Audio book listening recommendation method and platform based on big data intelligent recommendation

Abstract

The invention relates to an audio book listening recommendation method and platform based on big data intelligent recommendation, and the method comprises the steps: reading the historical book listening behavior data of a user through a book listening system background, and inputting the historical book listening behavior data into a pre-deployed audio book listening recommendation model; behavior characteristics in the historical book listening behavior data are recognized through an audio book listening recommendation model, and related historical book listening behavior characteristics are output; retrieving and generating a corresponding audio book listening list based on the historical book listening behavior characteristics, and recommending the audio book listening list to a front-end page; and activating and playing the selected audio book according to the selection information of the user on the audio book in the audio book listening list fed back by the front end. In combination with historical behavior characteristics (duration, interested interaction characteristics, channels, plots and the like) of the user who listens to the book, the interested audio book of the user is predicted and recommended for the user to select, so that the user can listen to the book taste, individualized and accurate book listening recommendation is flexibly performed, and the book listening service quality is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to an audio book recommendation method, platform and electronic device based on big data intelligent recommendation. Background Art

[0002] Audiobook listening mode refers to a reading mode in which users receive book content in an auditory manner through audio playback devices such as smart phones and smart speakers. Existing methods for recommending audiobook information to users mainly rely on the following technologies:

[0003] First, the platform will analyze the user's interests and preferences based on the user's browsing history and purchase history, and then recommend relevant books.

[0004] Secondly, through deep learning algorithms, the platform can accurately capture the user's interests and provide personalized content recommendations. For example, based on the user's listening history, ratings, favorites and other behavioral data, a user profile is constructed, so that when the user visits, books that meet their interests and needs can be recommended. In addition, some platforms will also recommend popular and high-quality books based on the popularity of books, user reviews and other information.

[0005] However, these technologies also have some shortcomings in application. First, the limitations of the recommendation algorithm may cause the types of books that users are exposed to to be too single, which fails to broaden the user's reading horizons. Second, over-reliance on the recommendation algorithm may reduce the user's ability to make independent choices, making the user lack the fun of active exploration in the process of listening to books (especially user portraits, which can only recommend fixed books to users). Third, for some specific types of audiobooks, such as professional books and niche books, since the user group is relatively small, the recommendation algorithm may not be able to accurately capture their needs, resulting in poor promotion effects for these books.

[0006] In addition, in the actual application of the audiobook mode, when recommending audiobooks, existing audiobook platforms can basically only consider the user's behavioral interest tags to make relevant popular audiobook recommendations, but cannot recommend personalized audiobook modes, quality, and listening environments to users based on the user's customized selection information for listening to books, and cannot provide users with better audiobook services. Summary of the invention

[0007] In order to solve the above problems, the present application proposes an audio book recommendation method, platform and electronic device based on big data intelligent recommendation.

[0008] On the one hand, the present application proposes a method for recommending audiobooks based on big data intelligent recommendation, comprising the following steps:

[0009] S1. The user logs in to the audiobook system backend;

[0010] S2, the audiobook system background reads the user's historical audiobook behavior data;

[0011] S3, inputting the historical listening behavior data into a pre-deployed audiobook recommendation model, identifying behavior features in the historical listening behavior data through the audiobook recommendation model and outputting relevant historical listening behavior features;

[0012] S4, retrieving and generating a corresponding audiobook listening list based on the historical listening behavior characteristics, and recommending the audiobook listening list to the front-end page;

[0013] S5. Activate and play the selected audio book in the audio book listening list according to the user's selection information fed back by the front end.

[0014] Preferably, the method for generating the audio book recommendation model comprises:

[0015] Data Collection

[0016] Collecting a number of historical audiobook listening behavior data of the user, wherein the audiobook listening behavior data includes interactive behavior data participated in during the audiobook listening process, the audiobook listening channel, the catalog summary of the audiobook listened to, and the audiobook audio type selected;

[0017] Feature Engineering

[0018] Analyzing and extracting historical listening behavior features from the listening behavior data, including: analyzing and extracting interactive features of the user's interest in the audiobook being listened to from the interactive behavior data, identifying and extracting corresponding channel keywords from the audiobook channel, analyzing and extracting background features or plot features of the audiobook being listened to from the catalog summary, identifying and extracting corresponding audio keywords from the audio type of the audiobook; collecting the historical listening behavior features and forming a feature set;

[0019] Model training and validation

[0020] Dividing the feature set into a training set and a validation set according to a preset ratio;

[0021] Input the training set into a preset random forest model to perform feature training and learning to generate an audiobook recommendation model;

[0022] The recommendation performance of the audiobook recommendation model is verified using the verification set:

[0023] If the verification is successful, the audiobook recommendation model is deployed on the audiobook system backend;

[0024] If the verification fails, the above steps are repeated to retrain and generate the audio book recommendation model.

[0025] Preferably, it also includes:

[0026] Feature Enhancements:

[0027] Inputting the feature set into a preset LLM large language model;

[0028] The feature set is recombined by the LLM large language model, and the background features or plot features in the feature set and their arrangement and combination with other features are constructed and generated to construct and generate the enhanced feature set: S:

[0029] S={S1, S2, S3, S4, S5, S6, S7, S8}

[0030] in,

[0031] S1 = {background features or plot features};

[0032] S2 = {background features or plot features, interactive features};

[0033] S3 = {background features or plot features, channel keywords};

[0034] S4 = {background features or plot features, audio keywords};

[0035] S5 = {background features or plot features, interactive features, channel keywords};

[0036] S6 = {background features or plot features, interactive features, audio keywords};

[0037] S7 = {background features or plot features, channel keywords, audio keywords};

[0038] S8 = {background features or plot features, interactive features, channel keywords, audio keywords}.

[0039] Preferably, the interactive feature includes a behavior preference score P:

[0040] P=(w C *C+w T *T+w F *F+w R *R+w M *M)e -λt ,

[0041] The meanings of each behavior variable are as follows:

[0042] C: number of clicks;

[0043] T: listening time;

[0044] F: number of collections;

[0045] R: Rating (between 1 to 5 or 1 to 10);

[0046] M: number of comments;

[0047] w C , w T , w F , w R , w M are the weight coefficients of the corresponding behavioral variables, which are set by system default;

[0048] e -λt is an exponential decay function, where t is the time difference between the user's listening behavior and the last time they listened to the book, and λ is the decay rate (with a value of 1-5).

[0049] Preferably, in step S2, when the audiobook system background reads the user's historical audiobook behavior data, it also includes:

[0050] Read the time when the user last listened to a book, and calculate the time difference from the last time the user listened to a book.

[0051] On the other hand, the present application proposes a platform for implementing the above-mentioned audio book recommendation method based on big data intelligent recommendation, comprising:

[0052] User management module, used to manage audiobook users;

[0053] Log module, used to record the user's historical listening behavior data;

[0054] The audiobook intelligent matching module is used to read the user's historical audiobook behavior data; input the historical audiobook behavior data into a pre-deployed audiobook recommendation model, identify the behavior features in the historical audiobook behavior data through the audiobook recommendation model, and output the relevant historical audiobook behavior features;

[0055] An audiobook recommendation module, for retrieving and generating a corresponding audiobook list based on the historical audiobook behavior characteristics, and recommending the audiobook list to the front-end page;

[0056] The audiobook execution module is used to activate and play the selected audiobook in the audiobook listening list according to the user's selection information of the audiobook in the audiobook listening list fed back by the front end, and the audiobook listening recommendation method based on big data intelligent recommendation.

[0057] In another aspect, the present application further provides an electronic device, comprising:

[0058] processor;

[0059] a memory for storing processor-executable instructions;

[0060] Among them, the processor is configured to implement the audio book recommendation method based on big data intelligent recommendation when executing the executable instructions.

[0061] Technical effects of the present invention:

[0062] The present invention reads the user's historical listening behavior data through the audiobook system background and inputs it into the pre-deployed audiobook recommendation model, identifies the behavior features in the historical listening behavior data through the audiobook recommendation model and outputs the relevant historical listening behavior features; retrieves and generates a corresponding audiobook list based on the historical listening behavior features, and recommends the audiobook list to the front-end page; activates and plays the selected audiobook according to the user's selection information on the audiobook in the audiobook list fed back by the front-end. It can combine the historical behavior features of the audiobook user (duration, interactive features of interest, channel, plot, etc.) to predict and recommend the audiobooks that the user is interested in for the user to choose, so as to meet the user's listening taste, flexibly perform individualized and accurate audiobook recommendations, and improve the quality of audiobook services.

[0063] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0065] Figure 1 It is shown as a schematic diagram of the implementation process of the present invention;

[0066] Figure 2 Shown is a schematic diagram of the model generation process of the present invention;

[0067] Figure 3 It is a schematic diagram of the platform architecture structure of the present invention;

[0068] Figure 4 The application diagram of the electronic device of the present invention is shown as follows DETAILED DESCRIPTION

[0069] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0070] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0071] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the main purpose of the present disclosure.

[0072] Example 1

[0073] The audiobook platform architecture adopted by the present invention can be understood and implemented in combination with the existing audiobook APP software, and the user can log in to the platform through the APP to listen to the book.

[0074] The present invention deploys an AI prediction model for listening to books on the listening platform (the application background of the APP listening software), which can make accurate recommendations based on the user's historical listening habits.

[0075] like Figure 1 As shown, on the one hand, the present application proposes a method for recommending audiobooks based on big data intelligent recommendation, comprising the following steps:

[0076] S1. The user logs in to the audiobook system backend;

[0077] S2, the audiobook system background reads the user's historical audiobook behavior data;

[0078] S3, inputting the historical listening behavior data into a pre-deployed audiobook recommendation model, identifying behavior features in the historical listening behavior data through the audiobook recommendation model and outputting relevant historical listening behavior features;

[0079] S4, retrieving and generating a corresponding audiobook listening list based on the historical listening behavior characteristics, and recommending the audiobook listening list to the front-end page;

[0080] S5. Activate and play the selected audio book in the audio book listening list according to the user's selection information fed back by the front end.

[0081] The present invention reads the user's historical listening behavior data through the audiobook system background and inputs it into the pre-deployed audiobook recommendation model, identifies the behavior features in the historical listening behavior data through the audiobook recommendation model and outputs the relevant historical listening behavior features; retrieves and generates a corresponding audiobook list based on the historical listening behavior features, and recommends the audiobook list to the front-end page; activates and plays the selected audiobook according to the user's selection information on the audiobook in the audiobook list fed back by the front-end. It can combine the historical behavior features of the audiobook user (duration, interactive features of interest, channel, plot, etc.) to predict and recommend the audiobooks that the user is interested in for the user to choose, so as to meet the user's listening taste, flexibly perform individualized and accurate audiobook recommendations, and improve the quality of audiobook services.

[0082] Combined with Figure 3 As shown, the present invention deploys an audiobook recommendation model (referred to as model or AI model) on the audiobook system to recommend the audiobook for the user and play it according to the user's choice. The following will describe the construction of the AI ​​model and its application principle.

[0083] The AI ​​model is built using the random forest model, so it can be understood and implemented in combination with the architecture and application principles of the random forest model. The structure of the random forest model and its training and application steps can be described as follows:

[0084] 1. Structure of the Random Forest Model

[0085] Random forest is an ensemble learning method consisting of many decision trees. Each tree is trained with a different subset of data and features to increase the diversity and accuracy of the model. Each tree in a random forest randomly samples a subset of the training data with replacement, known as bagging. Not only are the trees trained on different datasets, they also use different features to make decisions.

[0086] 2. Training steps of random forest model

[0087] 1. Data preparation: Prepare the dataset for training, including features and labels. Features are variables used for prediction, while labels are the results you want the model to predict. Make sure the dataset is cleaned and preprocessed, including handling missing values, outliers, and standardizing features.

[0088] 2. Dataset division: Divide the prepared dataset into training set and test set. Usually, cross-validation or holdout method is used to divide the dataset into training set and test set to ensure that the model is validated on unseen data.

[0089] 3. Model selection: Select Random Forest as the model. Random Forest makes predictions by building multiple decision trees, and then combines the results of multiple decision trees to get the final prediction result.

[0090] 4. Model training: Use the training set to train the random forest model. During the training process, the random forest will randomly select features and samples to build multiple decision trees, and use voting or averaging to combine the results of multiple decision trees.

[0091] 5. Model tuning: Tune the trained model, including adjusting hyperparameters, optimizing model structure, etc., to improve the performance and generalization ability of the model.

[0092] 3. Application steps of random forest model

[0093] 1. Model evaluation: Use the test set to evaluate the trained model. Usually, indicators such as accuracy, precision, recall, F1 value, etc. are used to evaluate the performance of the model.

[0094] 2. Model application: When the model passes the evaluation and meets the requirements, it can be used for actual prediction. In practical applications, attention should be paid to the deployment and monitoring of the model to ensure the continuous performance of the model.

[0095] 3. Feature importance evaluation: The random forest model also provides a feature importance evaluation function, which can evaluate the influence of each feature on the model prediction results, which is very valuable for understanding the data and the model's prediction results.

[0096] In summary, the random forest model improves prediction accuracy and generalization ability by integrating multiple decision trees, and has a wide range of uses in practical applications.

[0097] Combination Figure 2 As shown, preferably, the method for generating the audio book recommendation model includes:

[0098] Data Collection

[0099] Collecting a number of historical audiobook listening behavior data of the user, wherein the audiobook listening behavior data includes interactive behavior data participated in during the audiobook listening process, the audiobook listening channel, the catalog summary of the audiobook listened to, and the audiobook audio type selected;

[0100] Feature Engineering

[0101] Analyzing and extracting historical listening behavior features from the listening behavior data, including: analyzing and extracting interactive features of the user's interest in the audiobook being listened to from the interactive behavior data, identifying and extracting corresponding channel keywords from the audiobook channel, analyzing and extracting background features or plot features of the audiobook being listened to from the catalog summary, identifying and extracting corresponding audio keywords from the audio type of the audiobook; collecting the historical listening behavior features and forming a feature set;

[0102] Model training and validation

[0103] Dividing the feature set into a training set and a validation set according to a preset ratio;

[0104] Input the training set into a preset random forest model to perform feature training and learning to generate an audiobook recommendation model;

[0105] The recommendation performance of the audiobook recommendation model is verified using the verification set:

[0106] If the verification is successful, the audiobook recommendation model is deployed on the audiobook system backend;

[0107] If the verification fails, the above steps are repeated to retrain and generate the audio book recommendation model.

[0108] The present invention extracts the behavioral features that affect the user's choice of audiobooks by performing feature analysis on the user's historical audiobook behavior data, including the interactive behavior data in which the user participates in the audiobook listening process, such as collection, rating, etc., as well as extracting the audiobook channel features that the user is interested in, as well as the plot features or background features in the historical audiobook catalog summary, and the audiobook audio types that the user likes during the audiobook listening process. These behavioral features or content features of interest are analyzed and extracted to collect the user's historical audiobook behavior features, and a feature set is constructed for model training.

[0109] The collection and preprocessing of listening behavior data includes the following steps:

[0110] Collect users' listening behavior data, such as clicks, listening time, collections, ratings, comments, etc.

[0111] Perform preprocessing operations such as cleaning, denoising, and normalization on the data to ensure data quality.

[0112] Build user portraits:

[0113] Use machine learning algorithms, such as convolutional neural networks (CNN) and long short-term memory networks (LSTM) in deep learning, to analyze users' listening history and behavior patterns and build user portraits.

[0114] User portraits should include information such as user preferences, changes in interests, and listening habits.

[0115] Combined with intelligent speech synthesis technologies such as WaveNet and Tacotron, text is converted into natural and fluent speech, and the emotion and intonation of the speech are adjusted to enhance the user's listening experience.

[0116] Environmental Adaptation Reading:

[0117] Utilizing intelligent environmental perception technology, the system identifies the user's environment through audio analysis and adjusts the volume and clarity of the audio output to ensure that users can enjoy a barrier-free listening experience in any environment.

[0118] In the listening backend, several user listening data can be collected from the log database or the user historical listening behavior big data recorded in the log recording module and feature engineering can be performed, and the historical listening behavior big data can be further collected and cleaned to obtain training data. In this method, a large language model can also be used to input corresponding features and prompt words of data retrieval logic, so that the large language model can retrieve the corresponding historical listening behavior data from the log library based on the prompt words and perform preprocessing. For example, the backend can request a third party to call the large language model to automatically collect the corresponding listening and speaking behavior data and perform preprocessing, and the corresponding keywords and processing logic can be constructed by the administrator.

[0119] Feature engineering can be performed by a large language model or by using a text feature extraction algorithm. Please refer to the following method.

[0120] In order to analyze and extract the historical listening behavior features in the listening behavior data, the following steps can be performed based on the LLM large language model:

[0121] First, you need to combine feature keywords and feature extraction logic to form prompt words and input them into the LLM large language model. The feature keywords and feature extraction logic can be referred to as follows:

[0122] 1. Interactive behavior feature extraction

[0123] Keywords: interactive behavior data, interest

[0124] Analyze interaction data: Carefully review user interaction behavior data, including likes, comments, shares, favorites, repeat listening, and skipping.

[0125] Extract interactive features: Based on the above interactive data, identify which audiobook content users show strong interest in. For example, frequent likes, multiple comments, or repeated listening to a certain paragraph may indicate that the user is particularly interested in this part of the content.

[0126] Extraction logic: For example, "whether the comment involves the word "martial arts", "the keywords with the most likes in the comment", "the attributes of the protagonist mentioned in multiple comments by users", etc.

[0127] 2. Extract keywords from audiobook channels

[0128] Keywords: audiobook channel, channel keywords

[0129] Identify audiobook channels: Sort out the audiobook channels that users have visited and understand the user's preferred areas, such as science fiction, history, literature, etc.

[0130] Extract channel keywords: For each channel, extract the most representative keywords. These keywords should accurately reflect the theme and content characteristics of the channel.

[0131] 3. Catalog summary feature extraction

[0132] Keywords: Catalog summary, background characteristics, storyline characteristics

[0133] Analyze the table of contents and abstract: Study the table of contents and abstract of the audiobook in depth to understand the general content, structure and key plot of the book.

[0134] Extract background and plot features: Extract key information describing the book’s background, time setting, character relationships, and plot development from the catalog summary as important features reflecting the book’s characteristics.

[0135] 4. Extract keywords from audio book types

[0136] Keywords: audio type, audio keywords

[0137] Identify audio types: Classify audiobooks into categories such as narration, dialogue, sound effects, etc., to understand the audio presentation method preferred by users.

[0138] Extract audio keywords: For each audio type, summarize the keywords that describe its characteristics, such as "affectionate narration", "vivid dialogue", etc., to reflect the characteristics and style of the audio.

[0139] 5. Build a feature set

[0140] Collect features: Summarize all the features extracted in the above steps.

[0141] Forming a feature set: According to the category and nature of the features, they are organized in order to form a feature set that fully reflects the characteristics of the user's listening behavior. This feature set will serve as the basis for subsequent analysis, modeling, and recommendation.

[0142] The above extraction logic can be constructed by the administrator.

[0143] Through the above steps, we can systematically analyze and extract the historical listening behavior characteristics in the listening behavior data, providing strong support for building a more accurate and personalized audio book recommendation system.

[0144] You can also use feature analysis algorithms (such as SI FT feature detection and matching algorithms) to perform feature analysis and extraction. To use feature analysis algorithms to analyze and extract historical listening behavior features from the listening behavior data, you can follow the steps below:

[0145] First, it is necessary to analyze and extract the interactive features of users’ interest in the audiobooks they listen to from the interactive behavior data.

[0146] This can be achieved by analyzing the user's interactive behaviors such as likes, comments, shares, repeat listening, and skipping. For example, if a user frequently likes and comments on a certain type of audiobook, it can be considered that the user is more interested in this type of audiobook and used as part of the interactive features. Secondly, identify and extract the corresponding channel keywords from the audiobook channel.

[0147] Audiobook channels are usually classified by theme or type, such as science fiction, history, literature, etc. By analyzing the audiobook channels that users frequently visit, we can identify the user's preferred areas and extract the keywords of these channels as features. These keywords will help to understand the user's interests more accurately.

[0148] Next, the background features or plot features of the listened audiobook are analyzed and extracted from the catalog summary.

[0149] The catalog summary usually contains an overview of the main content and key plot of the audiobook. By deeply analyzing these summaries, key information such as the background setting, main characters, and storyline of the audiobook can be extracted as important features reflecting the characteristics of the book. This information will help to more fully understand the user's reading preferences.

[0150] In addition, it is also necessary to identify and extract corresponding audio keywords from the audio type of the audio book.

[0151] There are various audio types for audiobooks, including narration, dialogue, sound effects, etc. By analyzing users' preferences for different audio types, we can extract keywords that describe the characteristics of these audios, such as "affectionate narration", "lively dialogue", etc. These keywords will help us more accurately grasp users' auditory preferences, thereby providing users with more personalized recommendations.

[0152] Finally, all the above historical listening behavior features are collected and a feature set is formed.

[0153] The extracted interactive features, channel keywords, background or storyline features, and audio keywords are summarized and organized to form a feature set that fully reflects the characteristics of the user's listening behavior. This feature set will serve as the basis for subsequent analysis, modeling, and recommendations, providing strong data support.

[0154] In summary, by using feature analysis algorithms to conduct in-depth analysis of audiobook behavior data, we can effectively extract the user's historical audiobook behavior characteristics, laying a solid foundation for building a more accurate and personalized audiobook recommendation system.

[0155] Preferably, it also includes:

[0156] Feature Enhancements:

[0157] Inputting the feature set into a preset LLM large language model;

[0158] The feature set is recombined by the LLM large language model, and the background features or plot features in the feature set and their arrangement and combination with other features are constructed and generated to construct and generate the enhanced feature set: S:

[0159] S={S1, S2, S3, S4, S5, S6, S7, S8}

[0160] in,

[0161] S1 = {background features or plot features};

[0162] S2 = {background features or plot features, interactive features};

[0163] S3 = {background features or plot features, channel keywords};

[0164] S4 = {background features or plot features, audio keywords};

[0165] S5 = {background features or plot features, interactive features, channel keywords};

[0166] S6 = {background features or plot features, interactive features, audio keywords};

[0167] S7 = {background features or plot features, channel keywords, audio keywords};

[0168] S8 = {background features or plot features, interactive features, channel keywords, audio keywords}.

[0169] This section mainly recommends audiobooks based on the background features or plot features in the user's historical audiobook information (such as the background features of stories such as cultivation, martial arts, science fiction, etc., or the plot features of character development (such as the plot features of the protagonist being a genius from birth, suddenly becoming mediocre in the middle of the story, and then upgrading later)). The background or plot features of the audiobook can be analyzed from the concise catalog (abstract) of the audiobook, so that the system can match the audiobook with the corresponding features.

[0170] In order to enrich the recommendation results, the present invention arranges the features in combination to construct more recommended feature combinations, so that the system can combine user behavior, match, predict and output several groups of recommended features (that is, several historical listening behavior features predicted by the model based on user behavior data (multiple combined features can be output)), so that the system can perform component retrieval based on the combined features or perform corresponding retrieval on each feature in the combination and generate a recommended list (listening list). When recommending a list, perform feature retrieval and recommendation according to the feature direction from S8 to S1.

[0171] For example, if the model matches the feature sets S7, S4, and S3 based on the user's behavioral characteristics, the corresponding audio books will be retrieved based on the feature sets S7, S4, and S3 and written into the list in order, and then recommended to the user. In this way, the recommendations can be sorted by relevance (that is, the more matching features, the more the recommended audio books will suit the user's taste).

[0172] The feature set is input into the preset LLM (Large Language Model) and the features are recombined to construct an enhanced feature set. The following steps can be referred to:

[0173] 1. Input feature set to LLM large language model

[0174] Prepare feature sets: First, make sure that users have extracted and organized feature sets from the listening behavior data, including interactive features, channel keywords, background or storyline features, and audio keywords (the selected listening audio mode).

[0175] Input data: Input the organized feature set into the pre-set LLM large language model in an appropriate format (such as JSON, CSV, etc.). This usually involves encoding the feature set into a format that the model can understand and may include some preprocessing steps such as standardization, normalization, or vectorization.

[0176] 2. Feature Recombination and Permutation

[0177] Leveraging the generative capabilities of LLM: The LLM large language model has powerful text generation and understanding capabilities, which can be used to perform deep feature recombination on the input feature set.

[0178] Background or storyline features as the core: With background or storyline features as the core, try to make various permutations and combinations with other features (such as interactive features, channel keywords, audio keywords). This can be achieved by inputting specific instructions or prompts into the model, allowing the model to generate new feature combinations based on these instructions.

[0179] Explore diversity: Encourage the model to generate a variety of different feature combinations to increase the diversity and richness of the feature set. This can be achieved by adjusting the model's generation parameters or introducing some randomness.

[0180] 3. Build and generate enhanced feature sets

[0181] Evaluation and screening: Evaluate the feature combinations generated by the model and screen out those that are representative, can accurately reflect the user's listening behavior preferences, and are useful for the recommendation system.

[0182] Construct a new feature set: Combine the selected features to construct a new, enhanced feature set that is more comprehensive, accurate, and expressive than the original feature set.

[0183] Output and storage: The enhanced feature set is output and stored in an appropriate format for subsequent use in the recommendation system.

[0184] Continuous updating and optimization: As users’ listening behavior changes and audiobook content is updated, the feature set should be continuously updated and optimized to maintain the accuracy and effectiveness of the recommendation system.

[0185] Through the above steps, the LLM large language model can be used to recombine and rank the original feature set, so as to construct and generate an enhanced feature set, providing more comprehensive, accurate and personalized user feature information for the audio book recommendation system.

[0186] In addition to the above features, we can also conduct feature analysis on the user's listening behavior preferences. Combined with the user's listening time interval, we can introduce a listening attenuation function and construct behavioral preference features to facilitate targeted retrieval by the system based on the user's login time.

[0187] Preferably, the interactive feature includes a behavior preference score P:

[0188] P=(w C *C+w T *T+w F *F+w R *R+w M *M)e -λt ,

[0189] The meanings of each behavior variable are as follows:

[0190] C: number of clicks;

[0191] T: listening time;

[0192] F: number of collections;

[0193] R: Rating (between 1 to 5 or 1 to 10);

[0194] M: number of comments;

[0195] w C , w T , w F , w R , w M are the weight coefficients of the corresponding behavioral variables, which are set by system default;

[0196] e -λt is an exponential decay function, where t is the time difference between the user's listening behavior and the last time they listened to the book, and λ is the decay rate (with a value of 1-5).

[0197] Preferably, in step S2, when the audiobook system background reads the user's historical audiobook behavior data, it also includes:

[0198] Read the time when the user last listened to a book, and calculate the time difference from the last time the user listened to a book.

[0199] Define the behavior variables:

[0200] C: Clicks

[0201] T: Listening time (total minutes or total hours)

[0202] F: Number of favorites

[0203] R: Rating (usually a number between 1 and 5 or 1 and 10)

[0204] M: Number of comments

[0205] Assign a weight to each behavior:

[0206] P=(w C *C+w T *T+w F *F+w R *R+w M *M)e -λt

[0207] These weights can be determined based on actual business scenarios and data analysis, reflecting the importance of each behavior to user preferences.

[0208] Considering time decay:

[0209] If you want the user's recent behavior to have a greater impact on the preference score, you can introduce a time decay factor for each behavior.

[0210] For example, we can use an exponential decay function e -λt , where t is the time difference between each user's listening behavior and the last listening behavior, and λ is the decay rate (with a value of 1-5).

[0211] In this case, the contribution of each behavior will be multiplied by the corresponding time decay factor

[0212] Calculus approximation (theoretical explanation):

[0213] In the continuous case, if it is assumed that the user's behavior changes continuously over time, then integration can be used to accumulate the effects of these behaviors.

[0214] For example, for the listening time T(t) (as a function of time), its integral ∫T(t)d will give the total listening time.

[0215] However, in practical applications, usually only discrete data points are available, so using a weighted summation to approximate the integral is a more practical approach.

[0216] Practical Application:

[0217] In practical applications, the collected user behavior data is substituted into the above formula to calculate the behavior preference score of each user.

[0218] Subsequent models can be combined with users' behavioral preference scores to recommend audiobooks that better suit their preferences.

[0219] After detecting that the user has logged into the platform, the system can calculate the time interval between the previous and subsequent listening sessions based on the listening logs and the time difference, and calculate the score according to the above P-value calculation formula. In this way, the system can make comprehensive listening recommendations based on the user's listening attenuation characteristics.

[0220] After the platform recommends a list of audiobooks, the user selects and provides feedback through the front-end page of the APP, and the back-end plays the audio data of the corresponding audiobook based on the selection information.

[0221] Obviously, those skilled in the art should understand that the implementation of all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments. Those skilled in the art can understand that the implementation of all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (Flash Memory), a hard disk (Hard Disk Drive, abbreviated: HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

[0222] Example 2

[0223] like Figure 3 As shown, based on the implementation principle of Example 1, on the other hand, the present application proposes a platform for implementing the above-mentioned audio book recommendation method based on big data intelligent recommendation, including:

[0224] User management module, used to manage audiobook users;

[0225] Log module, used to record the user's historical listening behavior data;

[0226] The audiobook intelligent matching module is used to read the user's historical audiobook behavior data; input the historical audiobook behavior data into a pre-deployed audiobook recommendation model, identify the behavior features in the historical audiobook behavior data through the audiobook recommendation model, and output the relevant historical audiobook behavior features;

[0227] An audiobook recommendation module, for retrieving and generating a corresponding audiobook list based on the historical audiobook behavior characteristics, and recommending the audiobook list to the front-end page;

[0228] The audiobook execution module is used to activate and play the selected audiobook in the audiobook listening list according to the user's selection information of the audiobook in the audiobook listening list fed back by the front end, and the audiobook listening recommendation method based on big data intelligent recommendation.

[0229] Please understand the interaction between the backend and frontend of the above platform in conjunction with the corresponding steps in Example 1, and will not be repeated here.

[0230] The front-end and back-end design architecture of the platform can be understood in combination with the architecture of the existing audiobook platform and its interaction mode, and this embodiment does not limit it.

[0231] It can also be applied on existing audiobook platforms.

[0232] The modules or steps of the present invention described above can be implemented using a general computing platform, they can be concentrated on a single computing platform, or distributed on a network composed of multiple computing platforms, and optionally, they can be implemented using program codes executable by the computing platform, so that they can be stored in a storage platform and executed by the computing platform, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0233] Example 3

[0234] like Figure 4 As shown, further, in another aspect, the present application also proposes an electronic device, including:

[0235] processor;

[0236] a memory for storing processor-executable instructions;

[0237] Among them, the processor is configured to implement the audio book recommendation method based on big data intelligent recommendation described in Example 1 when executing the executable instructions.

[0238] The electronic device of the embodiment of the present disclosure includes a processor and a memory for storing instructions executable by the processor. The processor is configured to implement the audio book recommendation method based on big data intelligent recommendation described in the above embodiment 2 when executing the executable instructions.

[0239] Here, it should be noted that the number of processors can be one or more. At the same time, the electronic device of the embodiment of the present disclosure may also include an input platform and an output platform. Among them, the processor, memory, input platform and output platform may be connected through a bus or in other ways, which are not specifically limited here.

[0240] The memory, as a computer-readable storage medium, can be used to store software programs, computer executable programs and various modules, such as the program or module corresponding to the audiobook recommendation method based on big data intelligent recommendation in the embodiment of the present disclosure. The processor executes various functional applications and data processing of the electronic device by running the software programs or modules stored in the memory.

[0241] The input platform can be used to receive input numbers or signals. The signal can be a key signal related to user settings and function control of the device / terminal / server. The output platform can include display devices such as display screens.

[0242] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. An audiobook recommendation method based on big data intelligent recommendation, characterized in that: The steps include: S1. The user logs in to the audiobook system backend; S2, the audiobook system background reads the user's historical audiobook behavior data; S3, inputting the historical listening behavior data into a pre-deployed audiobook recommendation model, identifying behavior features in the historical listening behavior data through the audiobook recommendation model and outputting relevant historical listening behavior features; S4, retrieving and generating a corresponding audiobook listening list based on the historical listening behavior characteristics, and recommending the audiobook listening list to the front-end page; S5. Activate and play the selected audio book in the audio book listening list according to the user's selection information fed back by the front end.

2. The audio book recommendation method based on big data intelligent recommendation according to claim 1, characterized in that: The method for generating the audio book recommendation model comprises: Data Collection Collecting a number of historical audiobook listening behavior data of the user, wherein the audiobook listening behavior data includes interactive behavior data participated in during the audiobook listening process, the audiobook listening channel, the catalog summary of the audiobook listened to, and the audiobook audio type selected; Feature Engineering Analyzing and extracting historical listening behavior features from the listening behavior data, including: analyzing and extracting interactive features of the user's interest in the audiobook from the interactive behavior data, identifying and extracting corresponding channel keywords from the audiobook channel, analyzing and extracting background features or plot features of the audiobook from the catalog summary, identifying and extracting corresponding audio keywords from the audio type of the audiobook; collecting the historical listening behavior features and forming a feature set; Model training and validation Dividing the feature set into a training set and a validation set according to a preset ratio; Input the training set into a preset random forest model to perform feature training and learning to generate an audiobook recommendation model; The recommendation performance of the audiobook recommendation model is verified using the verification set: If the verification is successful, the audiobook recommendation model is deployed on the audiobook system backend; If the verification fails, the above steps are repeated to retrain and generate the audio book recommendation model.

3. The audio book recommendation method based on big data intelligent recommendation according to claim 2, characterized in that: Also includes: Feature Enhancements: Inputting the feature set into a preset LLM large language model; The feature set is recombined by the LLM large language model, and the background features or plot features in the feature set and their arrangement and combination with other features are performed to construct and generate the enhanced feature set: S: S={S1, S2, S3, S4, S5, S6, S7, S8} in, S1 = {background features or plot features}; S2 = {background features or plot features, interactive features}; S3 = {background features or plot features, channel keywords}; S4 = {background features or plot features, audio keywords}; S5 = {background features or plot features, interactive features, channel keywords}; S6 = {background features or plot features, interactive features, audio keywords}; S7 = {background features or plot features, channel keywords, audio keywords}; S8 = {background features or plot features, interactive features, channel keywords, audio keywords}.

4. The audio book recommendation method based on big data intelligent recommendation according to claim 3 is characterized in that: The interactive features include a behavior preference score P: P=(w C *C+w T *T+w F *F+w R *R+w M *M)e -λt , The meanings of each behavior variable are as follows: C: number of clicks; T: listening time; F: number of collections; R: Rating (between 1 to 5 or 1 to 10); M: number of comments; w C , w T , w F , w R , w M are the weight coefficients of the corresponding behavioral variables, which are set by system default; e -λt is an exponential decay function, where t is the time difference between the user's listening behavior and the last time they listened to the book, and λ is the decay rate (with a value of 1-5).

5. The audio book recommendation method based on big data intelligent recommendation according to claim 4 is characterized in that: In step S2, when the audiobook system background reads the user's historical audiobook behavior data, it also includes: Read the time when the user last listened to a book and calculate the time difference from the last time the user listened to a book.

6. A platform for implementing the audio book recommendation method based on big data intelligent recommendation as described in any one of claims 1 to 5, characterized in that: include: User management module, used to manage audiobook users; Log module, used to record the user's historical listening behavior data; The audiobook intelligent matching module is used to read the user's historical audiobook behavior data; Inputting the historical listening behavior data into a pre-deployed audiobook recommendation model, identifying behavior features in the historical listening behavior data through the audiobook recommendation model and outputting relevant historical listening behavior features; An audiobook recommendation module, for retrieving and generating a corresponding audiobook list based on the historical audiobook behavior characteristics, and recommending the audiobook list to a front-end page; The audiobook execution module is used to activate and play the selected audiobook in the audiobook listening list according to the user's selection information of the audiobook fed back by the front end.

7. An electronic device, characterized in that include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to implement the audio book recommendation method based on big data intelligent recommendation as described in any one of claims 1-5 when executing the executable instructions.

Images