep learning technologies to continuously update and optimize the reading preference prediction model, and update the reading preference prediction model by continuously collecting and analyzing reader behavior data.

[0025] Further, the construction method of the reading preference prediction model includes:

[0026] Collect a data training set, including input data and corresponding labels; divide the data set into a training set and a test set; use the historical feature data set of the data training set as the input and the corresponding reading preference data of readers as the output label to train the reading preference prediction model; the reading preference prediction model is a decision tree model.

[0027] During the training process of the reading preference prediction model, node splitting is performed according to different features of the historical feature dataset. A feature is selected from the historical feature dataset as the splitting feature of the current node, and the initial entropy of the current node is calculated using the formula in information theory to calculate entropy; when calculating each feature value as the splitting point, the entropy after splitting is calculated, and the information gain is obtained by calculating the difference between the initial entropy and the entropy after splitting.

[0028] The feature value with the maximum information gain is selected from the calculated information gains as the best splitting scheme for the current node; the maximum information gain corresponds to being able to maximize the purity of the readers' reading preferences; based on the best splitting scheme, child nodes are created, and the corresponding historical feature datasets are assigned to each child node.

[0029] For each child node, the above steps are recursively executed until the number of the historical feature dataset is lower than the preset number threshold of the historical feature dataset, and then the node splitting stops.

[0030] The Bayesian optimization method is used to tune the parameters in the reading preference prediction model. The specific steps are as follows:

[0031] Determine the minimum information gain parameter for feature selection in the decision tree model that needs to be tuned.

[0032] Prepare a set of initial data training sets, and evaluate the performance metrics of the model under the given parameter configuration by calculating the accuracy metric; the accuracy calculation formula is: where oj is the number of reader reading preference data predicted correctly; pw is the total number of reader reading preference data.

[0033] Using the initial data training set, establish a Gaussian process initial probability model between the parameter configuration and the performance metrics; start the iterative optimization process, and each iteration includes the following steps:

[0034] a. According to the current probability model, use Gaussian process sampling to select the next parameter configuration.

[0035] b. Using the selected parameter configuration, calculate its performance metrics by calculating the accuracy.

[0036] c. Add the new parameter configuration and performance metrics to the data training set.

[0037] d. Update the probability model, add the new data training set to the model training process to update the mapping relationship between the parameters and the performance.

[0038] e. Repeat steps a to d until the predetermined number of iterations is reached.

[0039] Output the optimal parameter configuration: After the iterative optimization is completed, according to the results of the evaluation function, select the parameter configuration with the best performance metrics as the final model configuration to obtain a trained reading preference prediction model. Use the trained reading preference prediction model to predict the current reader feature dataset to obtain reader reading preference data.

[0040] Further, the method for adjusting the inventory levels of different books according to the reader reading preference data includes:

[0041] The reader reading preference data includes the types of books preferred by the reader within a week, the number of searches for preferred books, and the search duration of preference data. Record the types of books preferred by the reader within a week, the number of searches for preferred books, and the search duration of preference data through the self-service book borrowing and returning machines set up in bookstores or libraries; perform standard deviation normalization on the obtained reader reading preference data and convert it into a standard normal distribution with a mean of 0 and a standard deviation of 1 to eliminate the dimensional influence between data, and finally organize and obtain the normalized reading preference dataset;

[0042] Check whether there are missing values in the reading preference dataset; for the columns containing missing values, regard them as the target columns to be filled; split the reading preference dataset into two parts: known data and missing data; the known data includes the types of books preferred by the reader within a week, the number of searches for preferred books, and the search duration of preference data, and the missing data only contains the types of books preferred by the reader within a week;

[0043] For the types of books preferred by the reader within a week in the missing data part, use the known data for Lagrange interpolation; construct a Lagrange interpolation polynomial; use the known types of books preferred by the reader within a week as the independent variable, and the corresponding number of searches for preferred books within a week and the search duration of preference data within a week as the dependent variables; use the constructed Lagrange interpolation polynomial to predict the types of books within a week in the missing data to obtain the corresponding number of searches for preferred books within a week and the search duration of preference data within a week, and fill the interpolated and predicted missing values back into the target columns of the reading preference dataset;

[0044] Conduct further data analysis and evaluation based on the filled reading preference dataset to verify the rationality and accuracy of the interpolation results;

[0045] Perform outlier processing on the reading preference dataset through the IQR statistical method: calculate the 25th percentile Q1 and 75th percentile Q3 of the dataset, and calculate IQR = Q3 - Q1; calculate the upper bound UB = Q3 + 1.5 * IQR and the lower bound LB = Q1 - 1.5 * IQR; check each data point in the reading feature dataset, if the data point is less than the lower bound or greater than the upper bound, mark it as an outlier, delete or correct the data points marked as outliers and finally organize them into a reading feature dataset;

[0046] Build a book inventory prediction model; the input data is a historical reading feature dataset, and the output label is the sales volume of different books. The intelligent control system for physical books records the obtained sales volume of different books and reminds the staff to adjust the inventory of different books based on the sales volume of different books.

[0047] Furthermore, the method for building the book inventory prediction model includes:

[0048] The book inventory prediction model is an LSTM model. Prepare the training dataset, including input data and corresponding labels; the input data is a historical reading feature dataset, and the output label is the sales volume of different books; divide the dataset into a training set and a test set to evaluate the performance of the model.

[0049] Convert the historical reader reading feature dataset into sequence data suitable for processing by the LSTM model; use time steps to represent each data point in the sequence, and construct input sequences and output sequences; use the deep learning framework TensorFlow to create an LSTM model; define the structure and parameter settings of the LSTM layer, including the number of LSTM cells and the size of the hidden layer, and adjust according to the scale and complexity of the historical reading feature dataset.

[0050] The method for converting the historical reader reading feature dataset into sequence data suitable for processing by the LSTM model includes:

[0051] The historical reader reading feature dataset is recorded once a week. Use the historical reader data of several weeks as the sequence length, and set the time step to 1 to keep consistent with the time scale of the historical reader reading feature data of one week. The model predicts the sales volume of different books in the next week based on the historical reader reading feature data of the past few weeks.

[0052] Use the test set to evaluate the performance of the trained LSTM model, and use the root mean square error as an indicator to measure the difference between the predicted sales volume of different books and the actual sales volume of different books; by using the Adam optimizer, adjust the weights and parameters according to the gradient information of the mean square error loss function to minimize the value of the loss function.

[0053] The method for adjusting the weights and parameters according to the gradient information of the mean square error loss function to minimize the value of the loss function includes:

[0054] Adjust the weights according to the gradient information of the mean square error loss function:

[0055] Initialize the weights of the LSTM model using the Xavier initialization method; input the training dataset into the LSTM model and calculate the predicted values of the model through forward propagation; calculate the mean squared error loss function between the predicted values and the actual values.

[0056] Calculate the gradient of the loss function with respect to the model weights and pass the gradient from the output layer to the input layer through backpropagation;

[0057] Update the model weights according to the gradient information and the rules of the Adam optimizer; use the following formula for weight update: PRX = PE - YH·η; where, PRX is the new weight of the LSTM model; PE is the old weight of the LSTM model; YH is the learning rate that controls the weight update speed; η is the gradient of the LSTM model weights;

[0058] Iterate the above steps repeatedly, gradually adjust the weights of the LSTM model to reduce the value of the loss function and improve the prediction accuracy of the model; use the trained LSTM model for book inventory prediction, input the current reading feature dataset, and make predictions through the model to obtain different book sales volumes.

[0059] Furthermore, the method for determining whether to generate a book inventory warning message by monitoring the book inventory includes:

[0060] Monitor the inventory levels of different books through the physical book intelligent management and control system, and compare the inventory levels of different books with the preset inventory thresholds for different books;

[0061] If the inventory level of a different book is less than or equal to the preset inventory threshold for different books, it is determined that a book inventory warning message is generated;

[0062] If the inventory level of a different book is greater than the preset inventory threshold for different books, it is determined that no book inventory warning message is generated;

[0063] An intelligent management and control method for physical books based on the Internet, which is implemented based on the described intelligent museum user management system, includes:

[0064] S1. Obtain reader behavior data;

[0065] S2. Obtain reader reading preference data according to the reader behavior data;

[0066] S3. Adjust the inventory levels of different books according to the reader reading preference data; determine whether to generate a book inventory warning message by monitoring the inventory levels of different books.

[0067] The technical effects and advantages of the intelligent management and control system for physical books based on the Internet according to the present invention:

[0068] Through the behavior data acquisition module, reader behavior data is obtained. By acquiring and processing the reader borrowing activity, reader evaluation data, and collection data in the reader behavior data, and constructing a reading preference prediction model to obtain the reader's reading preference, it helps to accurately understand the reader's reading preference and interest;

[0069] Through the inventory management module, according to the reader reading preference data, a book inventory prediction model is constructed to obtain the sales volume of different books. The physical book intelligent control system records the obtained sales volume of different books, and reminds the staff to adjust the inventory of different books based on the sales volume of different books, ensuring sufficient inventory supply and avoiding the situation of overstock or out-of-stock. Based on the book sales trend and prediction, the procurement plan, promotion strategy, and inventory management strategy of books can be adjusted to better meet the reader's needs and improve efficiency;

[0070] By using the physical book intelligent control system to monitor the inventory of different books, comparing the inventory of different books with the preset inventory threshold of different books to determine whether to generate a book inventory warning message, which helps to avoid the situation of insufficient inventory, ensure the smooth operation of the bookstore or library supply chain, improve the efficiency of inventory management, and reduce unnecessary cost expenditures; avoiding insufficient inventory or overstock can better meet the reader's needs and make inventory management more refined and economically reasonable. Brief Description of the Drawings

[0071] Figure 1 It is a schematic diagram of the module of a physical book intelligent control system based on the Internet according to the present invention;

[0072] Figure 2 It is a schematic diagram of the process of a physical book intelligent control method based on the Internet. Detailed Embodiments

[0073] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0074] Embodiment 1

[0075] Please refer to Figure 1 As shown, a physical book intelligent control system based on the Internet in this embodiment includes:

[0076] A behavior data acquisition module for acquiring reader behavior data;

[0077] A reading preference analysis module, which is used to obtain readers' reading preference data according to readers' behavior data;

[0078] An inventory management module, which is used to adjust the inventory of different books according to readers' reading preference data; by monitoring the inventory of different books, it determines whether to generate a warning message for the book inventory;

[0079] The various modules are connected by wired and / or wireless means to achieve data transmission between the modules.

[0080] The readers' behavior data includes readers' borrowing activity, readers' evaluation data and collection data.

[0081] The method for obtaining the readers' behavior data includes:

[0082] The readers' borrowing activity includes the borrowing duration of books by readers within a week, the number of books of the same type borrowed, and the persistence of borrowing books; when readers borrow books, they scan the RFID tags on the books through the self-service book borrowing and returning machines set in bookstores or libraries, and the self-service book borrowing and returning machines record the borrowing duration of books by readers within a week and the number of books of the same type borrowed;

[0083] The readers' borrowing activity is: Among them, ql is the borrowing duration of books by readers within a week; yk is the number of books of the same type borrowed within a week; zm is the persistence of borrowing books within a week;

[0084] Divide the borrowing duration of books by readers within a week by the number of books of the same type borrowed within a week to obtain the average borrowing duration of each book, and use the average borrowing duration to reflect the persistence of borrowing books; combine the borrowing duration of books by readers within a week ql and the number of books of the same type borrowed within a week yk to calculate the persistence of borrowing books within a week

[0085] The readers' evaluation data includes the scoring score, comment duration and comment quantity of readers on books within a week;

[0086] Set up a readers' feedback box at the service desk of the bookstore or library. Readers leave their post-view comments and scoring scores on the books in the feedback box, and the staff record the scoring scores and comment quantity of readers on the books within a week; record the comment duration of readers within a week through the cameras in the bookstore or library;

[0087] The collected data includes the time when readers collect books within a week, the return rate of the collected books, and the ways of collecting books; readers select their preferred books on the browsing shelves in the bookstore or library and notify the front desk staff to collect the preferred books, and the staff records the time when readers collect books within a week; by recording the number bx of books that readers cancel collection within a week and the total number jv of books collected within a week by the staff, the return rate of the collected books within a week is calculated. The ways of collecting books include collecting the books to be collected in a classified form.

[0088] The method of collecting the books to be collected in a classified form includes:

[0089] Classify the ways of collecting books through the K-means clustering algorithm to obtain q different types of clusters, and each cluster represents the ways of collecting books with different types of characteristics. The ways of collecting books with different types of characteristics include collecting books according to the theme or field, collecting books according to the author or editor, collecting books according to the publication time, collecting books according to the reading status, and collecting books according to the decoration style.

[0090] For example, collecting books according to the theme or field: a certain cluster may represent readers who have collected a large number of science fiction novels or fantasy books; they are more inclined to collect books related to science fiction and fantasy, and may be interested in themes such as space exploration and future technology;

[0091] Collecting books according to the author or editor: another cluster may represent readers who attach importance to the author or editor; they may collect multiple books published by the same author or the same editor because they have a strong preference and trust for these specific authors or editors;

[0092] Collecting books according to the publication time: a certain cluster may represent readers who pay attention to the publication time of books. They may prefer to collect the latest published books and follow the current hot topics and trends;

[0093] Collecting books according to the reading status: another cluster may represent readers who collect books according to the reading status; they may classify books into different statuses such as read, being read, and unread, and collect and manage them according to different statuses.

[0094] Collecting books according to the decoration style: the last cluster may represent readers who collect books according to the decoration style; these readers may pay more attention to the appearance and decoration of books and collect books with specific decoration styles, such as books with exquisite cover designs, special bindings, or deluxe editions.

[0095] The method of obtaining the reading preference data of readers according to the reader behavior data includes:

[0096] Perform standard deviation normalization on the obtained reader behavior data, convert it into a standard normal distribution with a mean of 0 and a standard deviation of 1, eliminate the dimensional influence between data, and finally organize and obtain the normalized feature dataset;

[0097] Construct a reading preference prediction model. The input of the reading preference prediction model is historical feature data; the output label is reader reading preference data; combine machine learning and deep learning technologies to continuously update and optimize the reading preference prediction model, and update the reading preference prediction model by continuously collecting and analyzing reader behavior data.

[0098] The construction method of the reading preference prediction model includes:

[0099] Collect a data training set, including input data and corresponding labels; divide the dataset into a training set and a test set; use the historical feature dataset of the data training set as the input and the corresponding reader reading preference data as the output label to train the reading preference prediction model; the reading preference prediction model is a decision tree model;

[0100] During the training process of the reading preference prediction model, perform node splitting according to different features of the historical feature dataset, select a feature from the historical feature dataset as the splitting feature of the current node, calculate the initial entropy of the current node, and calculate the entropy using the formula in information theory; calculate the entropy after division when each feature value is used as the splitting point, and obtain the information gain by calculating the difference between the initial entropy and the entropy after division; the information gain represents the degree to which the uncertainty of the node can be reduced through the division of specific feature values;

[0101] Select the feature value with the maximum information gain from the calculated information gains as the best division scheme for the current node; the maximum information gain corresponds to the purity that can maximize the improvement of the reader's reading preference; based on the best division scheme, create child nodes and allocate the corresponding historical feature datasets to each child node;

[0102] For each child node, recursively execute the above steps until the number of the historical feature dataset is lower than the preset number threshold of the historical feature dataset to stop node splitting;

[0103] Use the Bayesian optimization method to tune the parameters in the reading preference prediction model. The specific steps are:

[0104] Determine the minimum information gain parameter for feature selection in the decision tree model that needs to be tuned;

[0105] Prepare a set of initial data training sets, and evaluate the performance metrics of the model under the given parameter configuration by calculating the accuracy metric; the accuracy calculation formula is: Among them, oj is the number of reader reading preference data predicted correctly; pw is the total number of reader reading preference data;

[0106] Use the initial data training set to establish an initial Gaussian process probability model between parameter configurations and performance metrics; start the iterative optimization process, and each iteration includes the following steps:

[0107] a. According to the current probability model, use Gaussian process sampling to select the next parameter configuration;

[0108] b. Use the selected parameter configuration to calculate its performance metrics by calculating the accuracy rate;

[0109] c. Add the new parameter configuration and performance metrics to the data training set;

[0110] d. Update the probability model, add the new data training set to the model training process to update the mapping relationship between parameters and performance;

[0111] e. Repeat steps a to d until the predetermined number of iterations is reached;

[0112] Output the best parameter configuration: After the iterative optimization is completed, according to the results of the evaluation function, select the parameter configuration with the best performance metrics as the final model configuration to obtain a trained reading preference prediction model, and use the trained reading preference prediction model to predict the current reader feature data set to obtain reader reading preference data.

[0113] The method for adjusting the inventory levels of different books according to the reader reading preference data includes:

[0114] The reader reading preference data includes the types of books preferred by readers within a week, the number of searches for preferred books, and the search duration of preference data. Record the types of books preferred by readers within a week, the number of searches for preferred books, and the search duration of preference data through the self-service book borrowing and returning machines set up in bookstores or libraries; perform standard deviation normalization on the obtained reader reading preference data and convert it into a standard normal distribution with a mean of 0 and a standard deviation of 1 to eliminate the dimensionality impact between data, and finally organize and obtain the normalized reading preference data set;

[0115] Check whether there are missing values in the reading preference data set; for the columns containing missing values, regard them as the target columns to be filled; split the reading preference data set into two parts: known data and missing data; the known data includes the types of books preferred by readers within a week, the number of searches for preferred books, and the search duration of preference data, and the missing data only contains the types of books preferred by readers within a week;

[0116] For the book types preferred by readers in the missing data part within one week, use the known data for Lagrange interpolation; construct the Lagrange interpolation polynomial; use the known book types preferred by readers within one week as the independent variable, and the corresponding search times for preferred books and search durations for preferred data within one week as the dependent variables; use the constructed Lagrange interpolation polynomial to predict the book types within one week of the missing data, obtain the corresponding search times for preferred books and search durations for preferred data within one week, and fill the missing values obtained by interpolation prediction back into the target column of the reading preference dataset;

[0117] Conduct further data analysis and evaluation based on the filled reading preference dataset to verify the rationality and accuracy of the interpolation results;

[0118] Perform outlier processing on the reading preference dataset through the IQR statistical method: calculate the 25th percentile Q1 and 75th percentile Q3 of the dataset, and calculate IQR = Q3 - Q1; calculate the upper bound UB = Q3 + 1.5 * IQR and the lower bound LB = Q1 - 1.5 * IQR; check each data point in the reading feature dataset. If the data point is less than the lower bound or greater than the upper bound, mark it as an outlier, delete or correct the data points marked as outliers, and finally organize them into the reading feature dataset;

[0119] For example, there is the following reading feature dataset. Taking the search duration for preferred data as an example, [10, 15, 12, 18, 20, 25, 8, 30, 16, 14, 22, 27, 35, 40, 5]; calculate the 25th percentile and 75th percentile: the 25th percentile Q1 = 12, the 75th percentile Q3 = 27, calculate the upper bound and lower bound: IQR = 75th percentile Q3 - 25th percentile Q1 = 27 - 12 = 15, the upper bound = 27 + 1.5 * 15 = 49.5, the lower bound = 12 - 1.5 * 15 = -7.5, check each data point to determine whether it is an outlier:

[0120] 10 is not less than the lower bound and is not an outlier, 15 is not less than the lower bound and is not an outlier, 12 is not less than the lower bound and is not an outlier, 18 is not less than the lower bound and is not an outlier, 20 is not less than the lower bound and is not an outlier, 25 is not less than the lower bound and is not an outlier, 8 is less than the lower bound and is an outlier, 30 is not less than the lower bound and is not an outlier, 16 is not less than the lower bound and is not an outlier, 14 is not less than the lower bound and is not an outlier, 22 is not less than the lower bound and is not an outlier, 27 is not less than the lower bound and is not an outlier, 35 is not less than the lower bound and is not an outlier, 40 is not less than the lower bound and is not an outlier, 5 is less than the lower bound and is an outlier;

[0121] Delete or correct the data points marked as outliers: Delete outliers 8 and 5, and finally organize them into a reading feature dataset: [10, 15, 12, 18, 20, 25, 30, 16, 14, 22, 27, 35, 40].

[0122] Build a book inventory prediction model; the input data is the historical reading feature dataset, the output label is the sales volume of different books, the entity book intelligent control system records the obtained sales volume of different books, and reminds the staff to adjust the inventory of different books through the sales volume of different books.

[0123] The construction method of the book inventory prediction model includes:

[0124] The book inventory prediction model is an LSTM model. Prepare the training dataset, including input data and corresponding labels; the input data is the historical reading feature dataset, and the output label is the sales volume of different books; divide the dataset into a training set and a test set to evaluate the performance of the model;

[0125] Convert the reader reading feature dataset into sequence data suitable for processing by the LSTM model; use time steps to represent each data point in the sequence, and construct input sequences and output sequences; use the deep learning framework TensorFlow to create an LSTM model; define the structure and parameter settings of the LSTM layer, including the number of LSTM cells and the size of the hidden layer, and adjust according to the scale and complexity of the historical reading feature dataset.

[0126] The method of converting the historical reader reading feature dataset into sequence data suitable for processing by the LSTM model includes: The historical reader reading feature dataset is recorded once a week. Use the historical reader data of several weeks as the sequence length, and set the time step to 1 to keep the same time scale as the historical reader reading feature data of one week. The model predicts the sales volume of different books in the next week based on the historical reader reading feature data of the past few weeks.

[0127] Use the test set to evaluate the performance of the trained LSTM model, and use the root mean square error as an indicator to measure the difference between the predicted sales volume of different books and the actual sales volume of different books; by using the Adam optimizer, adjust the weights and parameters according to the gradient information of the mean square error loss function to minimize the value of the loss function.

[0128] The method of adjusting the weights and parameters according to the gradient information of the mean square error loss function to minimize the value of the loss function includes:

[0129] Adjust the weights according to the gradient information of the mean square error loss function:

[0130] Initialize the weights of the LSTM model using the Xavier initialization method; input the training dataset into the LSTM model and calculate the predicted values of the model through forward propagation; calculate the mean square error loss function between the predicted values and the actual values.

[0131] Calculate the gradient of the loss function with respect to the model weights and pass the gradient from the output layer to the input layer through backpropagation;

[0132] Update the model weights according to the gradient information and the rules of the Adam optimizer; use the following formula for weight update: PRX = PE - YH·η; where, PRX is the new weight of the LSTM model; PE is the old weight of the LSTM model; YH is the learning rate that controls the weight update speed; η is the gradient of the LSTM model weights;

[0133] Iterate the above steps repeatedly to gradually adjust the weights of the LSTM model, reduce the value of the loss function, and improve the prediction accuracy of the model; use the trained LSTM model for book inventory prediction, input the current reading feature dataset, and make predictions through the model to obtain the sales volumes of different books.

[0134] The method for monitoring the book inventory and determining whether to generate a book inventory warning message includes:

[0135] Monitor the inventory levels of different books through the physical book intelligent management and control system, and compare the inventory levels of different books with the preset inventory thresholds for different books;

[0136] If the inventory level of a different book is less than or equal to the preset inventory threshold for that different book, then it is determined that a book inventory warning message is generated;

[0137] If the inventory level of a different book is greater than the preset inventory threshold for that different book, then it is determined that no book inventory warning message is generated;

[0138] The preset inventory thresholds for different books are set by the staff. Record the inventory levels of different books through the physical book intelligent management and control system, collect multiple groups of inventory levels of different books, and take the average of multiple inventory levels of different books as the preset inventory thresholds for different books respectively.

[0139] In this embodiment, through the behavior data acquisition module, reader behavior data is acquired. By acquiring and processing the reader borrowing activity, reader evaluation data, and collection data in the reader behavior data, and constructing a reading preference prediction model, the reader's reading preference is obtained, which helps to accurately understand the reader's reading preference and interest;

[0140] Through the inventory management module, a book inventory prediction model is constructed based on readers' reading preference data to obtain the sales volumes of different books. The intelligent control system for physical books records the obtained sales volumes of different books and reminds the staff to adjust the inventory levels of different books based on the sales volumes of different books, ensuring sufficient inventory supply and avoiding overstocking or out-of-stock situations. Based on book sales trends and predictions, the procurement plan, promotion strategy, and inventory management strategy of books can be adjusted to better meet readers' needs and improve efficiency.

[0141] The intelligent control system for physical books monitors the inventory levels of different books, compares the inventory levels of different books with the preset inventory thresholds for different books, and determines whether to generate a warning message for book inventory levels, which helps avoid out-of-stock situations and ensures the smooth operation of the supply chain of bookstores or libraries. It can improve the efficiency of inventory management and reduce unnecessary cost expenditures; avoiding understocking or overstocking can better meet readers' needs and make inventory management more refined, economical, and reasonable.

[0142] Embodiment 2

[0143] Please refer to Figure 2 As shown, for the parts not described in detail in this embodiment, refer to the description content of Embodiment 1. A method for intelligent control of physical books based on the Internet is provided, including:

[0144] S1. Obtain readers' behavior data;

[0145] S2. Obtain readers' reading preference data based on readers' behavior data;

[0146] S3. Adjust the inventory levels of different books according to readers' reading preference data; by monitoring the inventory levels of different books, determine whether to generate a warning message for book inventory levels.

[0147] Since the electronic device introduced in this embodiment is the electronic device used in an intelligent control system for physical books based on the Internet in an embodiment of the present application, based on an intelligent control system for physical books based on the Internet introduced in an embodiment of the present application, those skilled in the art can understand the specific implementation manners and various variations of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiment of the present application will not be described in detail here. As long as those skilled in the art implement the electronic device used in an intelligent control system for physical books based on the Internet in an embodiment of the present application, it falls within the scope protected by the present application.

[0148] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain a formula that is closest to the actual situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0149] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. Any technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements should also be regarded as within the protection scope of the present invention.

Claims

1. An Internet-based intelligent control system for physical books, characterized in that, Including: A behavior data acquisition module, used to acquire reader behavior data; A reading preference analysis module, used to obtain reader reading preference data according to the reader behavior data; An inventory management module, used to adjust the inventory of different books according to the reader reading preference data; By monitoring the inventory of different books, judge whether to generate a warning message for the book inventory; The various modules are connected by wired and / or wireless means to realize data transmission between the modules.

2. The intelligent control system for physical books based on the Internet according to claim 1, wherein The reader behavior data includes reader borrowing activity, reader evaluation data and collection data.

3. The intelligent control system for physical books based on the Internet according to claim 2, wherein, The method for obtaining the reader behavior data includes: The reader borrowing activity includes the borrowing duration of books by the reader within a week, the number of books of the same type borrowed, and the persistence of borrowing books; when the reader borrows a book, the RFID tag on the book is scanned by a self-service book borrowing and returning machine set in the bookstore or library, and the self-service book borrowing and returning machine records the borrowing duration of books by the reader within a week and the number of books of the same type borrowed; The borrowing activity of the readers is as follows: where ql is the borrowing duration of books by readers within a week; yk is the number of books of the same type borrowed within a week; zm is the persistence of borrowing books within a week; Divide the borrowing duration of the books by a reader within a week by the number of books of the same type borrowed within a week to obtain the average borrowing duration of each book, and use the average borrowing duration to reflect the sustainability of borrowing books; combine the borrowing duration ql of the books by a reader within a week and the number yk of books of the same type borrowed within a week to calculate the sustainability of borrowing books within a week The reader evaluation data includes the score, comment duration and number of comments of the reader on the book within a week; A reader feedback box is set at the service desk of the bookstore or library. The reader leaves a post-view comment and score on the book in the feedback box, and the staff records the score and number of comments of the reader on the book within a week; the comment duration of the reader within a week is recorded by the camera in the bookstore or library; The collection data includes the collection time of books by the reader within a week, the return rate of the collected books and the collection method of the collected books; Readers select their preferred books from the browsing shelves in the bookstore or library, notify the front desk staff to collect the preferred books, and the staff record the time when the readers collect books within a week; by recording the number bx of books that the readers cancel the collection within a week and the total number jv of books collected within a week by the staff, calculate the return rate of the collected books within a week. The way of collecting books includes collecting the books to be pre-collected in a classified form.

4. An Internet-based intelligent control system for physical books according to claim 3, characterized in that, The method for collecting pre-collected books in a classified form includes: Classify the collection methods of the collected books through the K-means clustering algorithm to obtain q different types of clusters. Each cluster represents a collection method of books with different type characteristics. The collection methods of books with different type characteristics include collecting books according to the theme or field, collecting books according to the author or editor, collecting books according to the publication time, collecting books according to the reading status, and collecting books according to the decoration style.

5. An Internet-based intelligent control system for physical books according to claim 4, characterized in that, The method for obtaining the reader reading preference data according to the reader behavior data includes: Perform standard deviation normalization processing on the obtained reader behavior data, convert it into a standard normal distribution with a mean of 0 and a standard deviation of 1, eliminate the dimensional influence between the data, and finally organize and obtain the normalized feature data set; Construct a reading preference prediction model. The input of the reading preference prediction model is historical feature data; the output label is the reader reading preference data; combine machine learning and deep learning technologies to continuously update and optimize the reading preference prediction model, and update the reading preference prediction model by continuously collecting and analyzing reader behavior data.

6. The intelligent control system for physical books based on the Internet according to claim 5, characterized in that, The construction method of the reading preference prediction model includes: Collect a data training set, including input data and corresponding labels; divide the data set into a training set and a test set; use the historical feature data set of the data training set as the input and the corresponding reader reading preference data as the output label to train the reading preference prediction model; the reading preference prediction model is a decision tree model. During the training process of the reading preference prediction model, node splitting is performed according to different features of the historical feature dataset. A feature is selected from the historical feature dataset as the splitting feature of the current node, and the initial entropy of the current node is calculated using the formula in information theory to calculate entropy. When calculating each feature value as a splitting point, the entropy after division is calculated, and the information gain is obtained by calculating the difference between the initial entropy and the entropy after division. The feature value with the maximum information gain is selected from the calculated information gains as the best splitting scheme for the current node. The maximum information gain corresponds to being able to maximize the purity of the readers' reading preferences. Based on the best splitting scheme, child nodes are created, and the corresponding historical feature datasets are assigned to each child node. For each child node, the above steps are recursively executed until the number of the historical feature dataset is lower than the preset number threshold of the historical feature dataset, and then the node splitting stops. Use the Bayesian optimization method to tune the parameters in the reading preference prediction model. The specific steps are as follows: Determine the minimum information gain parameter for feature selection in the decision tree model that needs to be tuned. Prepare an initial set of data training sets, and evaluate the performance metrics of the model under a given parameter configuration by calculating the accuracy metric; the formula for calculating accuracy is: where oj is the number of reader reading preference data that is predicted correctly; pw is the total number of reader reading preference data; Use the initial data training set to establish a Gaussian process initial probability model between the parameter configuration and the performance metric. Start the iterative optimization process, and each iteration includes the following steps: a. According to the current probability model, use Gaussian process sampling to select the next parameter configuration. b. Use the selected parameter configuration to calculate its performance metric by calculating the accuracy. c. Add the new parameter configuration and performance metric to the data training set. d. Update the probability model by adding the new data training set to the model training process to update the mapping relationship between the parameters and the performance. e. Repeat steps a to d until the predetermined number of iterations is reached. Output the best parameter configuration: After the iterative optimization is completed, according to the result of the evaluation function, select the parameter configuration with the best performance metric as the final model configuration to obtain the trained reading preference prediction model, and use the trained reading preference prediction model to predict the current reader feature dataset to obtain the readers' reading preference data.

7. An Internet-based intelligent control system for physical books according to claim 6, characterized in that, The method for adjusting the inventory levels of different books according to the readers' reading preference data includes: The readers' reading preference data includes the types of books preferred by readers within a week, the number of searches for preferred books, and the search duration of preference data, which are recorded by setting up self-service book borrowing and returning machines in bookstores or libraries to record the types of books preferred by readers within a week, the number of searches for preferred books, and the search duration of preference data. Perform standard deviation normalization processing on the obtained readers' reading preference data, convert it into a standard normal distribution with a mean of 0 and a standard deviation of 1, eliminate the dimensionality impact between the data, and finally organize and obtain the normalized reading preference dataset. Check whether there are missing values in the reading preference dataset; for the columns containing missing values, regard them as the target columns to be filled; split the reading preference dataset into two parts: known data and missing data; the known data includes the types of books preferred by readers within a week, the number of searches for preferred books, and the search duration of preference data, and the missing data only contains the types of books preferred by readers within a week. For the types of books preferred by readers in the missing data part within a week, use the known data for Lagrange interpolation; construct the Lagrange interpolation polynomial; use the known types of books preferred by readers within a week as the independent variable, and the corresponding number of searches for preferred books within a week and the search duration of preferred data within a week as the dependent variables; use the constructed Lagrange interpolation polynomial to predict the types of books within a week of the missing data, obtain the corresponding number of searches for preferred books within a week and the search duration of preferred data within a week, and fill the missing values obtained by interpolation prediction back into the target column of the reading preference dataset; Conduct further data analysis and evaluation based on the filled reading preference dataset to verify the rationality and accuracy of the interpolation results; Perform outlier processing on the reading preference dataset through the IQR statistical method: calculate the 25th percentile Q1 and 75th percentile Q3 of the dataset, and calculate IQR = Q3 - Q1; calculate the upper bound UB = Q3 + 1.5 * IQR and the lower bound LB = Q1 - 1.5 * IQR; check each data point in the reading feature dataset, and if the data point is less than the lower bound or greater than the upper bound, mark it as an outlier, delete or correct the data points marked as outliers, and finally organize them into a reading feature dataset; Construct a book inventory prediction model; the input data is the historical reading feature dataset, the output label is the sales volume of different books, the entity book intelligent management and control system records the sales volume of different books obtained, and reminds the staff to adjust the inventory of different books through the sales volume of different books.

8. An Internet-based intelligent control system for physical books according to claim 7, characterized in that, The construction method of the book inventory prediction model includes: The book inventory prediction model is an LSTM model. Prepare the training dataset, including input data and corresponding labels; the input data is the historical reading feature dataset, and the output label is the sales volume of different books; divide the dataset into a training set and a test set to evaluate the performance of the model; Convert the historical reading feature dataset into sequence data suitable for processing by the LSTM model; use time steps to represent each data point in the sequence, and construct the input sequence and output sequence; use the deep learning framework TensorFlow to create an LSTM model; define the structure and parameter settings of the LSTM layer, including the number of LSTM units and the size of the hidden layer, and adjust according to the scale and complexity of the historical reading feature dataset; since the historical reading feature dataset is relatively small, select a smaller number of LSTM units to avoid overfitting, and the size of the hidden layer is comparable to the number of LSTM units; The method for converting the historical reading feature dataset into sequence data suitable for processing by the LSTM model includes: The historical reading feature dataset is recorded once a week. Use the historical reading feature dataset of several weeks as the sequence length, and set the time step to 1 to keep consistent with the time scale of the historical reading feature dataset of one week. The model predicts the sales volume of different books in the next week based on the historical reading feature dataset of the past few weeks; Evaluate the performance of the trained LSTM model using a test set, and use the root mean square error as an indicator to measure the difference between the predicted sales volumes of different books and the actual sales volumes of different books; by using the Adam optimizer, adjust the weights and parameters according to the gradient information of the mean square error loss function to minimize the value of the loss function; The method of adjusting the weights and parameters according to the gradient information of the mean square error loss function to minimize the value of the loss function includes: Adjust the weights according to the gradient information of the mean square error loss function: Initialize the weights of the LSTM model using the Xavier initialization method; input the training data set into the LSTM model, and calculate the predicted values of the model through forward propagation; calculate the mean square error loss function between the predicted values and the actual values. Calculate the gradient of the loss function with respect to the model weights, and transmit the gradient from the output layer to the input layer through backpropagation; Update the weights of the model according to the gradient information and the rules of the Adam optimizer; use the following formula for weight update: PRX = PE - YH·η; where, PRX is the new weight of the LSTM model; PE is the old weight of the LSTM model; YH is the learning rate that controls the weight update speed; η is the gradient of the LSTM model weights; Iterate the above steps repeatedly, gradually adjust the weights of the LSTM model to reduce the value of the loss function and improve the prediction accuracy of the model; use the trained LSTM model for book inventory prediction, input the current reading feature data set, and make predictions through the model to obtain the sales volumes of different books.

9. The intelligent control system for physical books based on the Internet according to claim 8, characterized in that, The method of judging whether to generate a book inventory warning message by monitoring the book inventory includes: Monitor the inventory levels of different books through the intelligent entity book management system, and compare the inventory levels of different books with the preset inventory thresholds of different books; If the inventory level of a different book is less than or equal to the preset inventory threshold of the different book, it is judged that a book inventory warning message is generated; If the inventory level of a different book is greater than the preset inventory threshold of the different book, it is judged that no book inventory warning message is generated; 10. An intelligent control method for physical books based on the Internet, which is implemented based on the intelligent control system for physical books based on the Internet described in any one of claims 1 to 9, and is characterized in that, including: S1. Obtain reader behavior data; S2. Obtain reader reading preference data according to the reader behavior data; S3. Adjust the inventory levels of different books according to the reader reading preference data; judge whether to generate a book inventory warning message by monitoring the inventory levels of different books.

Images