Personalized recommendation method of electronic book and related device
By constructing a three-dimensional emotional model that uses facial expression characteristics, combined with historical reading behavior data, the problem of not taking into account user emotions and reading habits in the existing e-book recommendation method is solved, and a more personalized and efficient e-book is achieved Recommended, improving user experience.
Patent Information
- Application Number
- CN202411003087.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing e-book recommendation methods fail to effectively consider users' reading habits and emotions, resulting in the recommended e-books that do not match users' needs, and the efficiency of obtaining emotional information through brain wave signals is low and the calculation amount is large, so it cannot provide personalization and convenience. Reading experience.
By obtaining the user's historical reading behavior data, extracting the keywords and reading order of reading content, building user reading portraits, combining facial expression characteristics to use the pleasure-activity-advantage three-dimensional emotional model to calculate the current emotional information, filter and adjust the e-book's Recommendation order and time, providing personalized e-book recommendations.
Improve the matching degree and user experience of e-book recommendations, provide e-book recommendations that are more in line with user emotions and reading habits, solve the problems of low brain wave signal acquisition efficiency and large amount of calculation, and enhance users' reading interest and satisfaction .
Smart Images

Figure CN118916546A_ABST
Abstract
Description
Technical Field
[0001] This invention relates primarily to the field of data processing technology, and more particularly to a personalized recommendation method and related apparatus for e-books. Background Technology
[0002] With the development of internet technology, e-book reading has gradually replaced traditional paper reading, becoming the preferred reading method for users. To improve the user reading experience, e-book service providers recommend e-books to users. Current e-book recommendation methods typically rely solely on users' age and gender, failing to consider the differences in each user's reading habits and preferences. This results in low matching rates between recommended e-books and user needs, failing to provide targeted recommendations and meet the personalized needs of e-book users. Furthermore, current e-book recommendation methods rarely consider incorporating user emotions into the recommendation process. Only a small number of e-book service providers do incorporate emotions into their recommendations, but this often involves obtaining emotional information through brainwave signals. While this method considers user emotions, brainwave signal collection requires specialized equipment, and the process of acquiring emotional information from brainwave signals is slow and computationally intensive, resulting in insufficient flexibility in e-book recommendations and failing to provide users with a convenient and satisfactory e-book recommendation experience. Furthermore, existing e-book recommendation methods directly recommend e-books to users' reading devices after obtaining them, without considering the recommended reading order and reading time. This fails to provide users with a comfortable reading experience, resulting in personalized e-book recommendations failing to achieve the expected results. Summary of the Invention
[0003] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a personalized recommendation method and related device for e-books, bringing users a comfortable reading experience and making the personalized recommendation of e-books achieve a more ideal effect.
[0004] To address the aforementioned technical problems, this invention provides a personalized recommendation method for e-books, the method comprising:
[0005] Obtain users' historical direct reading behavior data and historical indirect reading behavior data, and generate reading information tags based on the historical direct reading behavior data and historical indirect reading behavior data;
[0006] A user reading profile is constructed based on the reading information tags, and user e-book reading habit information and e-book reading preference information are generated based on the user reading profile.
[0007] Obtain the user's target image sequence, perform attention feature extraction on the target image sequence, and obtain facial expression features;
[0008] Based on the facial expression features, the current emotional parameters are calculated using a three-dimensional emotion model of pleasure, activation, and dominance, and the user's current emotional information is obtained based on the current emotional parameters.
[0009] Based on the current emotional information, e-book reading habit information, and e-book reading preference information, e-books are filtered to obtain a recommended e-book set;
[0010] The reading characteristics of similar user groups in the recommended e-book set are obtained, and the e-book reading order and e-book reading time are generated based on the reading characteristics. The e-book reading order and e-book reading time are adjusted based on the current emotional information and e-book reading habit information to obtain the recommended e-book reading order and recommended e-book reading time.
[0011] The recommended e-book set is sorted according to the recommended e-book reading order to obtain a sorted recommended e-book set, and the sorted recommended e-book set and the recommended e-book reading time are sent to the user's reading terminal.
[0012] Optionally, generating reading information tags based on the historical direct reading behavior data and historical indirect reading behavior data includes:
[0013] Based on the aforementioned historical reading direct behavior data and historical reading indirect behavior data, several keywords related to the reading content are extracted;
[0014] The weight of each keyword in the reading content is determined based on the word frequency algorithm and the maximum entropy algorithm. The reading content tags are generated based on several reading content keywords and the weight of each reading content keyword.
[0015] Based on the aforementioned historical direct reading behavior data and historical indirect reading behavior data, the reading time and reading order of different categories of e-books are extracted;
[0016] A reading time ratio matrix is generated based on the reading time of different types of e-books, a reading order ratio matrix is generated based on the reading order of different types of e-books, and reading behavior tags are generated based on the reading time ratio matrix and the reading order ratio matrix.
[0017] Reading information tags are generated based on the reading content tags and reading behavior tags.
[0018] Optionally, the step of constructing a user reading profile based on the reading information tags, and generating user e-book reading habit information and e-book reading preference information based on the user reading profile, includes:
[0019] The reading information tags are homomorphically encrypted to obtain ciphertext tags;
[0020] Obtain the authorization information of the encrypted tags, and construct a user reading profile based on the encrypted tags and authorization information;
[0021] Based on the user reading profile, reading feature information is determined, and information is associated and integrated based on the reading feature information to obtain the user's e-book reading habits and e-book reading preferences.
[0022] Optionally, the step of extracting attention features from the target image sequence to obtain facial expression features includes:
[0023] The target image sequence is preprocessed to obtain a preprocessed target image sequence;
[0024] Feature maps are extracted from the preprocessed target image sequence to obtain several feature maps, and each feature map is divided into regions to obtain several region feature maps.
[0025] Max pooling is performed on several regional feature maps to obtain the maximum feature map, and average pooling is performed on several regional feature maps to obtain the average feature map.
[0026] Based on the maximum feature map and the average feature map, a spatial attention feature map is obtained by splicing and dimensionality reduction, and an attention feature map is generated based on the spatial attention feature map.
[0027] Facial expression features are obtained using a preset facial motion unit based on the attention feature map.
[0028] Optionally, the step of calculating the current emotional parameters based on the facial expression features using a three-dimensional emotional model of pleasure-activation-dominance, and obtaining the user's current emotional information based on the current emotional parameters, includes:
[0029] Based on a preset nonlinear function, the facial expression features are mapped to a preset target emotion category space for emotion classification to obtain emotion classification information;
[0030] Construct a VGG-OCC model and calculate the sentiment space vector based on the VGG-OCC model;
[0031] Based on the aforementioned emotional space vector, parameter quantization mapping is performed to obtain a three-dimensional emotional model of pleasure, activation, and dominance.
[0032] Based on the aforementioned three-dimensional emotion model of pleasure-activation-dominance, the emotion dimension is calculated using the facial expression features and emotion classification information, and the current emotion parameters are calculated based on the emotion dimension.
[0033] The current sentiment information is generated based on the current sentiment parameters and sentiment classification information.
[0034] Optionally, the step of filtering ebooks based on the current emotional information, ebook reading habit information, and ebook reading preference information to obtain a recommended ebook set includes:
[0035] Obtain e-book feature information of several e-books, calculate the matching degree based on the e-book feature information and e-book reading preference information, obtain the matching degree calculation result, and perform initial e-book screening based on the matching degree calculation result to obtain the first e-book set;
[0036] The first set of e-books is adjusted based on the e-book reading habit information to obtain the second set of e-books;
[0037] The emotion annotations and e-book feature information of each e-book in the second e-book set are obtained, and the emotional information features of the e-books are generated based on the emotion annotations and e-book feature information of each e-book.
[0038] Obtain the correspondence between the emotional information features of e-books and the emotional information of users, and adjust the second set of e-books based on the correspondence and the current emotional information to obtain a recommended set of e-books.
[0039] Optionally, generating the e-book reading order and e-book reading time based on the reading features includes:
[0040] Each ebook in the recommended ebook set is split into segments using a preset splitting rule to obtain several ebook segments;
[0041] Obtain the comprehension difficulty coefficient of each ebook segment, and generate the ebook reading order and ebook reading time based on the comprehension difficulty coefficient of each ebook segment and the reading features.
[0042] In addition, the present invention also provides a personalized recommendation device for e-books, the device comprising:
[0043] Reading information tag generation module: used to acquire users' historical direct reading behavior data and historical indirect reading behavior data, and generate reading information tags based on the historical direct reading behavior data and historical indirect reading behavior data;
[0044] Reading habit information and preference information generation module: used to construct a user reading profile based on the reading information tags, and generate the user's e-book reading habit information and e-book reading preference information based on the user reading profile;
[0045] Facial expression feature acquisition module: used to acquire the user's target image sequence, extract attention features from the target image sequence, and obtain facial expression features;
[0046] Current Emotion Information Acquisition Module: Used to calculate current emotion parameters based on the facial expression features using a three-dimensional emotion model of pleasure-activation-dominance, and to acquire the user's current emotion information based on the current emotion parameters;
[0047] E-book filtering module: used to filter e-books based on the current sentiment information, e-book reading habit information, and e-book reading preference information to obtain a recommended e-book set;
[0048] Recommended reading order and time acquisition module: used to acquire the reading characteristics of similar user groups in the recommended e-book set, generate e-book reading order and e-book reading time based on the reading characteristics, and adjust the e-book reading order and e-book reading time based on the current emotional information and e-book reading habit information to obtain recommended e-book reading order and recommended e-book reading time;
[0049] E-book collection sorting and sending module: used to sort the recommended e-book collection based on the recommended e-book reading order, obtain the sorted recommended e-book collection, and send the sorted recommended e-book collection and recommended e-book reading time to the user's reading terminal.
[0050] In addition, the present invention also provides an electronic device, which includes a processor and a memory. The memory is used to store instructions, and the processor is used to call the instructions in the memory to cause the electronic device to execute the above-described personalized e-book recommendation method.
[0051] In addition, the present invention also provides a computer-readable storage medium that stores computer instructions that, when executed on an electronic device, cause the electronic device to perform the aforementioned personalized e-book recommendation method.
[0052] In this embodiment of the invention, reading information tags are generated based on reading content tags and reading behavior tags generated from historical direct reading behavior data and historical indirect reading behavior data. This makes the obtained reading information tags more comprehensive and better reflects the user's reading habits. Constructing a user reading profile based on ciphertext tags obtained through homomorphic encryption of the reading information tags and authorization information enhances the security and efficiency of the user profile configuration process, prevents the leakage of user reading profile information, and generates user e-book reading habit information and e-book reading preference information based on the user reading profile. This allows for a more comprehensive and accurate consideration of the user's reading needs, making the recommended e-book collection more suitable for the user's requirements. Based on facial expression features, a three-dimensional emotion model of pleasure, activation, and dominance is used to calculate current emotional parameters to obtain the user's current emotional information. Based on this current emotional information, e-book reading habits, and e-book reading preferences, a recommended e-book set is selected. By incorporating the user's current emotional information into the e-book selection process, e-books that match the user's current emotional state are recommended, improving the matching degree between the recommended e-books and the user's actual needs, further enhancing the user's reading interest, and making the recommended e-book set more closely aligned with the user's current actual needs. This also solves the problems of slow emotional information acquisition, high computational load, and insufficient flexibility in existing technologies that rely on dedicated equipment to collect EEG signals to identify user emotions. Based on the current emotional information and e-book reading habits, the e-book reading order and e-book reading time generated by the reading characteristics of similar user groups are adjusted to obtain a recommended e-book reading order for sorting the recommended e-book set. The sorted recommended e-book set and recommended reading time are sent to the user's reading terminal, providing a more personalized e-book experience that better meets the user's needs, bringing a comfortable reading experience, and achieving a more ideal effect in personalized e-book recommendations. Attached Figure Description
[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0054] Figure 1 This is a flowchart illustrating the personalized e-book recommendation method in an embodiment of the present invention.
[0055] Figure 2 This is a schematic diagram of the structure of the personalized recommendation device for e-books in an embodiment of the present invention;
[0056] Figure 3This is a schematic diagram of the structural composition of the electronic device in an embodiment of the present invention. Detailed Implementation
[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0058] Example 1
[0059] Please see Figure 1 , Figure 1 This is a flowchart illustrating a personalized e-book recommendation method according to an embodiment of the present invention. The method includes:
[0060] S11: Obtain the user's historical direct reading behavior data and historical indirect reading behavior data, and generate reading information tags based on the historical direct reading behavior data and historical indirect reading behavior data;
[0061] In the specific implementation of this invention, the step of generating reading information tags based on the historical direct reading behavior data and historical indirect reading behavior data includes: extracting several reading content keywords based on the historical direct reading behavior data and historical indirect reading behavior data; determining the weight of each reading content keyword based on a word frequency algorithm and a maximum entropy algorithm, and generating reading content tags based on the several reading content keywords and the weight of each reading content keyword; extracting the reading time and reading order of different categories of e-books based on the historical direct reading behavior data and historical indirect reading behavior data; generating a reading time ratio matrix based on the reading time of different categories of e-books, generating a reading order ratio matrix based on the reading order of different categories of e-books, and generating reading behavior tags based on the reading time ratio matrix and the reading order ratio matrix; and generating reading information tags based on the reading content tags and reading behavior tags.
[0062] Specifically, the system retrieves users' historical direct and indirect reading behavior data from the database. Direct reading behavior data includes ebook reading content, category, time, and order. Indirect reading behavior data includes ebook collection content, category, collection reading time, collection reading order, and number of collections. Based on this data, several reading content keywords are extracted. Word segmentation is then performed using a word segmentation model to extract words from the reading and collection data, resulting in several content words. The similarity of each word to other content words is calculated, and a word segmentation coverage rate is generated based on this similarity. The sum of the reciprocals of each word's positional order in the reading and collection content is calculated and used as a positional feature value. The topic probability distribution of each word is calculated. A comprehensive feature value is calculated based on the word segmentation coverage rate, positional feature value, and topic probability distribution. The content words with the highest predetermined number of comprehensive feature values are selected as the reading content keywords. The weights of each keyword in the reading content are determined using a term frequency (TF) algorithm and a maximum entropy algorithm. The TF algorithm employs a term frequency-inverse text frequency (IF-MRF) algorithm, while the maximum entropy algorithm is a criterion based on probabilistic model learning, aiming to find the probability distribution with maximum entropy under all constraints. The weights of each keyword are determined based on the results of both algorithms. Reading content tags are generated based on several keywords and their weights. Each keyword's weight is compared to a preset weight threshold; if a keyword's weight is less than the threshold, it is rejected. A frequent pattern mining algorithm is then used to select several keywords from the remaining keywords as reading content tags. Finally, reading time and order for different categories of ebooks are extracted based on historical direct and indirect reading behavior data. A reading time ratio matrix is generated based on the reading time of different categories of e-books. The ratio of the reading time of each category to the total reading time of all categories is calculated. A reading time ratio matrix is then constructed based on this ratio. A reading order ratio matrix is generated based on the reading order of different categories of e-books. A reading order ratio matrix is also generated based on the proportion of each category's reading order within all reading orders. Reading behavior tags are generated based on the reading time ratio matrix and the reading order ratio matrix. A reading tag matrix is obtained by calculating the reading tag weights using the reading time ratio matrix and the reading order ratio matrix. Reading behavior tags are then generated based on the reading tag matrix. Reading information tags are constructed based on the reading content tags and the reading behavior tags.
[0063] S12: Construct a user reading profile based on the reading information tags, and generate user e-book reading habit information and e-book reading preference information based on the user reading profile;
[0064] In a specific implementation of this invention, the step of constructing a user reading profile based on the reading information tags and generating user e-book reading habit information and e-book reading preference information based on the user reading profile includes: homomorphically encrypting the reading information tags to obtain ciphertext tags; obtaining authorization information for the ciphertext tags; constructing a user reading profile based on the ciphertext tags and authorization information; determining reading feature information based on the user reading profile; and performing information association and integration based on the reading feature information to obtain user e-book reading habit information and e-book reading preference information.
[0065] Specifically, the reading information tags are homomorphically encrypted using a homomorphic encryption algorithm to obtain ciphertext tags. This homomorphic encryption allows for the computation of multiple ciphertext tags before decryption, avoiding the computational overhead of decrypting and recalculating each tag individually, thus improving the efficiency of converting ciphertext tags into user profiles. It also enables keyless computation of ciphertext tags, reducing communication costs and shifting computational tasks, thereby balancing computational resources. Authorization information for the ciphertext tags is obtained from an authorization information database. A user reading profile is constructed based on the ciphertext tags and authorization information using a pre-defined user profile model. This user reading profile is also encrypted. Reading feature information, including reading record information, reading trajectory sequence information, and reading time information, is determined based on the user reading profile. Based on the reading feature information, information association and integration are performed, that is, the relationship between the reading trajectory order and reading time of the user's various types of e-books is obtained, and the reading feature information is integrated according to the relationship to avoid information clutter and obtain the user's e-book reading habits and e-book reading preferences.
[0066] S13: Obtain the user's target image sequence, extract attention features from the target image sequence, and obtain facial expression features;
[0067] In a specific implementation of this invention, the step of extracting attention features from the target image sequence to obtain facial expression features includes: preprocessing the target image sequence to obtain a preprocessed target image sequence; extracting feature maps from the preprocessed target image sequence to obtain several feature maps, and dividing each feature map into regions to obtain several region feature maps; performing max pooling on the several region feature maps to obtain a maximum feature map, and performing average pooling on the several region feature maps to obtain an average feature map; performing concatenation and dimensionality reduction processing based on the maximum feature map and the average feature map to obtain a spatial attention feature map, and generating an attention feature map based on the spatial attention feature map; and obtaining facial expression features using a preset facial action unit based on the attention feature map.
[0068] Specifically, the target image sequence is acquired using the camera built into the user's reading terminal; that is, the user's current image is captured through the camera without the need for dedicated equipment. The target image sequence is preprocessed by normalizing pixel values to obtain a normalized target image sequence. Face detection and alignment are then performed on the normalized target image sequence to obtain a preprocessed target image sequence. Feature map extraction is performed on the preprocessed target image sequence using a spatial pyramid. The spatial pyramid consists of four convolutional layers. The output features of the last residual block of each of the four convolutional layers are used as one layer of feature maps in the spatial pyramid, thus extracting multiple layers of feature maps, resulting in several feature maps. Each feature map is divided into regions using two 1x1 convolutions and one 3x3 convolution. The 3x3 convolution is positioned between the two 1x1 convolutions; the first 1x1 convolution is used for dimensionality reduction, and the second for dimensionality increase, ensuring that the output and input sizes are consistent. Batch normalization and ReLU activation are applied after each convolution. Local convolutions capture local appearance changes, and the learning weights for each region are updated independently, resulting in several region feature maps. Spatial pyramids help locate regions of feature variation. Max pooling is applied to these region feature maps along the channel axis to obtain the maximum value of each channel at each location. A maximum feature map is generated based on the maximum values of each channel. Average pooling is then applied to these region feature maps along the channel axis to obtain the average value of each channel at each location. An average feature map is generated based on the average values of each channel. Attention feature extraction emphasizes decisive features and suppresses invalid features. Based on the aforementioned maximum and average feature maps, concatenation and dimensionality reduction are performed to extract feature vectors from both maps. The importance of each feature vector within the entire facial space is determined, and a first and second feature maps are generated based on this importance. These first and second feature maps are then resized and concatenated to form a combined feature map. The sigmoid function is used to reduce the dimensionality of the combined feature map, resulting in a spatial attention feature map. An attention feature map is then generated based on this spatial attention feature map. The regional feature map is multiplied by the spatial attention feature map, and the product is added back to the regional feature map to obtain the final attention feature map. Facial expression features are then acquired using preset facial motion units (FMUs) based on this attention feature map. These FMUs contain various facial micro-expression features; combining the attention feature map with the preset FMUs yields the final facial expression features.
[0069] S14: Calculate the current emotional parameters based on the facial expression features using a three-dimensional emotional model of pleasure, activation, and dominance, and obtain the user's current emotional information based on the current emotional parameters;
[0070] In a specific implementation of this invention, the step of calculating current emotional parameters based on the facial expression features using a three-dimensional emotional model of pleasure, activation, and dominance, and obtaining the user's current emotional information based on the current emotional parameters, includes: mapping the facial expression features to a preset target emotional category space based on a preset nonlinear function to perform emotional classification and obtain emotional classification information; constructing a VGG-OCC model and calculating an emotional space vector based on the VGG-OCC model; performing parameter quantization mapping based on the emotional space vector to obtain a three-dimensional emotional model of pleasure, activation, and dominance; calculating an emotional dimension based on the facial expression features and emotional classification information using the three-dimensional emotional model of pleasure, activation, and dominance, and calculating current emotional parameters based on the emotional dimension; and generating current emotional information based on the current emotional parameters and emotional classification information.
[0071] Specifically, facial expression features are mapped to a preset target emotion category space based on a preset nonlinear function for emotion classification. A hyperbolic tangent tanh nonlinear mapping function is used to map facial expression features to the preset target emotion category space. Simultaneously, a softmax classifier is used to determine the emotion category. The softmax classifier uses the cross-entropy loss function as the optimization objective for model training. The gradient of the loss function is calculated through backpropagation, and relevant parameters in the model are continuously updated to obtain the optimal model, yielding the emotion category with the highest prediction probability, thus obtaining the emotion classification information. A VGG-OCC model is constructed. The VGG model is a deep convolutional neural network with powerful image feature extraction capabilities proposed by the Visual Geometry Group. The OCC model is a cognitive emotion evaluation model proposed by Ortony, Clore, and Collins, a structured model used in the field of artificial intelligence, which includes 22 emotion states. The VGG model and the OCC model are combined to construct the VGG-OCC model. Based on the VGG-OCC model, an emotion space vector is calculated. The feature sample set is sampled to obtain several sampled frames. These sampled frames are preprocessed to remove interference information, resulting in preprocessed sampled frames. The preprocessed sampled frames are then used to encode facial expressions using the VGG-OCC model to obtain the emotion space vector. Based on this emotion space vector, parameter quantization mapping is performed. A transformation relationship is established using linear regression based on the emotion space vector, and a transformation matrix is constructed to establish the quantization mapping relationship between the emotion space vector and the three-dimensional emotion space. Based on the transformation relationship and the transformation matrix, a three-dimensional emotion model of pleasure, activation, and dominance is constructed. Based on the aforementioned three-dimensional emotion model of pleasure, activation, and dominance, the emotion dimension is calculated using facial expression features and emotion classification information. The emotion value within the three-dimensional emotion model is then calculated based on these features. Based on the emotion value, a preset emotion scale is used to obtain the emotion dimension, and the current emotion parameter is calculated based on this dimension. Instantaneous emotion parameters are calculated using an instantaneous key emotion array based on the emotion dimension. Long-term emotion parameters are calculated using a long-term emotion experience function based on the emotion dimension. Finally, the user's current emotion parameter is calculated using preset weighting ratios based on the instantaneous and long-term emotion parameters. Based on the current emotion parameter and emotion classification information, current emotion information is generated. The emotion classification information reflects whether the user's current emotion is positive or negative, and the current emotion parameter reflects the degree of positivity or negativity of the user's current emotion.
[0072] S15: Based on the current emotional information, e-book reading habit information, and e-book reading preference information, e-books are filtered to obtain a recommended e-book set;
[0073] In a specific implementation of this invention, the step of filtering ebooks based on the current emotional information, ebook reading habit information, and ebook reading preference information to obtain a recommended ebook set includes: acquiring ebook feature information of several ebooks; calculating a matching degree based on the ebook feature information and ebook reading preference information to obtain a matching degree calculation result; and performing initial ebook filtering based on the matching degree calculation result to obtain a first ebook set; adjusting the first ebook set based on the ebook reading habit information to obtain a second ebook set; acquiring the emotion labeling and ebook feature information of each ebook in the second ebook set, and generating ebook emotional information features based on the emotion labeling and ebook feature information of each ebook; acquiring the correspondence between the ebook emotional information features and the user's emotional information, and adjusting the second ebook set based on the correspondence using the current emotional information to obtain a recommended ebook set.
[0074] Specifically, the process involves acquiring ebook feature information for several ebooks, including content feature word segmentation and comment feature word segmentation. A matching degree is calculated based on this feature information and ebook reading preference information. This matching degree can be calculated using a neural network recommendation model. Based on the matching degree calculation results, an initial ebook selection is performed, ranking the ebooks from highest to lowest matching degree. A predetermined number of ebooks at the top of the ranking are selected to form an ebook set, i.e., the first ebook set. The first ebook set is then adjusted based on the ebook reading habit information, specifically by removing ebooks with fewer reading categories from the first ebook set, resulting in a second ebook set. The process also involves acquiring the sentiment annotations and ebook feature information for each ebook in the second ebook set. The database stores the sentiment annotations for each ebook, and based on these, ebook sentiment information features are generated. Furthermore, ebook sentiment keywords are generated based on the keyword features in the sentiment annotations and ebook feature information, and these keywords are used as the ebook sentiment information features. The system obtains the correspondence between the emotional information features of e-books and the user's emotional information, and adjusts the second set of e-books based on the correspondence using the current emotional information. E-books that do not match the user's current emotional information are removed. For example, if the user's current emotional information is negative, e-books labeled with sadness or grief are removed, thus obtaining a recommended set of e-books.
[0075] S16: Obtain the reading characteristics of similar user groups in the recommended e-book set, generate e-book reading order and e-book reading time based on the reading characteristics, and adjust the e-book reading order and e-book reading time based on the current emotional information and e-book reading habit information to obtain the recommended e-book reading order and recommended e-book reading time;
[0076] In the specific implementation of this invention, generating the e-book reading order and e-book reading time based on the reading features includes: splitting each e-book in the recommended e-book set into segments using a preset splitting rule to obtain several e-book segments; obtaining the comprehension difficulty coefficient of each e-book segment; and generating the e-book reading order and e-book reading time based on the comprehension difficulty coefficient of each e-book segment combined with the reading features.
[0077] Specifically, the reading characteristics of similar user groups within the recommended ebook set are obtained. The server aggregates similar user groups with the same recommended ebook set, and their reading characteristics may include the reading order and reading time of the ebooks. Each ebook in the recommended ebook set is segmented using preset segmentation rules. The length of each ebook in the recommended ebook set is analyzed, and the ebook content is divided into segments, pages, or words to obtain several ebook segments. The comprehension difficulty coefficient of each ebook segment is obtained. Based on the comprehension difficulty coefficient of each ebook segment and the reading characteristics, the ebook reading order and ebook reading time are generated. Ebooks with more segments of higher comprehension difficulty coefficients are placed later in the reading order, increasing their reading time. The ebook reading order and ebook reading time are adjusted based on the current emotional information and ebook reading habit information. After the initial adjustment of the ebook reading order and reading time based on the ebook reading habit information, ebooks suitable for the user's current emotional information are placed at the top, and a recommended ebook reading time is generated based on the user's current emotional information, thus obtaining the recommended ebook reading order and recommended ebook reading time.
[0078] S17: Sort the recommended e-book set according to the recommended e-book reading order to obtain the sorted recommended e-book set, and send the sorted recommended e-book set and the recommended e-book reading time to the user's reading terminal.
[0079] In the specific implementation of this invention, the recommended e-book set is sorted based on the recommended e-book reading order to obtain a sorted recommended e-book set. If the recommended e-books are directly recommended to the user's reading terminal after being obtained, and the user sees that the e-books at the top do not match their current mood or reading habits, they may not continue reading the e-books at the back of the recommended e-book set. This may lead to the user losing interest in the recommended e-book set and believing that the entire recommended e-book set does not meet their actual needs. Therefore, sorting the recommended e-book set based on the recommended e-book reading order can provide e-books that better meet the user's personalized e-book needs, bringing the user a comfortable reading experience. Sending the sorted recommended e-book set and recommended e-book reading time to the user's reading terminal provides the user with a personalized reading plan, taking into account the user's current reading needs from multiple aspects, so that the personalized recommendation of e-books achieves a more ideal effect.
[0080] In this embodiment of the invention, reading information tags are generated based on reading content tags and reading behavior tags generated from historical direct reading behavior data and historical indirect reading behavior data. This makes the obtained reading information tags more comprehensive and better reflects the user's reading habits. Constructing a user reading profile based on ciphertext tags obtained through homomorphic encryption of the reading information tags and authorization information enhances the security and efficiency of the user profile configuration process, prevents the leakage of user reading profile information, and generates user e-book reading habit information and e-book reading preference information based on the user reading profile. This allows for a more comprehensive and accurate consideration of the user's reading needs, making the recommended e-book collection more suitable for the user's requirements. Based on facial expression features, a three-dimensional emotion model of pleasure, activation, and dominance is used to calculate current emotional parameters to obtain the user's current emotional information. Based on this current emotional information, e-book reading habits, and e-book reading preferences, a recommended e-book set is selected. By incorporating the user's current emotional information into the e-book selection process, e-books that match the user's current emotional state are recommended, improving the matching degree between the recommended e-books and the user's actual needs, further enhancing the user's reading interest, and making the recommended e-book set more closely aligned with the user's current actual needs. This also solves the problems of slow emotional information acquisition, high computational load, and insufficient flexibility in existing technologies that rely on dedicated equipment to collect EEG signals to identify user emotions. Based on the current emotional information and e-book reading habits, the e-book reading order and e-book reading time generated by the reading characteristics of similar user groups are adjusted to obtain a recommended e-book reading order for sorting the recommended e-book set. The sorted recommended e-book set and recommended reading time are sent to the user's reading terminal, providing a more personalized e-book experience that better meets the user's needs, bringing a comfortable reading experience, and achieving a more ideal effect in personalized e-book recommendations.
[0081] Example 2
[0082] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of the personalized e-book recommendation device in an embodiment of the present invention. The device includes:
[0083] Reading information tag generation module 21: used to acquire the user's historical direct reading behavior data and historical indirect reading behavior data, and generate reading information tags based on the historical direct reading behavior data and historical indirect reading behavior data;
[0084] Reading habit information and preference information generation module 22: used to construct a user reading profile based on the reading information tags, and generate the user's e-book reading habit information and e-book reading preference information based on the user reading profile;
[0085] Facial expression feature acquisition module 23: used to acquire the user's target image sequence, extract attention features from the target image sequence, and obtain facial expression features;
[0086] Current Emotion Information Acquisition Module 24: Used to calculate current emotion parameters based on the facial expression features using a three-dimensional emotion model of pleasure-activation-dominance, and to acquire the user's current emotion information based on the current emotion parameters;
[0087] E-book filtering module 25: used to filter e-books based on the current emotional information, e-book reading habit information and e-book reading preference information, and obtain a recommended e-book set;
[0088] Recommended reading order and time acquisition module 26: used to acquire the reading characteristics of similar user groups of the recommended e-book set, generate e-book reading order and e-book reading time based on the reading characteristics, and adjust the e-book reading order and e-book reading time based on the current emotional information and e-book reading habit information to obtain recommended e-book reading order and recommended e-book reading time;
[0089] E-book set sorting and sending module 27: used to sort the recommended e-book set based on the recommended e-book reading order, obtain the sorted recommended e-book set, and send the sorted recommended e-book set and recommended e-book reading time to the user's reading terminal.
[0090] In the specific implementation of this invention, the specific implementation of the device item can be referred to the implementation of the method item above, and will not be repeated here.
[0091] In this embodiment of the invention, reading information tags are generated based on reading content tags and reading behavior tags generated from historical direct reading behavior data and historical indirect reading behavior data. This makes the obtained reading information tags more comprehensive and better reflects the user's reading habits. Constructing a user reading profile based on ciphertext tags obtained through homomorphic encryption of the reading information tags and authorization information enhances the security and efficiency of the user profile configuration process, prevents the leakage of user reading profile information, and generates user e-book reading habit information and e-book reading preference information based on the user reading profile. This allows for a more comprehensive and accurate consideration of the user's reading needs, making the recommended e-book collection more suitable for the user's requirements. Based on facial expression features, a three-dimensional emotion model of pleasure, activation, and dominance is used to calculate current emotional parameters to obtain the user's current emotional information. Based on this current emotional information, e-book reading habits, and e-book reading preferences, a recommended e-book set is selected. By incorporating the user's current emotional information into the e-book selection process, e-books that match the user's current emotional state are recommended, improving the matching degree between the recommended e-books and the user's actual needs, further enhancing the user's reading interest, and making the recommended e-book set more closely aligned with the user's current actual needs. This also solves the problems of slow emotional information acquisition, high computational load, and insufficient flexibility in existing technologies that rely on dedicated equipment to collect EEG signals to identify user emotions. Based on the current emotional information and e-book reading habits, the e-book reading order and e-book reading time generated by the reading characteristics of similar user groups are adjusted to obtain a recommended e-book reading order for sorting the recommended e-book set. The sorted recommended e-book set and recommended reading time are sent to the user's reading terminal, providing a more personalized e-book experience that better meets the user's needs, bringing a comfortable reading experience, and achieving a more ideal effect in personalized e-book recommendations.
[0092] This invention provides a computer-readable storage medium storing a computer program. When executed by a processor, this program implements the personalized e-book recommendation method of any of the above embodiments. The computer-readable storage medium includes, but is not limited to, any type of disk (including floppy disk, hard disk, optical disk, CD-ROM, and magneto-optical disk), ROM (Read-Only Memory), RAM (Random Access Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory, magnetic cards, or optical cards. In other words, the storage device includes any medium that stores or transmits information in a readable form by a device (e.g., a computer, a mobile phone), and can be a read-only memory, a disk, or an optical disk, etc.
[0093] Example 3
[0094] Please see Figure 3 , Figure 3 This is a schematic diagram of the structural composition of the electronic device in an embodiment of the present invention.
[0095] This invention also provides an electronic device, such as... Figure 3 As shown, the electronic device includes a memory 31, a processor 33, and a computer program 32 stored in the memory 31 and executable on the processor 33. Those skilled in the art will understand that... Figure 3The illustrated electronic device does not constitute a limitation on all devices and may include more or fewer components than illustrated, or combine certain components. Memory 31 can be used to store computer program 32 and various functional modules. Processor 33 runs the computer program 32 stored in memory 31, thereby performing various functional applications and data processing of the device. Memory can be internal memory or external memory, or both. Internal memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, or random access memory. External memory may include hard disks, floppy disks, ZIP disks, USB flash drives, magnetic tapes, etc. Processor 33 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, a single-chip microcomputer, or a processor 33, or any conventional processor, etc. The processors and memories disclosed in this invention include, but are not limited to, these types of processors and memories. The processors and memories disclosed in this invention are merely examples and not intended to be limiting.
[0096] As one embodiment, the electronic device includes: one or more processors 33, a memory 31, and one or more computer programs 32, wherein the one or more computer programs 32 are stored in the memory 31 and configured to be executed by the one or more processors 33, and the one or more computer programs 32 are configured to execute the personalized recommendation method for e-books in any of the above embodiments. For specific implementation details, please refer to the above embodiments, which will not be repeated here.
[0097] In this embodiment of the invention, reading information tags are generated based on reading content tags and reading behavior tags generated from historical direct reading behavior data and historical indirect reading behavior data. This makes the obtained reading information tags more comprehensive and better reflects the user's reading habits. Constructing a user reading profile based on ciphertext tags obtained through homomorphic encryption of the reading information tags and authorization information enhances the security and efficiency of the user profile configuration process, prevents the leakage of user reading profile information, and generates user e-book reading habit information and e-book reading preference information based on the user reading profile. This allows for a more comprehensive and accurate consideration of the user's reading needs, making the recommended e-book collection more suitable for the user's requirements. Based on facial expression features, a three-dimensional emotion model of pleasure, activation, and dominance is used to calculate current emotional parameters to obtain the user's current emotional information. Based on this current emotional information, e-book reading habits, and e-book reading preferences, a recommended e-book set is selected. By incorporating the user's current emotional information into the e-book selection process, e-books that match the user's current emotional state are recommended, improving the matching degree between the recommended e-books and the user's actual needs, further enhancing the user's reading interest, and making the recommended e-book set more closely aligned with the user's current actual needs. This also solves the problems of slow emotional information acquisition, high computational load, and insufficient flexibility in existing technologies that rely on dedicated equipment to collect EEG signals to identify user emotions. Based on the current emotional information and e-book reading habits, the e-book reading order and e-book reading time generated by the reading characteristics of similar user groups are adjusted to obtain a recommended e-book reading order for sorting the recommended e-book set. The sorted recommended e-book set and recommended reading time are sent to the user's reading terminal, providing a more personalized e-book experience that better meets the user's needs, bringing a comfortable reading experience, and achieving a more ideal effect in personalized e-book recommendations.
[0098] Furthermore, the above provides a detailed description of the personalized e-book recommendation method and related apparatus provided by the embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A personalized recommendation method for e-books, It is characterized in that The method comprises: Acquire the user's historical reading direct behavior data and historical reading indirect behavior data, and generate a reading information tag based on the historical reading direct behavior data and historical reading indirect behavior data; Building a user reading profile based on the reading information tag, and generating the user's e-book reading habit information and e-book reading preference information based on the user reading profile; Acquire a target image sequence of a user, extract attention features from the target image sequence, and obtain facial expression features; Calculating current emotion parameters using a three-dimensional emotion model of pleasure-activation-dominance based on the facial expression features, and acquiring current emotion information of the user based on the current emotion parameters; Screening e-books based on the current emotion information, e-book reading habit information and e-book reading preference information to obtain a recommended e-book collection; Acquire reading characteristics of a similar user group of the recommended e-book set, generate an e-book reading order and an e-book reading time based on the reading characteristics, and adjust the e-book reading order and the e-book reading time based on the current emotion information and the e-book reading habit information to obtain a recommended e-book reading order and a recommended e-book reading time; The recommended e-book set is sorted based on the recommended e-book reading order to obtain a sorted recommended e-book set, and the sorted recommended e-book set and the recommended e-book reading time are sent to a reading terminal of the user.
2. The personalized recommendation method for electronic books according to claim 1, It is characterized in that The generating of the reading information tag based on the historical reading direct behavior data and the historical reading indirect behavior data includes: Extracting a number of reading content keywords based on the historical reading direct behavior data and the historical reading indirect behavior data; Determine the weight of each reading content keyword based on a word frequency algorithm and a maximum entropy algorithm, and generate a reading content label based on a number of reading content keywords and the weight of each reading content keyword; Extracting the reading time and reading order of different categories of e-books based on the historical reading direct behavior data and the historical reading indirect behavior data; Generate a reading time ratio matrix based on the reading time of different categories of e-books, generate a reading order ratio matrix based on the reading order of different categories of e-books, and generate a reading behavior label based on the reading time ratio matrix and the reading order ratio matrix; A reading information tag is generated based on the reading content tag and the reading behavior tag.
3. The personalized recommendation method for electronic books according to claim 1, It is characterized in that The step of constructing a user reading profile based on the reading information tag, and generating the user's e-book reading habit information and e-book reading preference information based on the user reading profile includes: Homomorphically encrypt the reading information tag to obtain a ciphertext tag; Obtaining authorization information of the ciphertext label, and constructing a user reading profile based on the ciphertext label and the authorization information; Reading feature information is determined based on the user reading portrait, and information association and integration are performed based on the reading feature information to obtain the user's e-book reading habit information and e-book reading preference information.
4. The personalized recommendation method for electronic books according to claim 1, It is characterized in that The step of extracting attention features from the target image sequence to obtain facial expression features includes: Preprocessing the target image sequence to obtain a preprocessed target image sequence; Extracting feature maps from the preprocessed target image sequence to obtain several feature maps, and dividing each feature map into regions to obtain several regional feature maps; Performing maximum pooling processing on several regional feature maps to obtain a maximum feature map, and performing average pooling processing on several regional feature maps to obtain an average feature map; Performing concatenation and dimensionality reduction processing based on the maximum feature map and the average feature map to obtain a spatial attention feature map, and generating an attention feature map based on the spatial attention feature map; Facial expression features are acquired using preset facial action units based on the attention feature map.
5. The personalized recommendation method for electronic books according to claim 1, It is characterized in that The method of calculating the current emotion parameter by using the pleasure-activation-dominance three-dimensional emotion model based on the facial expression feature, and obtaining the current emotion information of the user based on the current emotion parameter, includes: Based on a preset nonlinear function, the facial expression features are mapped to a preset target emotion category space for emotion classification to obtain emotion classification information; Constructing a VGG-OCC model, and calculating an emotion space vector based on the VGG-OCC model; Perform parameter quantization mapping based on the emotion space vector to obtain a three-dimensional emotion model of pleasure-activation-dominance; Calculating the emotion dimension based on the pleasure-activation-dominance three-dimensional emotion model using the facial expression features and emotion classification information, and calculating the current emotion parameter based on the emotion dimension; Current emotion information is generated based on the current emotion parameter and the emotion classification information.
6. The personalized recommendation method for electronic books according to claim 1, It is characterized in that The step of screening e-books based on the current emotion information, the e-book reading habit information and the e-book reading preference information to obtain a recommended e-book collection includes: Acquiring electronic book characteristic information of a plurality of electronic books, performing matching degree calculation based on the electronic book characteristic information and the electronic book reading preference information to obtain matching degree calculation results, and performing initial electronic book screening based on the matching degree calculation results to obtain a first electronic book set; Adjusting the first electronic book collection based on the electronic book reading habit information to obtain a second electronic book collection; Acquire the emotion annotation and the electronic book characteristic information of each electronic book in the second electronic book set, and generate the electronic book emotion information feature based on the emotion annotation and the electronic book characteristic information of each electronic book; The correspondence between the e-book emotion information feature and the user emotion information is obtained, and based on the correspondence, the second e-book collection is adjusted using the current emotion information to obtain a recommended e-book collection.
7. The personalized recommendation method for electronic books according to claim 1, It is characterized in that The generating the e-book reading order and the e-book reading time based on the reading characteristics includes: Splitting each e-book in the recommended e-book collection into segments using a preset segmentation rule to obtain a plurality of e-book segments; The comprehension difficulty coefficient of each e-book segment is obtained, and the e-book reading order and e-book reading time are generated based on the comprehension difficulty coefficient of each e-book segment and the reading characteristics.
8. A personalized recommendation device for electronic books, It is characterized in that The device comprises: A reading information tag generation module: used to obtain the user's historical reading direct behavior data and historical reading indirect behavior data, and generate a reading information tag based on the historical reading direct behavior data and historical reading indirect behavior data; A reading habit information and preference information generation module: used to construct a user reading portrait based on the reading information tag, and to generate the user's e-book reading habit information and e-book reading preference information based on the user reading portrait; Facial expression feature acquisition module: used to acquire the user's target image sequence, extract attention features from the target image sequence, and obtain facial expression features; Current emotion information acquisition module: used to calculate current emotion parameters based on the facial expression features using a three-dimensional emotion model of pleasure-activation-dominance, and to acquire the user's current emotion information based on the current emotion parameters; An e-book screening module: used to screen e-books based on the current emotion information, e-book reading habit information and e-book reading preference information to obtain a recommended e-book collection; Recommended reading order and time acquisition module: used to acquire reading characteristics of the same user group of the recommended e-book set, generate e-book reading order and e-book reading time based on the reading characteristics, and adjust the e-book reading order and e-book reading time based on the current emotional information and e-book reading habit information to obtain the recommended e-book reading order and recommended e-book reading time; The electronic book collection sorting and sending module is used to sort the recommended electronic book collection based on the recommended electronic book reading order, obtain the sorted recommended electronic book collection, and send the sorted recommended electronic book collection and the recommended electronic book reading time to the user's reading terminal.
9. An electronic device, comprising a processor and a memory, It is characterized in that The memory is used to store instructions, and the processor is used to call the instructions in the memory, so that the electronic device executes the personalized recommendation method for e-books according to any one of claims 1 to 7.
10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores computer instructions, and when the computer instructions are executed on an electronic device, the electronic device executes the personalized recommendation method for electronic books according to any one of claims 1 to 7.
Citation Information
Patent Citations
Electronic book reading plan recommendation system, method thereof, terminal and server
CN106657581A
E-book recommendation method and system, electronic equipment and computer storage medium
CN111274381A
Emotion-based OCC-PAD-OCEAN federal cognitive modeling method
CN113420591A
Intelligent marketing method and system, equipment terminal and readable storage medium
CN116485424A
Expression feature extraction method, device and equipment and computer readable storage medium
CN117422797A
Cited By
Reading management method and reading management device for electronic document
CN120162433A
