A method and system for novel recommendation

Abstract

The present invention discloses a method and system for novel recommendation, and relates to the technical field of novel recommendation. The present invention comprises step S1, data collection, step S2, emotion information collection and analysis, step S3, text analysis and emotion search, step S4, optimization recommendation, step S5, recommendation display, step S6, continuous detection, real-time feedback. The present invention sets a recommendation engine module, and can search for matching novel emotion fragments in a novel emotion context analysis module according to the current emotion state and intensity obtained by a real-time emotion monitoring module of the reader. At the same time, the module can also combine the user preference and the user comprehensive preference portrait in the historical data integration module to screen and sort the matched novels, and give priority to recommending novels that are both in line with the current emotion match and highly consistent with the user's historical preferences, thereby reducing the time for users to find their favorite works and accurately providing users with more favorite works.

Description

Technical Field

[0001] The present invention belongs to the technical field of novel recommendation, and in particular relates to a method and system for novel recommendation. Background Art

[0002] As the scale of electronic novels continues to expand, their number and types are becoming more and more diverse. In this case, when users are looking for their favorite novels, they often need to spend a lot of time and energy browsing through a large number of works, which will undoubtedly bring users a poor reading experience. It is precisely to solve this dilemma that the recommendation system came into being. The recommendation system can push novels that users may be interested in based on their interest characteristics, effectively improving users' reading experience and efficiency.

[0003] In the existing novel recommendation system, the analysis of novel texts often remains at the level of simple keyword matching or text classification based on machine learning, and it is unable to accurately provide users with more desirable works based on their emotions. To address the above issues, the following solutions are proposed. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for novel recommendation. By setting up a recommendation engine module, the current emotional state and intensity obtained by the reader's real-time emotion monitoring module can be used to search for matching novel emotional fragments in the novel emotional context analysis module, thereby solving the problem that the existing technology cannot accurately provide users with more desirable works.

[0005] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0006] The present invention is a method for novel recommendation, and the recommendation method comprises:

[0007] Step S1, data collection: collecting data related to novel recommendation from multiple data sources, including user data and novel data;

[0008] The step S1, data collection, also includes the following contents:

[0009] Step S11: collecting user data;

[0010] Step S12, novel data collection;

[0011] Step S2, emotional information collection and analysis: When a reader reads a novel or enters a reading platform, the system automatically starts a reader emotional monitoring module to collect multimodal data and analyze the reader's real-time emotional state;

[0012] Step S3, text analysis and emotion search: Analyze the collected novel text data to identify the emotional tendency in the novel. The recommendation engine module performs matching search in the novel emotional context analysis module according to the current emotional state, and performs screening and sorting based on user preferences to generate a preliminary recommendation list.

[0013] Step S4, optimized recommendation: combining user preferences with the preference information of readers in the historical data integration module, sorting multiple novels, and updating and optimizing the user's comprehensive preference portrait;

[0014] Step S5, recommendation display: display the recommendation list on the reader's reading interface or recommendation page;

[0015] Step S6, continuous detection and real-time feedback: During the reading process, the system continuously monitors the reader's emotional changes and dynamically optimizes the recommendation list based on the emotions. When the reader emotion monitoring module detects that Xiao Li's emotions fluctuate slightly, the system re-updates the recommendation list according to the matching search and screening sorting process. At the same time, it records Xiao Li's feedback on the recommended novels and provides real-time feedback to the user preference and historical data integration module.

[0016] Preferably, in the step S3, text analysis and emotion search, the specific method for identifying the emotional tendency in the novel includes:

[0017] Construct a sentiment dictionary, where each sentiment word in the dictionary corresponds to a sentiment vector F m =(e1,e2,…,e n ), where n is the number of emotion categories, e i Indicates the weight of the sentiment word on the i-th sentiment;

[0018] For each sentence T in the novel text, segment it into words to get the word sequence and calculate the sentiment vector of the sentence Where l is an index variable used to traverse the vocabulary in the sentence. It starts counting from 1 and increases by 1 each time until it reaches w; F ml Represents the sentiment vector corresponding to the lth word;

[0019] A recurrent neural network is used to process sentence sequences to capture the contextual emotional information of the text. The state update formula of the hidden layer in the RNN is set to p r =tanh(M p p r-1 +M x F Tr +b p ), where M p and M x is the weight matrix, b p is the bias vector, p ris the hidden layer state at the rth time step, tanh is the hyperbolic tangent function, and the sentiment category probability distribution H of each paragraph H is obtained through a fully connected layer. emo (H)=(h1,h2,…,h n )=\softmax(M fc p R +b fc ), where p R is the hidden layer state of the last sentence in the paragraph, M fc and h fc are the weights and biases of the fully connected layer, the \softmax function is used to convert the output into a probability distribution, h i It represents the probability that paragraph H belongs to the i-th sentiment category and defines the sentiment intensity in, Used to measure the concentration of emotion.

[0020] Preferably, in the step S2, emotion collection and analysis, the method for analyzing the reader's emotional state includes:

[0021] For facial expression analysis, a convolutional neural network model based on deep learning is used. The input image is L. After a series of convolutional layers, pooling layers, and fully connected layers, the probability distribution H of the emotional category of facial expressions is obtained. face (L)=(h f1 ,h f2 ,…,h fn )=\softmax(M fcn x+b fcn ), where x is the output vector of the last fully connected layer, M fcn and b fcn are the corresponding weights and biases, h fn is the probability that the facial expression belongs to the nth emotion category;

[0022] For speech data, speech recognition is performed to obtain text content, and the sentiment vector F of the speech text is calculated using the novel sentiment analysis method. voice , a small RNN model is used to obtain the probability distribution H of speech emotion categories voice (V)=(h v1 ,h v2 ,…,h vn ), where h vn is the probability that the speech belongs to the nth emotion category;

[0023] For heart rate data HR, set the heart rate threshold range to correspond to different emotions. high When the tension weight otherwise Among them, Plow and P high The heart rate thresholds corresponding to calm and nervous emotions are calculated using the same method for the weights of other emotions.

[0024] The probability distribution of readers’ emotion categories is obtained by integrating multimodal data:

[0025] H reader =(h s1 ,h s2 ,…,h sn )=αH face +βH voice +γH HR , where α, β, γ are weight coefficients of different modal data, which can be adjusted according to experiments or user feedback, and α+β+γ=1, h sn The probability that the reader belongs to the nth emotion category; calculate the reader's emotion intensity in And set the time interval for emotion updates.

[0026] Preferably, in the step S3, text analysis and sentiment search, the specific formula for matching search is:

[0027] Define the sentiment matching formula: Where H is the novel paragraph, P is the reader's emotional state, h si is the probability of the reader having the i-th emotion, h i is the probability of paragraph H in the i-th sentiment, L reader For the reader's emotional intensity, L emo (H) is the sentiment intensity of paragraph H.

[0028] Preferably, in the step S4, optimization recommendation, the specific method of sorting the multiple novels includes:

[0029] The comprehensive score is calculated by the formula:

[0030] C=m emo ×W emo +m pref ×H pref , where C represents the comprehensive score of the novel, m emo is the sentiment matching weight, W emo is the value of sentiment matching, m pref is the user preference weight, H pref It is a score related to user preferences, representing the quantitative score of the user's preference factors such as novel types. The comprehensive score of each novel is calculated and the novels are sorted according to the comprehensive score.

[0031] Preferably, the present invention is a system for novel recommendation, the recommendation system comprising a recommendation engine module, the recommendation engine module being bidirectionally connected to an emotion detection module, a novel emotion context analysis module, a user preference and history integration module and a reading interface display module, the input end of the novel emotion context analysis module being connected to a data acquisition module, and the output end of the data acquisition module being unidirectionally connected to the input end of the user preference and history integration module.

[0032] Preferably, the emotion detection module includes a mobile terminal and a sports bracelet, the mobile terminal is connected to the sports bracelet via Bluetooth, the mobile terminal includes a camera and a microphone, and the sports bracelet is used to collect the user's heart rate data.

[0033] Preferably, the data collection module is used to collect data related to novel recommendations from multiple data sources, including user data and novel data; the emotion monitoring module is used to collect multimodal data of readers when reading, and analyze and transform it through an emotion recognition algorithm to obtain a quantifiable reader emotional state to provide a basis for novel recommendations; the novel emotional context analysis module is used to analyze the novel, identify the emotional tendencies contained in different plot paragraphs in the novel, and mark the emotional intensity; the recommendation engine module is used to search for matching novel emotional fragments in the novel emotional context analysis module based on the current emotional state and intensity obtained by the reader's real-time emotion monitoring module; the user preference and historical data integration module is used to analyze the user's past preferences and record the user's selection tendency for different types of novels under different emotional states; the reading interface display module is used to display recommended novel information and record the reader's interaction with the recommended novel.

[0034] The present invention has the following beneficial effects:

[0035] 1. The present invention sets a recommendation engine module, which can search for matching novel emotion fragments in the novel emotion context analysis module according to the current emotion state and intensity obtained by the reader real-time emotion monitoring module. At the same time, the module can also combine the user's comprehensive preference portrait in the user preference and historical data integration module to screen and sort the matched novels, and give priority to recommending novels that are both in line with the current emotion and highly consistent with the user's historical preferences, thereby reducing the time users spend looking for their favorite works and accurately providing users with more favorite works.

[0036] 2. During the reading process of users, the system of the present invention continuously monitors the changes in readers' emotions, and further optimizes the recommendation list according to the dynamic adjustment of emotions, thereby replacing some novels in the original recommendation list that do not quite match the current emotional changes, so as to better meet the user's emotions and preference needs during the reading process. At the same time, the system records the user's feedback on the recommended novels, and feeds these feedback data back to the user preference and historical data integration module in real time, so as to continuously optimize the user's comprehensive preference portrait and recommendation algorithm, and improve the accuracy and quality of subsequent recommendations.

[0037] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0039] Figure 1 A framework diagram of a system for novel recommendation according to the present invention;

[0040] Figure 2 The present invention is a flow chart of a method for novel recommendation. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] See also Figure 2 As shown, the present invention is a method for novel recommendation, and the recommendation method includes:

[0043] Step S1, data collection: collecting data related to novel recommendation from multiple data sources, including user data and novel data;

[0044] Step S1, data collection also includes the following contents:

[0045] Step S11: collecting user data;

[0046] Step S12, novel data collection;

[0047] Step S2, emotional information collection and analysis: When a reader reads a novel or enters a reading platform, the system automatically starts a reader emotional monitoring module to collect multimodal data and analyze the reader's real-time emotional state;

[0048] Step S3, text analysis and emotion search: Analyze the collected novel text data to identify the emotional tendency in the novel. The recommendation engine module performs matching search in the novel emotional context analysis module according to the current emotional state, and performs screening and sorting based on user preferences to generate a preliminary recommendation list.

[0049] Step S4, optimized recommendation: combining user preferences with the preference information of readers in the historical data integration module, sorting multiple novels, and updating and optimizing the user's comprehensive preference portrait;

[0050] Step S5, recommendation display: display the recommendation list on the reader's reading interface or recommendation page;

[0051] Step S6, continuous detection and real-time feedback: During the reading process, the system continuously monitors the reader's emotional changes and dynamically optimizes the recommendation list based on the emotions. When the reader emotion monitoring module detects that Xiao Li's emotions fluctuate slightly, the system re-updates the recommendation list according to the matching search and screening sorting process. At the same time, it records Xiao Li's feedback on the recommended novels and provides real-time feedback to the user preference and historical data integration module.

[0052] In step S3, text analysis and sentiment search, the specific method for identifying the sentiment tendency in the novel includes:

[0053] Construct a sentiment dictionary, where each sentiment word in the dictionary corresponds to a sentiment vector F m =(e1,e2,…,e n ), where n is the number of emotion categories, e i Indicates the weight of the sentiment word on the i-th sentiment;

[0054] For each sentence T in the novel text, segment it into words to get the word sequence and calculate the sentiment vector of the sentence Where l is an index variable used to traverse the vocabulary in the sentence. It starts counting from 1 and increases by 1 each time until it reaches w; F ml Represents the sentiment vector corresponding to the lth word;

[0055] A recurrent neural network is used to process sentence sequences to capture the contextual emotional information of the text. The state update formula of the hidden layer in the RNN is set to p r =tanh(M p p r-1 +M x F Tr +b p ), where Mp and M x is the weight matrix, b p is the bias vector, p r is the hidden layer state at the rth time step, tanh is the hyperbolic tangent function, and the sentiment category probability distribution H of each paragraph H is obtained through a fully connected layer. emo (H)=(h1,h2,…,h n )=\softmax(M fc p R +b fc ), where p R is the hidden layer state of the last sentence in the paragraph, M fc and h fc are the weights and biases of the fully connected layer, the \softmax function is used to convert the output into a probability distribution, h i It represents the probability that paragraph H belongs to the i-th sentiment category and defines the sentiment intensity in, Used to measure the concentration of emotion.

[0056] In step S2, emotion collection and analysis, the method for analyzing the reader's emotional state includes:

[0057] For facial expression analysis, a convolutional neural network model based on deep learning is used. The input image is L. After a series of convolutional layers, pooling layers, and fully connected layers, the probability distribution H of the emotional category of facial expressions is obtained. face (L)=(h f1 ,h f2 ,…,h fn )=\softmax(M fcn x+b fcn ), where x is the output vector of the last fully connected layer, M fcn and b fcn are the corresponding weights and biases, h fn is the probability that the facial expression belongs to the nth emotion category;

[0058] For speech data, speech recognition is performed to obtain text content, and the sentiment vector F of the speech text is calculated using the novel sentiment analysis method. voice , a small RNN model is used to obtain the probability distribution H of speech emotion categories voice (V)=(h v1 ,h v2 ,…,h vn ), where h vn is the probability that the speech belongs to the nth emotion category;

[0059] For heart rate data HR, set the heart rate threshold range to correspond to different emotions.high When the tension weight otherwise Among them, P low and P high The heart rate thresholds corresponding to calm and nervous emotions are calculated using the same method for the weights of other emotions.

[0060] The probability distribution of readers’ emotion categories is obtained by integrating multimodal data:

[0061] H reader =(h s1 ,h s2 ,…,h sn )=αH face +βH voice +γH HR , where α, β, γ are weight coefficients of different modal data, which can be adjusted according to experiments or user feedback, and α+β+γ=1, h sn The probability that the reader belongs to the nth emotion category; calculate the reader's emotion intensity in And set the time interval for emotion updates.

[0062] In step S3, text analysis and sentiment search, the specific formula for matching search is:

[0063] Define the sentiment matching formula: Where H is the novel paragraph, P is the reader's emotional state, h si is the probability of the reader having the i-th emotion, h i is the probability of paragraph H in the i-th sentiment, L reader For the reader's emotional intensity, L emo (H) is the sentiment intensity of paragraph H.

[0064] In step S4, in the optimization recommendation, the specific method for sorting the multiple novels includes:

[0065] The comprehensive score is calculated by the formula:

[0066] C=m emo ×W emo +m pref ×H pref , where C represents the comprehensive score of the novel, m emo is the sentiment matching weight, W emo is the value of sentiment matching, m pref is the user preference weight, H pref It is a score related to user preferences, representing the quantitative score of the user's preference factors such as novel types. The comprehensive score of each novel is calculated and the novels are sorted according to the comprehensive score.

[0067] See also Figure 1 As shown, the present invention is a system for novel recommendation, and the recommendation system includes: a recommendation engine module, the recommendation engine module is bidirectionally connected to an emotion detection module, a novel emotion context analysis module, a user preference and history integration module and a reading interface display module, the input end of the novel emotion context analysis module is connected to a data acquisition module, and the output end of the data acquisition module is unidirectionally connected to the input end of the user preference and history integration module.

[0068] The emotion detection module includes a mobile terminal and a sports bracelet. The mobile terminal is connected to the sports bracelet via Bluetooth. The mobile terminal includes a camera and a microphone. The sports bracelet is used to collect the user's heart rate data.

[0069] The data collection module is used to collect data related to novel recommendations from multiple data sources, including user data and novel data; the emotion monitoring module is used to collect multimodal data of readers when reading, and analyze and transform it through emotion recognition algorithm to obtain quantifiable reader emotional state, providing a basis for novel recommendation; the novel emotional context analysis module is used to analyze the novel, identify the emotional tendencies contained in different plot paragraphs in the novel, and mark the emotional intensity; the recommendation engine module is used to search for matching novel emotional fragments in the novel emotional context analysis module based on the current emotional state and intensity obtained by the reader's real-time emotion monitoring module; the user preference and historical data integration module is used to analyze the user's past preferences and record the user's selection tendency for different types of novels under different emotional states; the reading interface display module is used to display recommended novel information and record the interaction between readers and recommended novels.

[0070] A specific application of this embodiment is:

[0071] Step S1, data collection: the data collection module is responsible for collecting data related to novel recommendation from multiple data sources, including user data and novel data;

[0072] Step S11: User data collection:

[0073] Collect basic information about users, such as age, gender, and region. This information helps to gain a preliminary understanding of the potential differences in users' reading preferences.

[0074] Collect users' reading history records, including detailed records of the titles of novels that users have read, reading time, reading progress, whether they have finished reading, whether they have rated and commented on them, and other behavioral data. By analyzing reading time, we can determine the user's level of involvement in a novel. Reading progress and whether they have finished reading can reflect the novel's appeal to users. Ratings and comments directly reflect users' preferences and opinions on the novel.

[0075] Collect other user interaction data on the platform, such as search keywords, dwell time and click behavior when browsing novel lists, etc. The keywords searched by users can reveal the topics or novel types they are currently interested in, and the dwell time and click behavior when browsing can reflect the user's initial interest in different novels;

[0076] Step S12: novel data collection:

[0077] Collect basic information about the novel, such as title, author, year of publication, word count, novel type tags (such as fantasy, romance, martial arts, science fiction, etc.), theme tags (such as revenge, growth, time travel, etc.), style tags (such as humorous, serious, profound, poignant, etc.), etc. These tags help to classify and describe the characteristics of the novel, so as to facilitate matching with user preferences;

[0078] Collect the text content data of the novel for subsequent analysis of the emotional context of the novel, store and organize the chapter content, paragraphs, sentences and other text information of the novel, so as to analyze the expression of emotion, emotional change trend, emotional intensity, etc. in the novel;

[0079] Step S2, emotion collection and analysis: When the reader starts to read a novel or enters the reading platform, the system automatically starts the reader emotion monitoring module, starts to collect multimodal data and analyzes the reader's real-time emotional state. For example, reader Xiao Li opens the reading application and starts to read a novel. The system captures Xiao Li's facial expression through the front camera of the mobile phone. The microphone does not detect obvious emotional voice tone changes. The heart rate monitoring data shows that the heart rate is stable. Combining these multimodal data, the reader emotion monitoring module analyzes that Xiao Li is currently in a calm emotional state. The probability distribution of emotion categories may be [calm: 0.8, joy: 0.1, sadness: 0.05, anger: 0.05], and the emotion intensity is relatively low, set to 0.2;

[0080] Among them, the reader emotion monitoring module: connects with sensors of wearable devices or smart terminals (such as cameras, microphones, heart rate monitors, etc.) to obtain multimodal data of readers during the reading process. The camera can capture the reader's facial expressions, the microphone collects voice and tone information, and the heart rate monitor obtains changes in physiological signals.

[0081] Computer vision technology is used to analyze facial expression features, speech recognition and emotion analysis technology is used to process speech data, and physiological signal analysis algorithms are used to interpret heart rate and other data to comprehensively judge the reader's current emotional state and emotional intensity;

[0082] For facial expression analysis, a convolutional neural network (CNN) model based on deep learning is used. The input image is L. After a series of convolutional layers, pooling layers, and fully connected layers, the emotional category probability distribution H of facial expressions is obtained.face (L)=(h f1 ,h f2 ,…,h fn )=\softmax(M fcn x+b fcn ), where x is the output vector of the last fully connected layer, M fcn and b fcn are the corresponding weights and biases;

[0083] For speech data, we first perform speech recognition to obtain the text content, and then calculate the sentiment vector F of the speech text similar to sentiment analysis of novels. voice , and obtain the speech emotion category probability distribution H through a small RNN model voice (V)=(h v1 ,h v2 ,…,h vn );

[0084] For the heart rate data HR, the heart rate threshold range is set to correspond to different emotions. For example, when HR>P high When the tension weight otherwise (P low and P high is the heart rate threshold corresponding to calm and nervous emotions), and the weights of other emotions are calculated similarly

[0085] The final probability distribution of the reader's emotion category is obtained by integrating multimodal data:

[0086] H reader =(h s1 ,h s2 ,…,h sn )=αH face +βH voice +γH HR , where α, β, γ are weight coefficients of different modal data, which can be adjusted according to experiments or user feedback, and α+β+γ=1; at the same time, the reader's emotional intensity is calculated in Set the time interval for sentiment updates, for example, update readers' real-time sentiment data every 5 minutes;

[0087] Step S3, text analysis and emotion search: Analyze the collected novel text data to identify the emotional tendencies contained in different plot paragraphs in the novel. The recommendation engine module performs matching search in the novel emotional context analysis module according to the current emotional state, and performs screening and sorting in combination with user preferences to generate a preliminary recommendation list. Assume that there is a novel "The Story of a Quiet Town" in the novel emotional context analysis module, and the emotional context analysis result of its opening chapter is the emotional category probability distribution [Calm: 0.7, Joy: 0.2, Sadness: 0.05, Angry: 0.05], and the emotional intensity is 0.15. The emotional matching degree is calculated according to the formula:

[0088] The current emotion matching degree of Xiao Li can be calculated as:

[0089]

[0090] Among them, the novel sentiment analysis module uses natural language processing technology to conduct sentiment analysis on a large amount of novel text data sentence by sentence and paragraph by paragraph. It uses sentiment dictionaries combined with machine learning algorithms (such as recurrent neural networks in deep learning) to identify the emotional tendencies contained in different plot paragraphs in the novel, such as joy, sadness, anger, calmness, tension, etc., and mark the intensity of the emotions.

[0091] First, we need to build a sentiment dictionary. Each sentiment word in the dictionary corresponds to a sentiment vector F. m =(e1,e2,…,e n ), where n is the number of emotion categories (e.g. n = 5, representing joy, sadness, anger, calmness, and tension, respectively), e i Indicates the weight of the sentiment word on the i-th sentiment (value range 0-1);

[0092] For each sentence T in the novel text, segment it into words to get the word sequence and calculate the sentiment vector of the sentence (l is an index variable used to traverse the vocabulary in the sentence. It starts counting from 1 and increases by 1 each time until it reaches w; F ml represents the sentiment vector corresponding to the lth word);

[0093] A recurrent neural network (RNN) is used to process sentence sequences to capture the contextual emotional information of the text. The state update formula of the hidden layer in the RNN is set to p r =tanh(M p p r-1 +M x F Tr +b p ), where M p and M x is the weight matrix, b pis the bias vector, p r is the hidden layer state of the rth time step (sentence), tanh is the hyperbolic tangent function, and finally the sentiment category probability distribution H of each paragraph H is obtained through a fully connected layer. emo (H)=(h1,h2,…,h n )=\softmax(M fc p R +b fc ), where p R is the hidden layer state of the last sentence in the paragraph, M fc and h fc are the weights and biases of the fully connected layer, the \softmax function is used to convert the output into a probability distribution, h i It represents the probability that paragraph H belongs to the i-th sentiment category and defines the sentiment intensity in, Used to measure the concentration of emotions;

[0094] Divide each novel into chapters or certain text fragments, construct a novel emotional context map, and record the curve of emotional type and intensity as the text progresses;

[0095] Recommendation engine module: Based on the current emotional state and intensity obtained by the real-time emotional monitoring module, the novel emotional context analysis module searches for matching novel emotional fragments. The matching basis can be the consistency of emotional type and the similarity of emotional intensity.

[0096] Define the sentiment matching formula: Where H is the novel paragraph, P is the reader's emotional state, h si is the probability of the reader having the i-th emotion, h i is the probability of paragraph H in the i-th sentiment, L reader For the reader's emotional intensity, L emo (H) is the sentiment intensity of paragraph H. This formula takes into account both the matching of sentiment types (h si h i Item), and the difference in emotional intensity is also taken into account. The smaller the difference in intensity, the higher the match degree;

[0097] Combine the user preference and the user's comprehensive preference portrait in the historical data integration module to filter and sort the matched novels, and give priority to recommending novels that match the current mood and are highly consistent with the user's historical preferences. For example, if the user has historically preferred ancient style romance novels and is currently in a slightly sad mood, then the recommended ancient style romance novels with similar emotions will be ranked first;

[0098] Step S4, optimized recommendation: The recommendation engine module combines the user preference and the preference information of Xiao Li in the historical data integration module, and finds that Xiao Li used to like to read rural life novels, and "The Story of a Quiet Town" happens to belong to this category. In the screening process, assuming that the emotional matching weight is set to 0.6 and the user preference weight is set to 0.4, through C = m emo ×W emo +m pref ×H pref The comprehensive score of the novel is calculated as 0.61×0.6+0.4×0.8 (assuming that the weight of the pastoral life category in Xiao Li’s preference is 0.8)≈0.69, where C represents the comprehensive score of the novel, m emo is the sentiment matching weight, W emo is the value of sentiment matching, m pref is the user preference weight, H pref It is a score related to user preferences, representing the quantitative score of the user's preference factors such as novel types. After sorting multiple novels through similar calculations, when generating a preliminary recommendation list, "The Story of a Quiet Town" is ranked at the top;

[0099] The user preference and historical data integration module is used to analyze the conventional preference factors such as the novel types, themes, and authors that users have preferred in the past, and to record the user's preference tendency for different types of novels under different emotional states. For example, when a user is in a sad mood, is he more inclined to read healing novels? These emotion-related preference data are integrated and stored with the conventional preference data to form a comprehensive preference portrait of the user.

[0100] Step S5, recommendation display: the recommendation list is displayed on the reader's reading interface or a special recommendation page through the reading interface display module, and the recommendation information includes the title of the novel "The Story of a Quiet Town", the introduction "tells a heartwarming story that happened in a quiet town, full of the atmosphere of pastoral life...", the cover picture (showing the pastoral scenery of the town) and a brief description of the reason that matches the reader's current mood "the emotional tone of the opening of this novel is calm and peaceful, which fits your current calm mood and meets your previous preference for pastoral life novels". At the same time, some key evaluation indicators of the recommended novel are displayed, such as the average reader rating of 4.5 stars (out of 5 stars), the praise rate of 80%, etc., and a personalized recommendation label "the choice of tranquility of the soul" is also provided;

[0101] The reading interface display module is used to receive the recommendation list from the recommendation engine module, display the recommended novel information (title, introduction, cover image, matching reason, evaluation index, personalized label), record the reader's interaction with the recommended novel (click to view details, start reading, etc.), and transmit the reader's interactive behavior data to the user preference and historical data integration module;

[0102] Step S6, continuous detection, real-time feedback: Readers can choose to view the details of the recommended novels and start reading. During the reading process, the system continuously monitors the reader's emotional changes and further optimizes the recommendation list based on the dynamic adjustment of emotions. For example, after Xiao Li has read for 10 minutes, the novel has some minor twists and turns. The reader emotion monitoring module detects that Xiao Li's emotions fluctuate slightly, and the probability distribution of emotion categories becomes [calm: 0.6, joy: 0.1, sadness: 0.2, anger: 0.1], and the emotion intensity rises to 0.3. The system re-updates the recommendation list according to the above matching search and screening sorting process, and may recommend some novels with certain emotional fluctuations but still strong. Novels within Xiao Li's preference range, such as "A Small Town Story: Wind and Rain Changes", have plots that are calm but with some emotional changes caused by twists and turns, which better match Xiao Li's emotional state at this time, thereby replacing some novels in the original recommendation list that do not quite match the current emotional changes, in order to better meet Xiao Li's emotional and preference needs during the reading process. At the same time, Xiao Li's feedback on the recommended novels is recorded, such as whether he clicks to view details, whether he starts reading, how long he reads, etc. These feedback data are fed back to the user preference and historical data integration module in real time, so as to continuously optimize the user's comprehensive preference portrait and recommendation algorithm, and improve the accuracy and quality of subsequent recommendations.

[0103] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0104] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for novel recommendation, characterized in that: The recommended methods include: Step S1, data collection: collecting data related to novel recommendation from multiple data sources, including user data and novel data; The step S1, data collection, also includes the following contents: Step S11: collecting user data; Step S12, novel data collection; Step S2, emotional information collection and analysis: When a reader reads a novel or enters a reading platform, the system automatically starts a reader emotional monitoring module to collect multimodal data and analyze the reader's real-time emotional state; Step S3, text analysis and emotion search: Analyze the collected novel text data to identify the emotional tendency in the novel. The recommendation engine module performs matching search in the novel emotional context analysis module according to the current emotional state, and performs screening and sorting based on user preferences to generate a preliminary recommendation list. Step S4, optimized recommendation: combining user preferences with the preference information of readers in the historical data integration module, sorting multiple novels, and updating and optimizing the user's comprehensive preference portrait; Step S5, recommendation display: display the recommendation list on the reader's reading interface or recommendation page; Step S6, continuous detection and real-time feedback: During the reading process, the system continuously monitors the reader's emotional changes and dynamically optimizes the recommendation list based on the emotions. When the reader emotion monitoring module detects that Xiao Li's emotions fluctuate slightly, the system re-updates the recommendation list according to the matching search and screening sorting process. At the same time, it records Xiao Li's feedback on the recommended novels and provides real-time feedback to the user preference and historical data integration module.

2. A method for novel recommendation according to claim 1, characterized in that: In the step S3, text analysis and sentiment search, the specific method for identifying the sentiment tendency in the novel includes: Construct a sentiment dictionary, where each sentiment word in the dictionary corresponds to a sentiment vector F m =(e1,e2,…,e n ), where n is the number of emotion categories, e i Indicates the weight of the sentiment word on the i-th sentiment; For each sentence T in the novel text, segment it into words to get the word sequence and calculate the sentiment vector of the sentence Where l is an index variable used to traverse the vocabulary in the sentence. It starts counting from 1 and increases by 1 each time until it reaches w; F ml Represents the sentiment vector corresponding to the lth word; A recurrent neural network is used to process sentence sequences to capture the contextual emotional information of the text. The state update formula of the hidden layer in the RNN is set to p r =tanh(M p p r-1 +M x F Tr +b p ), where M p and M x is the weight matrix, b p is the bias vector, p r is the hidden layer state at the rth time step, tanh is the hyperbolic tangent function, and the sentiment category probability distribution H of each paragraph H is obtained through a fully connected layer. emo (H)=(h1,h2,…,h n )=\softmax(M fc p R +b fc ), where p R is the hidden layer state of the last sentence in the paragraph, M fc and h fc are the weights and biases of the fully connected layer, the \softmax function is used to convert the output into a probability distribution, h i It represents the probability that paragraph H belongs to the i-th sentiment category and defines the sentiment intensity in, Used to measure the concentration of emotion.

3. A method for novel recommendation according to claim 1, characterized in that: In the step S2, emotion collection and analysis, the method for analyzing the reader's emotional state includes: For facial expression analysis, a convolutional neural network model based on deep learning is used. The input image is L. After a series of convolutional layers, pooling layers, and fully connected layers, the probability distribution H of the emotional category of facial expressions is obtained. face (L)=(h f1 ,h f2 ,…,h fn )=\softmax(M fcn x+b fcn ), where x is the output vector of the last fully connected layer, M fcn and b fcn are the corresponding weights and biases, h fn is the probability that the facial expression belongs to the nth emotion category; For speech data, speech recognition is performed to obtain text content, and the sentiment vector F of the speech text is calculated using the novel sentiment analysis method. voice , a small RNN model is used to obtain the probability distribution H of speech emotion categories voice (V)=(h v1 ,h v2 ,…,h vn ), where h vn is the probability that the speech belongs to the nth emotion category; For heart rate data HR, set the heart rate threshold range to correspond to different emotions. high When the tension weight otherwise Among them, P low and P high The heart rate thresholds corresponding to calm and nervous emotions are calculated using the same method for the weights of other emotions. The probability distribution of readers’ emotion categories is obtained by integrating multimodal data: H reader =(h s1 ,h s2 ,…,h sn )=αH face +βH voice +γH HR , where α, β, γ are weight coefficients of different modal data, which can be adjusted according to experiments or user feedback, and α+β+γ=1, h sn The probability that the reader belongs to the nth emotion category; calculate the reader's emotion intensity in And set the time interval for emotion updating.

4. A method for novel recommendation according to claim 1, characterized in that: In the step S3, text analysis and sentiment search, the specific formula for matching search is: Define the sentiment matching formula: Where H is the novel paragraph, P is the reader's emotional state, h si is the probability of the reader having the i-th emotion, h i is the probability of paragraph H in the i-th sentiment, L reader For the reader's emotional intensity, L emo (H) is the sentiment intensity of paragraph H.

5. The method for novel recommendation according to claim 1, characterized in that: In the step S4, optimization recommendation, the specific method for sorting the multiple novels includes: The comprehensive score is calculated by the formula: C=m emo ×W emo +m pref ×H pref , where C represents the comprehensive score of the novel, m emo is the sentiment matching weight, W emo is the value of sentiment matching, m pref is the user preference weight, H pref It is a score related to user preferences, representing the quantitative score of the user's preference factors such as novel types. The comprehensive score of each novel is calculated and the novels are sorted according to the comprehensive score.

6. A system for novel recommendation, the recommendation system comprising a recommendation engine module, characterized in that: The recommendation engine module is bidirectionally connected to the emotion detection module, the novel emotion context analysis module, the user preference and history integration module and the reading interface display module; the input end of the novel emotion context analysis module is connected to the data acquisition module; the output end of the data acquisition module is unidirectionally connected to the input end of the user preference and history integration module.

7. A system for novel recommendation according to claim 6, characterized in that: The emotion detection module includes a mobile terminal and a sports bracelet. The mobile terminal is connected to the sports bracelet via Bluetooth. The mobile terminal includes a camera and a microphone. The sports bracelet is used to collect the user's heart rate data.

8. A system for novel recommendation according to claim 6, characterized in that: The data collection module is used to collect data related to novel recommendations from multiple data sources, including user data and novel data. The emotion monitoring module is used to collect multimodal data of readers when reading, and analyze and transform it through emotion recognition algorithm to obtain quantifiable reader emotional states to provide a basis for novel recommendations. The novel emotional context analysis module is used to analyze novels, identify emotional tendencies contained in different plot paragraphs in the novel, and mark emotional intensities. The recommendation engine module is used to search for matching novel emotional fragments in the novel emotional context analysis module based on the current emotional state and intensity obtained by the reader's real-time emotion monitoring module. The user preference and historical data integration module is used to analyze the user's past preferences and record the user's selection tendencies for different types of novels under different emotional states. The reading interface display module is used to display recommended novel information and record the interaction between readers and recommended novels.

Images