Book classification method, device, equipment and storage medium

Abstract

The application discloses a book classification method, device and equipment and a storage medium. The method comprises the following steps: determining the first similarity of a to-be-screened book and a reference book according to the picture features of the to-be-screened book and the picture features of the reference book; determining a similar target book of the reference book in the to-be-screened book according to the first similarity; and updating the classification category of the similar target book according to the classification category of the reference book. The application improves the accuracy of book classification.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a book classification method, apparatus, device, and storage medium. Background Technology

[0002] When e-reading platforms recommend books of a certain type, the recommendations often include books outside the recommended category. For example, gardening books may appear in the category of educational supplements. These books outside the recommended category are usually due to errors in manual or automatic labeling, which cause them to be mixed into the recommended category. Furthermore, e-reading platforms cannot exclude these incorrectly categorized books when classifying them, and therefore cannot guarantee the accuracy of book classification. Summary of the Invention

[0003] The main objective of this invention is to provide a book classification method, apparatus, device, and storage medium, aiming to solve the problem of how to improve the accuracy of book classification.

[0004] To achieve the above objectives, the present invention provides a book classification method, which includes the following steps:

[0005] Based on the image features of the books to be screened and the image features of the reference books, the first similarity between the books to be screened and the reference books is determined;

[0006] Based on the first similarity, similar target books to the reference books are determined among the books to be screened;

[0007] Update the classification of the similar target books based on the classification of the reference books.

[0008] Optionally, before the step of determining the first similarity between the book to be screened and the reference book based on the image features of the book to be screened and the image features of the reference book, the method further includes:

[0009] Input the original book into the first classification model to obtain the first feature of the original book;

[0010] Input the reference book into the first classification model to obtain the second feature of the reference book;

[0011] The books to be screened are determined from the original books based on the second similarity between the first feature and the second feature.

[0012] Optionally, before the step of determining the first similarity between the book to be screened and the reference book based on the image features of the book to be screened and the image features of the reference book, the method further includes:

[0013] The books to be screened are input into the second classification model to obtain the third feature of the books to be screened;

[0014] The reference book is input into the second classification model to obtain the fourth feature of the reference book;

[0015] The image features of the books to be screened are determined based on the third feature, and the image features of the reference books are determined based on the fourth feature.

[0016] Optionally, the step of determining the image features of the book to be screened based on the third feature, and determining the image features of the reference book based on the fourth feature, includes:

[0017] Based on the first feature and the third feature of the books to be screened, the image features of the books to be screened are fused together.

[0018] The image features of the reference book are obtained by fusing the second and fourth features of the reference book.

[0019] Optionally, before the step of determining the book to be screened from the original books based on the second similarity between the first feature and the second feature, the method further includes:

[0020] Based on the multi-dimensional feature information of the first feature, the one-dimensional feature information of the original book is determined;

[0021] Based on the multi-dimensional feature information of the second feature, the one-dimensional feature information of the reference book is determined;

[0022] Based on the one-dimensional feature information of the original book and the one-dimensional feature information of the reference book, the second similarity between the first feature and the second feature is determined.

[0023] Optionally, before the step of inputting the original book into the first classification model to obtain the first feature of the original book, the method further includes:

[0024] A pre-defined neural network model is trained based on source domain samples and target domain samples;

[0025] If the classification accuracy of the neural network model is greater than a preset threshold, then the first classification model is determined based on the neural network model.

[0026] Optionally, after the step of updating the classification category of the similar target books according to the classification category of the reference books, the method further includes:

[0027] Determine the target recommendation list where the reference books are located;

[0028] Add the similar target books after updating the category to the target recommendation list.

[0029] To achieve the above objectives, the present invention also provides a book classification device, which includes a memory, a processor, and a book classification program stored in the memory and executable on the processor. When the book classification program is executed by the processor, it implements the various steps of the book classification method as described above.

[0030] To achieve the above objectives, the present invention also provides a computer-readable storage medium storing a book classification program, which, when executed by a processor, implements the various steps of the book classification method described above.

[0031] This invention provides a book classification method, apparatus, device, and storage medium. Based on image features of a book to be screened and image features of a reference book, a first similarity is determined between the book to be screened and the reference book. Based on the first similarity, similar target books are identified among the books to be screened. The classification category of the similar target books is updated based on the classification category of the reference books. By identifying similar target books through the first similarity of image features between the book to be screened and the reference book, and correcting the classification category of the similar target books based on the classification category of the reference book, the book classification becomes more accurate. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the hardware structure of the book classification device according to an embodiment of the present invention;

[0033] Figure 2 This is a flowchart illustrating the first embodiment of the book classification method of the present invention;

[0034] Figure 3 This is a flowchart illustrating the second embodiment of the book classification method of the present invention;

[0035] Figure 4 This is a flowchart illustrating the third embodiment of the book classification method of the present invention;

[0036] Figure 5 This is a schematic diagram of the logical structure of the book classification device according to an embodiment of the present invention.

[0037] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0038] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0039] The main solution of this invention is as follows: determine the first similarity between the book to be screened and the reference book based on the image features of the book to be screened and the image features of the reference book; determine similar target books among the books to be screened based on the first similarity; and update the classification category of the similar target books based on the classification category of the reference book.

[0040] Similar target books are identified by the first similarity of the image features of the books to be screened and the reference books. The classification of the similar target books is then corrected based on the classification of the reference books, making the book classification more accurate.

[0041] As one implementation solution, book sorting equipment can be like... Figure 1 As shown.

[0042] The present invention relates to a book sorting device, which includes: a processor 101, such as a CPU, a memory 102, and a communication bus 103. The communication bus 103 is used to enable communication between these components.

[0043] Memory 102 can be high-speed RAM or stable memory (non-volatile memory), such as disk storage. Figure 1 As shown, the memory 102, which is a computer-readable storage medium, may include a book classification program; and the processor 101 can be used to call the book classification program stored in the memory 102 and perform the following operations:

[0044] Based on the image features of the books to be screened and the image features of the reference books, the first similarity between the books to be screened and the reference books is determined;

[0045] Based on the first similarity, similar target books to the reference books are determined among the books to be screened;

[0046] Update the classification of the similar target books based on the classification of the reference books.

[0047] Optionally, the processor 101 can be used to call a book classification program stored in the memory 102 and perform the following operations:

[0048] Input the original book into the first classification model to obtain the first feature of the original book;

[0049] Input the reference book into the first classification model to obtain the second feature of the reference book;

[0050] The books to be screened are determined from the original books based on the second similarity between the first feature and the second feature.

[0051] Optionally, the processor 101 can be used to call a book classification program stored in the memory 102 and perform the following operations:

[0052] The books to be screened are input into the second classification model to obtain the third feature of the books to be screened;

[0053] The reference book is input into the second classification model to obtain the fourth feature of the reference book;

[0054] The image features of the books to be screened are determined based on the third feature, and the image features of the reference books are determined based on the fourth feature.

[0055] Optionally, the processor 101 can be used to call a book classification program stored in the memory 102 and perform the following operations:

[0056] Based on the first feature and the third feature of the books to be screened, the image features of the books to be screened are fused together.

[0057] The image features of the reference book are obtained by fusing the second and fourth features of the reference book.

[0058] Optionally, the processor 101 can be used to call a book classification program stored in the memory 102 and perform the following operations:

[0059] Based on the multi-dimensional feature information of the first feature, the one-dimensional feature information of the original book is determined;

[0060] Based on the multi-dimensional feature information of the second feature, the one-dimensional feature information of the reference book is determined;

[0061] Based on the one-dimensional feature information of the original book and the one-dimensional feature information of the reference book, the second similarity between the first feature and the second feature is determined.

[0062] Optionally, the processor 101 can be used to call a book classification program stored in the memory 102 and perform the following operations:

[0063] A pre-defined neural network model is trained based on source domain samples and target domain samples;

[0064] If the classification accuracy of the neural network model is greater than a preset threshold, then the first classification model is determined based on the neural network model.

[0065] Optionally, the processor 101 can be used to call a book classification program stored in the memory 102 and perform the following operations:

[0066] Determine the target recommendation list where the reference books are located;

[0067] Add the similar target books after updating the category to the target recommendation list.

[0068] Based on the hardware architecture of the aforementioned book classification equipment, embodiments of the book classification method of the present invention are proposed.

[0069] Reference Figure 2 , Figure 2 This is a first embodiment of the book classification method of the present invention, which includes the following steps:

[0070] Step S10: Determine the first similarity between the book to be screened and the reference book based on the image features of the book to be screened and the image features of the reference book;

[0071] Optionally, the reference book, or positive sample, is a book that has been identified as belonging to a preset category a and has been manually or actually determined to belong to the preset category. For example, the reference book is a book that has been classified as an educational supplement and has been manually or actually determined to belong to the educational supplement category.

[0072] Optionally, the books to be screened are negative samples, meaning books that have been categorized into a non-preset category. For example, the books to be screened might be categorized as "non-educational supplements." Since the categorization of the books to be screened has not been manually determined or actually verified, some books may belong to the preset category 'a' but are classified as non-preset categories.

[0073] Optionally, the image features of the books to be screened and the reference books can be image features of the book covers or image features of the book catalog.

[0074] Optionally, the books to be screened are those similar to the reference books selected from the original books. The original books, i.e., the negative samples, are books that have been classified into categories other than the pre-defined categories.

[0075] Optionally, the first similarity can be the cosine similarity between the image features of the book to be screened and the image features of the reference book, obtained by calculating the cosine of the angle between the feature vectors of the image features of the book to be screened and the feature vectors of the image features of the reference book. Optionally, the first similarity can be obtained by calculating the Euclidean distance between the image features of the book to be screened and the image features of the reference book.

[0076] Step S20: Based on the first similarity, determine the similar target books of the reference books among the books to be screened;

[0077] Optionally, when the first similarity between the book to be screened and the reference book is greater than a preset threshold, the book to be screened is determined to be a similar target book, wherein the similar target book is a book that should have been identified as a preset category a, but was identified as a book that is not a preset category.

[0078] Optionally, when the first similarity between the book to be screened and the reference book is less than or equal to a preset threshold, the book to be screened is determined to be a book in a non-preset category.

[0079] Optionally, similar target books can be multiple books whose first similarity to the reference book is greater than a preset threshold, such as the top 200 books whose first similarity to the reference book is greater than the preset threshold.

[0080] Step S30: Update the classification category of the similar target books according to the classification category of the reference books.

[0081] Optionally, similar target books that should have been identified as books in the preset category a are identified as books in a different category. Therefore, it is necessary to correct the classification of similar target books based on the classification of the reference books to make the book classification more accurate.

[0082] Optionally, after step S30, the method further includes: determining the target recommendation list where the reference book is located; adding similar target books with updated classification categories to the target recommendation list, and recommending similar target books. Optionally, upon receiving a recommendation request for a preset classification category, book recommendations are made based on the target recommendation lists where the reference book and similar target books are located.

[0083] In this embodiment, the technical solution involves determining a first similarity between the book to be screened and the reference book based on their image features; identifying similar target books among the books to be screened based on the first similarity; and updating the classification category of the similar target books based on the classification category of the reference books. By determining similar target books through the first similarity of their image features and correcting the classification category of the similar target books based on the classification category of the reference books, the book classification becomes more accurate.

[0084] Reference Figure 3 , Figure 3 This is a second embodiment of the book classification method of the present invention. Based on the first embodiment, before step S10, it further includes:

[0085] Step S40: Input the original book into the first classification model to obtain the first feature of the original book;

[0086] Step S50: Input the reference book into the first classification model to obtain the second feature of the reference book;

[0087] Step S60: Based on the second similarity between the first feature and the second feature, determine the book to be screened from the original books.

[0088] Optionally, the original books are those classified into non-preset categories. Since the classification results of the original books have not been manually determined or actually verified, some books in the original books may belong to the preset category 'a' but are classified as non-preset categories. The original books include the books to be screened, and the number of original books is greater than the number of books to be screened. Optionally, the books to be screened are those selected from the original books that have a certain relevance to the reference books.

[0089] Optionally, the first feature of the original book is an image feature, and the second feature of the reference book is an image feature. Optionally, the image feature can be an image feature of the book cover or an image feature of the book's table of contents.

[0090] Optionally, the second similarity can be the cosine similarity between the image features of the original book and the image features of the reference book, obtained by calculating the cosine of the angle between the feature vectors of the image features of the original book and the feature vectors of the image features of the reference book. Optionally, the second similarity can be obtained by calculating the Euclidean distance between the image features of the original book and the image features of the reference book.

[0091] Optionally, when the second similarity between the original book and the reference book is greater than a preset threshold, the original book is determined to be a book to be screened, wherein the book to be screened is a book that should have been identified as a preset category a, but was identified as a book that is not a preset category, and the book to be screened is related to the reference book.

[0092] Optionally, if the second similarity between the original book and the reference book is less than or equal to a preset threshold, the original book is determined not to be a book to be screened.

[0093] Optionally, the books to be screened can be multiple books whose second similarity to the reference book is greater than a preset threshold, and these multiple books can be sorted according to their similarity. For example, the top 1500 books whose second similarity to the reference book is greater than the preset threshold can be used as the books to be screened.

[0094] Optionally, a preset neural network model is trained based on source domain samples and target domain samples; if the classification accuracy of the neural network model is greater than a preset threshold, then the first classification model is determined based on the neural network model. Optionally, the first classification model is a model obtained through transfer learning; optionally, the source domain is the original domain of the transferred model in transfer learning, and the target domain is the intended domain of the transferred model. For example, the original domain of the transferred model is for classifying posters, and the intended domain is for classifying books.

[0095] Optionally, a ResNet (Deep Residual Network)-50 model is used for transfer learning. This involves obtaining a ResNet-50 deep neural network model already trained on the ImageNet dataset (the source domain samples). This neural network model achieves over 80% accuracy on the classification task of the dataset. The ImageNet dataset is a computer vision dataset containing 1.28 million images and 1000 categories. Since transfer learning can solve similar problems based on generalizable machine learning models, and the basic task of the ResNet-50 model is the same as the binary classification task of determining whether an image belongs to a predefined classification list, transfer learning can be performed using a deep learning model whose parameters have already been trained on the dataset and which has achieved high accuracy.

[0096] To improve the accuracy of the ResNet-50 model on the book dataset, i.e., the target domain samples, some parts of the network need to be adjusted to suit the target domain samples. The ResNet-50 model consists of four residual blocks, pooling layers, and fully connected layers. Since the original training dataset used by the ResNet-50 model was the ImageNet dataset, which has 1000 categories, the final fully connected layer of the ResNet-50 model had a dimension of 1000. When transferring the ResNet-50 model to the book classification task, to adapt to the characteristics of binary classification, the 1000-dimensional fully connected layer needs to be removed and replaced with a fully connected layer more suitable for binary classification.

[0097] After the ResNet-50 model is trained, the output of the ResNet-50 model residual block is taken. This output is the image features of the entire book learned by the residual network. In the ResNet-50 model, this output is a three-dimensional matrix. In order to facilitate the calculation of the similarity between books, while preserving as much image information as possible, the feature data of the entire three-dimensional matrix is ​​flattened, that is, the high-dimensional tensor is transformed into a one-dimensional tensor.

[0098] Optionally, before step S60, the method further includes: determining the one-dimensional feature information of the original book based on the multi-dimensional feature information of the first feature; determining the one-dimensional feature information of the reference book based on the multi-dimensional feature information of the second feature; and determining the second similarity between the first feature and the second feature based on the one-dimensional feature information of the original book and the one-dimensional feature information of the reference book. Optionally, a flattening operation, i.e., a tensor flattening operation, is performed on the first feature and the second feature to transform the high-dimensional feature information into one-dimensional feature information.

[0099] In this embodiment, the original books are input into a first classification model to obtain the first feature of the original books; reference books are input into the first classification model to obtain the second feature of the reference books; and books to be screened are determined from the original books based on the second similarity between the first and second features. This initial screening of the original books using the second similarity improves the efficiency of subsequent, more precise screening by obtaining books that are similar to the reference books.

[0100] Reference Figure 4 , Figure 4 This is a third embodiment of the book classification method of the present invention. Based on the first or second embodiment, before step S10, it further includes:

[0101] Step S70: Input the books to be screened into the second classification model to obtain the third feature of the books to be screened;

[0102] Step S80: Input the reference book into the second classification model to obtain the fourth feature of the reference book;

[0103] Step S90: Determine the image features of the book to be screened based on the third feature, and determine the image features of the reference book based on the fourth feature.

[0104] Optionally, the third feature can be used as the image feature of the books to be screened, and the fourth feature can be used as the image feature of the reference books.

[0105] Optionally, the image features of the books to be screened are obtained by fusing the first and third features of the books to be screened; and the image features of the reference books are obtained by fusing the second and fourth features of the reference books.

[0106] Optionally, the classification accuracy of the second classification model is greater than that of the first classification model. Since the classification efficiency of the second classification model is lower than that of the first classification model, the image features output by the first classification model are used to initially screen the original books, and then the image features output by the second classification model are used to further screen the books to be screened.

[0107] Optionally, the second classification model uses a deep learning model with Vision Transformer as its backbone network to further filter the books to be selected, in order to accurately determine similar target books. The distributed ResNet-50 model can initially and quickly filter out books associated with reference books based on image similarity. However, as a model using CNN as its backbone network, ResNet-50 relies on convolutional kernels for computation. While reducing parameters and computational cost, this is not conducive to capturing global features, resulting in insufficient accuracy in classification results. The deep learning model with Vision Transformer as its backbone can capture global information about the book images and can capture long-distance dependencies, but its computational cost is much higher than that of the ResNet-50 model. The ResNet-50 model (the first classification model) is used to initially filter the original books to obtain the books to be selected, and the deep learning model with Vision Transformer as its backbone is used to further filter these books.

[0108] In this embodiment, the books to be screened are input into a second classification model to obtain a third feature; reference books are input into the second classification model to obtain a fourth feature; image features of the books to be screened are determined based on the third feature, and image features of the reference books are determined based on the fourth feature. A first similarity is determined using the image features of the books to be screened and the image features of the reference books, facilitating further screening of the books to be screened based on the first similarity, thus improving the accuracy of the classification of the books to be screened.

[0109] Reference Figure 5 The present invention also provides a book classification device, the book classification device comprising:

[0110] The calculation module 100 is used to determine the first similarity between the book to be screened and the reference book based on the image features of the book to be screened and the image features of the reference book.

[0111] The determining module 200 is used to determine, based on the first similarity, similar target books among the books to be screened;

[0112] The update module 300 is used to update the classification category of the similar target books according to the classification category of the reference books.

[0113] Optionally, before the step of determining the first similarity between the book to be screened and the reference book based on the image features of the book to be screened and the image features of the reference book, the method further includes:

[0114] Input the original book into the first classification model to obtain the first feature of the original book;

[0115] Input the reference book into the first classification model to obtain the second feature of the reference book;

[0116] The books to be screened are determined from the original books based on the second similarity between the first feature and the second feature.

[0117] Optionally, before the step of determining the first similarity between the book to be screened and the reference book based on the image features of the book to be screened and the image features of the reference book, the method further includes:

[0118] The books to be screened are input into the second classification model to obtain the third feature of the books to be screened;

[0119] The reference book is input into the second classification model to obtain the fourth feature of the reference book;

[0120] The image features of the books to be screened are determined based on the third feature, and the image features of the reference books are determined based on the fourth feature.

[0121] Optionally, the step of determining the image features of the book to be screened based on the third feature, and determining the image features of the reference book based on the fourth feature, includes:

[0122] Based on the first feature and the third feature of the books to be screened, the image features of the books to be screened are fused together.

[0123] The image features of the reference book are obtained by fusing the second and fourth features of the reference book.

[0124] Optionally, before the step of determining the book to be screened from the original books based on the second similarity between the first feature and the second feature, the method further includes:

[0125] Based on the multi-dimensional feature information of the first feature, the one-dimensional feature information of the original book is determined;

[0126] Based on the multi-dimensional feature information of the second feature, the one-dimensional feature information of the reference book is determined;

[0127] Based on the one-dimensional feature information of the original book and the one-dimensional feature information of the reference book, the second similarity between the first feature and the second feature is determined.

[0128] Optionally, before the step of inputting the original book into the first classification model to obtain the first feature of the original book, the method further includes:

[0129] A pre-defined neural network model is trained based on source domain samples and target domain samples;

[0130] If the classification accuracy of the neural network model is greater than a preset threshold, then the first classification model is determined based on the neural network model.

[0131] Optionally, after the step of updating the classification category of the similar target books according to the classification category of the reference books, the method further includes:

[0132] Determine the target recommendation list where the reference books are located;

[0133] Add the similar target books after updating the category to the target recommendation list.

[0134] The present invention also provides a book classification device, the book classification device including a memory, a processor, and a book classification program stored in the memory and executable on the processor, wherein when the book classification program is executed by the processor, it implements the various steps of the book classification method as described in the above embodiments.

[0135] The present invention also provides a computer-readable storage medium storing a book classification program, which, when executed by a processor, implements the various steps of the book classification method described in the above embodiments.

[0136] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0137] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, system, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, system, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, system, article, or apparatus that includes that element.

[0138] Through the above description of the embodiments, those skilled in the art can clearly understand that the systems described in the embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a computer-readable storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, parking management device, air conditioner, or network device, etc.) to execute the systems described in the various embodiments of the present invention.

[0139] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A book classification method, characterized in that, The book classification methods include: The original book is input into the first classification model to obtain the first feature of the original book; the first classification model is a convolutional neural network model. The reference book is input into the first classification model to obtain the second feature of the reference book; the first feature and the second feature are image features, which are the image features of the book cover or the image features of the book table of contents; the reference book is a positive sample, which is a book that has been identified as a preset classification category and has been manually determined or actually identified as a preset classification category; Based on the second similarity between the first feature and the second feature, books to be screened are determined from the original books; The books to be screened are input into the second classification model to obtain the third feature of the books to be screened; the books to be screened are negative samples, which are books that are classified into non-preset categories; the second classification model is a visual converter model; the classification accuracy of the second classification model is greater than the classification accuracy of the first classification model; The reference book is input into the second classification model to obtain the fourth feature of the reference book; Based on the first feature and the third feature of the book to be screened, the image features of the book to be screened are fused together. Based on the second and fourth features of the reference book, the image features of the reference book are obtained by fusing them. Based on the image features of the books to be screened and the image features of the reference books, the first similarity between the books to be screened and the reference books is determined; Based on the first similarity, similar target books to the reference books are determined among the books to be screened; Update the classification of the similar target books based on the classification of the reference books.

2. The book classification method as described in claim 1, characterized in that, Before the step of determining the book to be screened in the original books based on the second similarity between the first feature and the second feature, the method further includes: Based on the multi-dimensional feature information of the first feature, the one-dimensional feature information of the original book is determined; Based on the multi-dimensional feature information of the second feature, the one-dimensional feature information of the reference book is determined; Based on the one-dimensional feature information of the original book and the one-dimensional feature information of the reference book, the second similarity between the first feature and the second feature is determined.

3. The book classification method as described in claim 1, characterized in that, Before the step of inputting the original book into the first classification model to obtain the first feature of the original book, the method further includes: A pre-defined neural network model is trained based on source domain samples and target domain samples; If the classification accuracy of the neural network model is greater than a preset threshold, then the first classification model is determined based on the neural network model.

4. The book classification method as described in claim 1, characterized in that, After the step of updating the classification category of the similar target books according to the classification category of the reference books, the method further includes: Determine the target recommendation list where the reference books are located; Add the similar target books after updating the category to the target recommendation list.

5. A book sorting device, characterized in that, The book sorting device includes: The calculation module is used to input the original books into a first classification model to obtain the first feature of the original books; the first classification model is a convolutional neural network model; input reference books into the first classification model to obtain the second feature of the reference books; the first feature and the second feature are image features, the image features being the image features of the book cover or the image features of the book catalog; the reference books are positive samples, which are books identified as belonging to a preset classification category and have been manually determined or actually identified as belonging to the preset classification category; based on the second similarity between the first feature and the second feature, determine the books to be screened from the original books; input the books to be screened into a second classification model to obtain the images to be screened. The third feature of the book; the book to be screened is a negative sample, which is a book classified into a non-preset category; the second classification model is a visual converter model; the classification accuracy of the second classification model is greater than the classification accuracy of the first classification model; the reference book is input into the second classification model to obtain the fourth feature of the reference book; the image features of the book to be screened are obtained by fusing the first and third features of the book to be screened; the image features of the reference book are obtained by fusing the second and fourth features of the reference book; the first similarity between the book to be screened and the reference book is determined based on the image features of the book to be screened and the image features of the reference book. The determining module is used to determine similar target books to the reference books among the books to be screened based on the first similarity. The update module is used to update the classification category of the similar target books according to the classification category of the reference books.

6. A book sorting device, characterized in that, The book classification device includes a memory, a processor, and a book classification program stored in the memory and executable on the processor. When executed by the processor, the book classification program implements the various steps of the book classification method as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a book classification program, which, when executed by a processor, implements the steps of the book classification method as described in any one of claims 1-4.

Images