A book procurement prediction method based on the changing trends of user and item popularity

Abstract

The present invention discloses a book procurement prediction method based on the changing trends of user and item popularity. This method involves data collection. The data collected includes user and book information. User information includes user gender and borrowing records. The preference status of users for the same type of books is obtained through the category and keyword tag information of the books. The borrowing start and end times are used to facilitate the statistics of the number of borrowed copies of books within a time period, and the borrowing popularity trend of the book is obtained. The present invention captures the changing trend of popularity by collecting data including user and book information, using one-hot encoding to process book categories and keyword tags, and using a sliding window method to process continuous features. This method combines the Wide&Deep model in the recommendation field and makes improvements. The Wide part is used to capture the interactions of large-scale sparse features, and the Deep part is used to learn the deep feature expressions of users and books. Finally, the quantity of books to be procured is adjusted according to the predicted value, so as to provide an auxiliary decision for book procurement.

Description

Technical Field

[0001] The present invention relates to the technical field of big data and collaborative filtering recommendation systems, and particularly to a book procurement prediction method based on the changing trends of user and item popularity. Background Art

[0002] In existing book management systems, procurement decisions mainly rely on traditional statistical analysis and empirical judgment, which has certain limitations. Due to the rapid changes in market dynamics and user preferences, traditional methods often struggle to accurately capture these changes. Additionally, existing recommendation systems mainly focus on the short-term popularity of books, while ignoring the long-term trends of popularity and users' responses to these trends, resulting in some books being in short supply or oversupplied, causing unreasonable utilization of book resources. Summary of the Invention

[0003] To solve the above problems, the present invention provides a book procurement prediction method based on the changing trends of user and item popularity, which can more comprehensively predict the market demand for books and improve the accuracy and efficiency of procurement.

[0004] The present invention adopts the following technical solutions to achieve the above invention objectives: A book procurement prediction method based on the changing trends of user and item popularity, the method comprising the following steps:

[0005] 1) Data collection. Collect user and book information, where user information includes user gender and borrowing records to capture the borrowing trends of users for books; book information includes category, search times, borrowing start and end times, book keyword tags, and the number of copies. By the category and keyword tag information of the book, obtain the preference status of users for the same type of books. The borrowing start and end times are convenient for counting the number of borrowed copies of the book within a statistical time period to obtain the borrowing popularity trend of the book.

[0006] 2) Item category feature processing. Encode the book category and keyword tags by one-hot encoding, convert the categorical variables into a form that can be understood and processed by a computer, so that the model can use this information for learning. Each category is represented as a vector, where only one element is 1 and the rest are 0. The position of this element represents the index of the category, so each category has a unique encoding.

[0007] 3) Item continuous feature processing. Use the sliding window method to measure the changing trend of popularity. This method is based on a fixed-length time window, and captures the trend by calculating the popularity change of the item within each window. The following are the general steps for measuring the changing trend of popularity using the sliding window method:

[0008] Select the window size: First, it is necessary to select an appropriate window size, that is, the time span △t for each slide. The window size can be adjusted according to the characteristics and requirements of the data, and the average borrowing duration of books by users can be used as the window size.

[0009] Construct a sliding window: Divide the entire time range into multiple windows according to the selected window size. Each window contains continuous data for a period of time.

[0010] Calculate the popularity index: In each window, calculate the current search count as the search popularity of the current window to form a search popularity vector. In addition, calculate the borrowing trend based on the start time and end time of borrowing. If the borrowing period overlaps with the current window, it will be counted as borrowing popularity, and count the borrowing popularity of each book in the current window to form a borrowing popularity vector. Finally, calculate the number of books borrowed in the current window, and the maximum number of borrowed books is the label.

[0011] 4) User feature processing: Encode the user's gender and borrowing record features using one-hot encoding.

[0012] 5) Improve the output layer of the Wide&Deep model. Change the output layer of the model from the softmax or sigmoid function in the classification problem to a linear output layer suitable for the regression problem, and directly output a real value.

[0013] 6) The loss function of the Wide&Deep model uses the mean squared error to calculate the loss between the predicted required number of copies and the actual maximum number of copies.

[0014] 7) During model training, the Wide part inputs the borrowing records of users and books to capture the interaction of large-scale sparse features. The Deep part mainly learns the feature vectors of items and user feature vectors to learn the deep feature expressions of users and books.

[0015] 8) Adjust the output range: According to the requirements of the maximum and minimum number of copies of all current books, it may be necessary to adjust the output range of the model.

[0016] 9) Purchase prediction: After the model training is completed, predict the demand for each book. If the predicted value is greater than 80% of the current maximum quantity of the book, it is recommended to appropriately increase the quantity of the book, and the increase can be calculated according to the following formula:

[0017] Specific increased number of copies = current data * n.

[0018] Beneficial effects:

[0019] The present invention captures the trend of popularity changes by collecting data including user and book information, processing book categories and keyword tags using one-hot encoding, and processing continuous features in a sliding window manner. This method combines and improves the Wide&Deep model in the recommendation field. The Wide part is used to capture the interactions of large-scale sparse features, and the Deep part is used to learn the deep feature expressions of users and books. The quantity of books to be purchased is adjusted according to the predicted value, so as to provide an auxiliary decision for book procurement. Brief Description of the Drawings

[0020] Figure 1 It is a structural diagram of the improved Wide&Deep model based on the trend of popularity changes of users and items in the present invention.

[0021] Figure 2 It is a flowchart of the book procurement prediction method based on the trend of popularity changes of users and items in the present invention. Detailed Embodiment

[0022] The following further details the present invention in conjunction with the drawings of the specification.

[0023] As Figure 1 and Figure 2 shown, the present invention provides a book procurement prediction method based on the trend of popularity changes of users and items. The method includes the following steps:

[0024] 1) Data collection. Collect data including user and book information. The user information includes user gender and borrowing records to capture the borrowing trend of users for books. The book information includes category, search times, borrowing start and end times, book keyword tags, and the number of copies. The preference status of users for the same type of books is obtained through the category and keyword tag information of the books. The borrowing start and end times are used to conveniently count the number of borrowed copies of books within a time period to obtain the borrowing popularity trend of the book.

[0025] 2) Processing of item category features. Encode the book category and keyword tags in a one-hot encoding manner to convert the categorical variables into a form that can be understood and processed by a computer, so that the model can use this information for learning. Each category is represented as a vector, where only one element is 1 and the rest are 0. The position of this element represents the index of the category, so each category has a unique encoding.

[0026] 3) Processing of item continuous features. Use the sliding window method to measure the trend of popularity changes. This method is based on a fixed-length time window and captures the trend by calculating the popularity changes of items within each window. The following are the general steps for measuring the trend of popularity changes using the sliding window method:

[0027] Select window size: First, it is necessary to select an appropriate window size, that is, the time span △t for each slide. The window size can be adjusted according to the characteristics and requirements of the data, and the average borrowing duration of users for books can be used as the window size.

[0028] Construct a sliding window: Divide the entire time range into multiple windows according to the selected window size. Each window contains continuous data for a period of time.

[0029] Calculate popularity metrics: Within each window, calculate the current search count as the search popularity of the current window to form a search popularity vector. Additionally, calculate the borrowing trend based on the start time and end time of borrowing. If the borrowing period overlaps with the current window, it will be counted as borrowing popularity, and count the borrowing popularity of each book in the current window to form a borrowing popularity vector. Finally, calculate the number of books on loan within the current window, and the maximum number of borrowed books is the label.

[0030] 4) User feature processing: Encode the user gender and borrowing record features using one-hot encoding.

[0031] 5) Improve the output layer of the Wide&Deep model. Change the output layer of the model from the softmax or sigmoid function in classification problems to a linear output layer suitable for regression problems, and directly output a real value.

[0032] 6) Use the mean squared error for the loss function of the Wide&Deep model to calculate the loss between the predicted required number of copies and the actual maximum number of copies.

[0033] 7) During model training, the Wide part inputs the borrowing records of users and books to capture the interactions of large-scale sparse features. The Deep part mainly learns the feature vectors of items and user feature vectors to learn the deep feature representations of users and books.

[0034] 8) Adjust the output range: According to the requirements of the maximum and minimum number of copies among all current books, it may be necessary to adjust the output range of the model.

[0035] 9) Procurement prediction: After the model training is completed, predict the demand for each book. If the predicted value is greater than 80% of the current maximum quantity of the book, it is recommended to appropriately increase the quantity of the book, and the increase can be based on the following formula:

[0036] Specific increase in the number of copies = current data * n.

[0037] The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A book procurement prediction method based on the changing trends of user and item popularity, characterized in that, The method includes the following steps: 1) Data collection; Collect information including users and books. The user information includes user gender and borrowing records, which are used to capture the borrowing trends of users for books. The book information includes category, search times, borrowing start and end times, book keyword tags, and the number of copies. The preference status of users for the same type of books is obtained through the category and keyword tag information of the books. The borrowing start and end times are used to count the number of borrowed copies of the book within a time period to obtain the borrowing popularity trend of the book; 2) Item category feature processing; Encode the book category and keyword tags by one-hot encoding, convert the categorical variables into a form that can be understood and processed by a computer, so that the model can use this information for learning. Each category is represented as a vector, where only one element is 1 and the rest are 0. The position of this element represents the index of the category, and each category has a unique encoding; 3) Item continuous feature processing; Use the sliding window method to measure the popularity change trend. Based on a fixed-length time window, capture the trend by calculating the popularity change of items within each window. The following are the steps to measure the popularity change trend using the sliding window method: Select the window size: First, it is necessary to select an appropriate window size, that is, the time span △t of each slide. The window size is adjusted according to the characteristics and requirements of the data, and the average borrowing duration of users for books is used as the window size; Construct the sliding window: Divide the entire time range into multiple windows according to the selected window size. Each window contains continuous data for a period of time; Calculate the popularity index: Within each window, calculate the current search times as the search popularity of the current window to form a search popularity vector. In addition, calculate the borrowing trend based on the start time and end time of borrowing. If the borrowing period overlaps with the current window, count the borrowing popularity of each book in the current window to form a borrowing popularity vector. Finally, calculate the number of copies borrowed within the current window, and the maximum number of borrowed copies is the label label; 4) User feature processing: Encode the user gender and borrowing record features using one-hot encoding; 5) Improve the output layer of the Wide&Deep model. Change the output layer of the model from the output layer of the softmax or sigmoid function suitable for classification problems to a linear output layer suitable for regression problems, and directly output a real value; 6) Use the mean squared error for the loss function of the Wide&Deep model to calculate the loss between the predicted required number of copies and the actual maximum number of copies; 7) During model training, the Wide part inputs the borrowing records of users and books to capture the interaction of large-scale sparse features. The Deep part mainly learns the feature vectors of items and user feature vectors to learn the deep feature expressions of users and books; 8) Adjust the output range: Adjust the output range of the model according to the requirements of the maximum and minimum number of copies among all current books; 9) Purchase prediction: After the model training is completed, predict the demand for each type of book. If the predicted value is greater than 80% of the current maximum quantity of the book, increase the quantity of the book.

Images