New media operation content recommendation system and method based on big data

Abstract

The invention discloses a new media operation content recommendation system based on big data. The system comprises an operation content classification module used for classifying new media operation contents; the user behavior data acquisition module is used for acquiring behavior data contents of the user on the new media platform; the user behavior classification module is used for classifying the collected user behavior data content; the content and behavior matching module is used for grading the matching of each operation content and the user behavior data content according to the matching degree; the comprehensive score calculation module is used for obtaining the operation content with the highest comprehensive score according to a matching scoring result; the content recommendation module is used for recommending the operation content with the highest comprehensive score to the user; the system can effectively improve the personalized recommendation capability of the new media operation content, improve the user experience, enhance the user stickiness, and bring higher user activeness and conversion rate to an operator.

Description

TECHNICAL FIELD

[0001] The present application relates to a new media operation content recommendation system and method based on big data. BACKGROUND

[0002] With the rapid development of new media platforms, the production and dissemination of content have become more diversified, with various content forms such as audio, text, and images coexisting. How to improve the accuracy of personalized content recommendation and enhance user experience has become a core issue in new media operations. Currently, recommendation systems have a wide range of applications in personalized content recommendation, especially in e-commerce, social platforms, and news delivery. However, there are still many technical challenges.

[0003] Most current content recommendation systems are based on the analysis of a single content form, such as processing only text, audio, or images. Although some technologies can handle multiple content forms, there are still many technical challenges in efficiently and accurately splitting and classifying different types of content (audio, text, and images) in specific applications. For example, the diversity of images, the complexity of audio, and the richness of text content all pose high requirements for classification accuracy. Traditional classification methods cannot efficiently process these multi-modal data and cannot fully exploit the deep features of different types of content.

[0004] Although modern big data technology can collect user behavior data on the platform to some extent, the effectiveness and accuracy of these data are still limited. The amount of user behavior data such as clicks, plays, comments, and shares is huge, and how to extract valuable information from these massive data and accurately identify user interests is still a technical bottleneck. For example, it is difficult to accurately capture users' implicit interests and long-term preferences from short-term behavior, which greatly reduces the accuracy of the recommendation results. Existing behavior data analysis methods cannot adapt to changes in user needs in real time, resulting in low recommendation accuracy and poor user experience.

[0005] Existing recommendation systems typically use single content form matching algorithms when processing multi-modal data (such as audio, text, and images). However, in actual applications, different types of content have different effects on user behavior, and how to effectively integrate the features of these content forms to achieve high matching degrees is a difficult problem that current technology cannot solve. For example, users' preferences for audio content may differ from their preferences for images and text. How to find a reasonable matching mechanism between multi-modal content to ensure the relevance and diversity of recommended content is still a technical difficulty.

[0006] The core of a recommendation system is to score content through algorithms and recommend to users based on the scores. Currently, most recommendation systems rely on simple scoring mechanisms, such as similarity-based scoring or collaborative filtering-based scoring. However, this scoring mechanism is often too simplistic and fails to fully consider the heterogeneity between different content types and the complexity of user behavior. For example, a user's interest in a certain type of content may not be single-dimensional, but multi-dimensional, which requires the recommendation system to consider multiple behavioral characteristics of the user when scoring and make fine-tuned adjustments. In addition, the scoring model of the recommendation system often lacks flexibility and real-time performance, resulting in the inability to effectively adapt to changes in user demand during the recommendation process, thereby affecting the recommendation accuracy.

[0007] New media platforms generate massive amounts of user behavior data and operational content data every day. How to quickly and efficiently process massive amounts of data is an important problem in current technology. Although existing big data processing frameworks and technologies can support batch processing of data to some extent, there are still significant bottlenecks in real-time performance, computational efficiency, and system load. For example, real-time processing of user behavior data and immediate recommendation requires processors with sufficient computing and storage capabilities, but existing technologies often cannot achieve efficient processing of massive amounts of data while ensuring real-time recommendation. In addition, how to maintain system stability and scalability while ensuring efficient processing of the recommendation system is still a problem to be solved.

[0008] Although personalized recommendation has achieved certain results, it still faces some important challenges, such as how to dynamically adjust the recommendation strategy, how to timely capture changes in user interest, and how to optimize the recommendation system based on user feedback. Existing recommendation systems rely heavily on historical behavior data, but often ignore rapid changes in user interest, resulting in poor long-term relevance of recommended content and reduced user stickiness. To address this issue, existing technical frameworks and algorithms are still insufficient to achieve a more intelligent and adaptive recommendation mechanism. SUMMARY

[0009] The purpose of the present application is to provide a new media operational content recommendation system and method based on big data, which can effectively improve the personalized recommendation capability of new media operational content, improve user experience, enhance user stickiness, and bring higher user activity and conversion rate to the operator.

[0010] The technical solution adopted by the present application to solve its technical problems is:

[0011] The new media operational content recommendation system based on big data comprises:

[0012] The operational content classification module is used to split the new media operational content into operational audio content, operational text content, and operational image content, and classify each type of content.

[0013] A user behavior data collection module is configured to collect user behavior data on a new media platform, including audio content, text content, and image content of the user;

[0014] A user behavior classification module is configured to classify the collected user behavior data into audio content, text content, and image content of the user, and further classify each type of user behavior data;

[0015] A content and behavior matching module is configured to match the operation audio content, operation text content, and operation image content with the user audio content, user text content, and user image content, and score the matching of each operation content and user behavior data according to the matching degree;

[0016] A comprehensive score calculation module is configured to calculate the comprehensive score of each operation content according to the matching score, and obtain the operation content with the highest comprehensive score;

[0017] A content recommendation module is configured to recommend the operation content with the highest comprehensive score to the user.

[0018] Preferably, the operation audio content, operation text content, and operation image content are classified based on content type, theme, and emotional color; the user behavior data collection module includes data types of user playback records, click behavior, comment behavior, sharing behavior, and collection behavior; and the user audio content, user text content, and user image content are classified according to user interest preferences, historical behavior, and social interaction data.

[0019] Another technical problem to be solved by the present application is to provide a new media operation content recommendation method based on big data, including the following steps:

[0020] The new media operation content is divided into operation audio content, operation text content, and operation image content, and each type of content is classified;

[0021] The user behavior data on the new media platform is collected, including audio content, text content, and image content of the user, and the behavior data is divided into user audio content, user text content, and user image content;

[0022] The user behavior data is classified, and the user audio content, user text content, and user image content are further classified according to user interest preferences and historical behavior data;

[0023] The operation audio content, operation text content and operation image content are matched with the user audio content, user text content and user image content, and a score is given to the matching of each operation content and user behavior data content based on the matching degree;

[0024] The comprehensive score of each operation content is calculated, and the operation content with the highest comprehensive score is obtained;

[0025] The operation content with the highest comprehensive score is recommended to the user.

[0026] As preferred, the new media operation content is divided into operation audio content, operation text content and operation image content, and the classification method for each type of content is as follows:

[0027] The frequency spectrum features of the audio are extracted using the audio signal processing method;

[0028] A deep learning model is trained to classify the audio, and the specific formula is as follows:

[0029] Audio Category=f audio (C audio )

[0030] Wherein, f audio is the audio classification function, C audio is the audio content;

[0031] TF-IDF is used to convert the text into vector representation;

[0032] A deep learning model is used to classify the text, and the specific formula is as follows:

[0033] Text Category=f text (C text )

[0034] Wherein, f text is the text classification function, C text is the text content;

[0035] Convolutional neural network is used to extract high-level features from images;

[0036] A deep learning model is used to classify the image, and the specific formula is as follows:

[0037] Image Category=f image (C image )

[0038] Wherein, f image is the image classification function, C image is the image content;

[0039] The comprehensive classification result is represented as:

[0040] C audio → {Audio Category1, Audio Category2, …}

[0041] C text → {Text Category1, Text Category2, …}

[0042] C image → {Image Category1, Image Category2, …}.

[0043] wherein the classification result of each type of content contains multiple categories.

[0044] As preferred, the behavior data content of the user on the new media platform is obtained, including the audio content, the text content and the image content of the user, and the method for splitting these behavior data into the user audio content, the user text content and the user image content is:

[0045] If it is a voice message or a voice comment, it is converted into text through voice recognition technology and classified as audio behavior data;

[0046] If it is an audio file, the metadata and the audio of the audio file can be directly stored;

[0047] C audio = {Audio Content1, Audio Content2, …}

[0048] wherein Caudio represents the audio content of the user, and AudioContenti is each piece of audio content published by the user;

[0049] The text content is extracted from the interaction record of the user;

[0050] The extracted text content is cleaned;

[0051] C text = {Text Content1, Text Content2, …}

[0052] wherein Ctext represents the text content of the user, and TextContenti is each piece of text content published by the user;

[0053] The image content is extracted from the upload record of the user;

[0054] A pre-trained deep learning model is used for image classification;

[0055] C image={Image Content1, Image Content2, …}

[0056] Wherein, Cimage represents the image content of the user, ImageContenti is each image uploaded by the user;

[0057] Through the above steps, the user's behavior data is divided into three categories of audio, text and image, each category of data contains the content uploaded or interacted by the user, and each category of data is further stored and processed;

[0058] C user ={C audio , C text , C image}

[0059] Wherein, Cuser represents all the behavior data of the user, including audio data Caudio, text data Ctext and image data Cimage.

[0060] As a preferred, the user behavior data content is classified, and the method for further classifying the user audio content, the user text content and the user image content according to the user's interest preference and historical behavior data is:

[0061] Based on the user historical behavior data, a model of user interest preference is established;

[0062] P user =f(H audio , H text , H image )

[0063] Wherein, Puser is the interest preference of the user, Haudio, Htext, Himage respectively represent the historical behavior data of the user in audio, text and image;

[0064] The audio feature extraction technology is used to extract the content features of the audio, and the deep learning model is used to classify the audio based on these features;

[0065] C audio ={Musi, Speech, Noise, …}

[0066] The text content is analyzed by natural language processing, first, the word segmentation and word embedding are performed, and then the machine learning model is used to classify the text;

[0067] C text ={Positive, Negative, Neutral, …} using convolutional neural network for image classification, feature extraction and classification of images through deep learning model;

[0068] C image ={Landscape, Portrait, Abstract, …}

[0069] Combine the user's interest preferences with the classification model to further refine the user's classification of audio, text, and image content;

[0070] C audio =f(P user , H audio )

[0071] C text =f(P user , H text )

[0072] C image =f(P user , H image )

[0073] where Puser represents the user's interest preferences, Haudio, Htext, and Himage are the user's audio, text, and image history behavior data, and Caudio, Ctext, and Cimage are the classified audio, text, and image content categories.

[0074] As a preferred method, the operating audio content, operating text content, and operating image content are matched with the user audio content, user text content, and user image content, and a scoring method is used to score the matching of each operating content and user behavior data content based on the matching degree:

[0075] The operating audio content, operating text content, and operating image content are converted into vector representations with the user audio content, user text content, and user image content;

[0076] Compare the operating content with the user content, and calculate their similarity through cosine similarity;

[0077]

[0078] where A·B is the dot product of two vectors, ‖A‖ and ‖B‖ are the norms of the vectors, and the cosine value of the angle between the two vectors is calculated. The closer the cosine value is to 1, the higher the similarity;

[0079] Assign different weights w audio ,w text ,w image to each content type, and then calculate the weighted comprehensive score based on the matching degree;

[0080] S total (U) = w audio × S audio(U)+w text x S text (U)+w image x S image (U)

[0081] wherein Saudio(U), Stext(U), Simage(U) are the matching degree scores of audio, text and image content and user behavior data respectively, w audio ,w text ,w image are weight coefficients used to adjust the relative importance of each content type.

[0082] As a preferred, the comprehensive score of each operation content is calculated, and the method for obtaining the operation content with the highest comprehensive score is as follows:

[0083] The formula of the comprehensive matching degree score is as follows:

[0084]

[0085] wherein A audio ,A text ,A image are vectors of operation audio, text and image content respectively, U audio ,U text ,U image are vectors of user audio, text and image content respectively, w audio ,w text ,w image are weights of various contents.

[0086] Another technical problem to be solved by the application is to provide an electronic device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the big data-based new media operation content recommendation method according to any one of the above when executing the program.

[0087] Another technical problem to be solved by the application is to provide a computer readable storage medium, which stores a computer program, wherein the program is executed by a processor to implement the big data-based new media operation content recommendation method according to any one of the above.

[0088] The application has the following beneficial effects:

[0089] Through in-depth analysis of user behavior data (including splitting and classification of audio, text and image content), the system can better understand each user's preferences for different content types, thus achieving precise personalized recommendations. Users can obtain highly relevant content to their interests, improving user experience; multi-dimensional matching of operational content (audio, text, images) with user behavior not only considers single content types, but also integrates multiple content forms for recommendations, avoiding recommendations based on only a single dimension (such as text or audio), increasing the diversity and relevance of recommended content.

[0090] The system can intelligently recommend corresponding content based on user interest in audio, text and image content, meeting user needs in different scenarios. For example, when a user is listening to audio, they may have lower interest in text content, while when browsing articles, their preference for audio and images may be higher; based on accurate matching, users can see more content that may interest them, thereby improving platform content exposure and user engagement.

[0091] Through intelligent content classification and user behavior analysis, the platform can more efficiently operate content, avoiding resource waste, and operational staff can target content push and promotion based on the high-matching content provided by the recommendation system, improving content dissemination effectiveness; the comprehensive scoring calculation module accurately scores each operational content, helping the operational team quickly understand which content is most popular, thus making more scientific decisions in future content creation and promotion; based on user behavior data and content matching, advertisers can more accurately select advertising content for placement through the recommendation system, thereby improving advertising conversion and effectiveness, and avoiding ineffective advertising exposure. Specific implementation method

[0092] The principles and features of the present application are described below, and the examples provided are only for the purpose of explaining the present application and are not intended to limit the scope of the present application. In the following paragraphs, the present application is described in more detail by way of example. The advantages and features of the present application will become more apparent from the following description and claims.

[0093] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terminology used in the specification of the present application herein is only for the purpose of describing specific embodiments and is not intended to limit the present application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Embodiments

[0094] The new media operational content recommendation system based on big data includes:

[0095] An operation content classification module is configured to split new media operation content into operation audio content, operation text content, and operation image content, and classify each type of content.

[0096] A user behavior data collection module is configured to collect user behavior data on a new media platform, including user audio content, text content, and image content.

[0097] A user behavior classification module is configured to split the collected user behavior data into user audio content, user text content, and user image content, and classify each type of user behavior data.

[0098] A content and behavior matching module is configured to match operation audio content, operation text content, and operation image content with user audio content, user text content, and user image content, and score the matching of each operation content and user behavior data content based on the matching degree.

[0099] A comprehensive score calculation module is configured to calculate the comprehensive score of each operation content based on the matching score results, and obtain the operation content with the highest comprehensive score.

[0100] A content recommendation module is configured to recommend the operation content with the highest comprehensive score to the user.

[0101] The classification of operation audio content, operation text content, and operation image content is based on content type, theme, and emotional color; the user behavior data collection module includes data types such as user playback records, click behavior, comment behavior, sharing behavior, and collection behavior; the classification of user audio content, user text content, and user image content is based on user interest preferences, historical behavior, and social interaction data.

[0102] Through classification based on content type, theme, and emotional color, combined with user interest preferences, historical behavior, and social interaction data, the system can provide highly personalized content recommendations for each user. Users can see audio, text, and image content that best suits their tastes, thereby increasing user satisfaction and platform stickiness.

[0103] Collecting and analyzing user behavior data (such as playback records, clicks, comments, shares, collections, etc.) can provide in-depth understanding of user interests and behavior patterns. Through these data, the platform can actively push content that users may be interested in, promote more interaction and social sharing, and improve user activity and engagement on the platform.

[0104] Based on user behavior data and content classification analysis, the platform can identify which types of content are most popular, which themes and emotional color content can resonate with users. This provides data support for content creators and operation teams, helping them adjust their creative direction, optimize content strategy, and improve content quality and overall platform performance.

[0105] A new media operation content recommendation method based on big data includes the following steps:

[0106] The new media operation content is divided into operation audio content, operation text content and operation image content, and each type of content is classified;

[0107] Obtain user behavior data content on the new media platform, including user audio content, text content and image content, and split these behavior data into user audio content, user text content and user image content;

[0108] Classify user behavior data content, further classify user audio content, user text content and user image content according to user interest preferences and historical behavior data;

[0109] Match operation audio content, operation text content and operation image content with user audio content, user text content and user image content, and score each operation content and user behavior data content based on matching degree;

[0110] Calculate the comprehensive score of each operation content to get the operation content with the highest comprehensive score;

[0111] Recommend the operation content with the highest comprehensive score to the user.

[0112] Through detailed analysis of user behavior data, including interest preferences of audio, text and image content, the unique needs and interests of each user can be accurately identified. This fine-grained personalized recommendation can significantly improve user experience, allowing users to see the most relevant content to their interests, increasing user satisfaction and platform stickiness.

[0113] This method classifies and matches different types of content to ensure that the recommended content is highly relevant to the user's interests, avoiding irrelevant content interference. This not only improves the accuracy of content recommendation, but also helps the platform improve user engagement and reduce user churn.

[0114] Based on big data and real-time behavior data analysis, the recommendation system can dynamically adjust content recommendation strategies. This means that as user behavior changes, the recommendation system will continuously optimize recommended content, providing the latest and most relevant content to improve platform activity and long-term user retention.

[0115] The new media operation content is divided into operation audio content, operation text content and operation image content, and the method for classifying each type of content is:

[0116] The frequency spectrum features of the audio are extracted using an audio signal processing method;

[0117] A deep learning model is trained to classify the audio, and the specific formula is as follows:

[0118] Audio Category=f audio (C audio )

[0119] Wherein, f audio is an audio classification function, and C audio is audio content;

[0120] The text is converted into a vector representation using TF-IDF;

[0121] A deep learning model is used to classify the text, and the specific formula is as follows:

[0122] Text Category=f text (C text )

[0123] Wherein, f text is a text classification function, and C text is text content;

[0124] High-level features are extracted from the image using a convolutional neural network;

[0125] A deep learning model is used to classify the image, and the specific formula is as follows:

[0126] Image Category=f image (C image )

[0127] Wherein, f image is an image classification function, and C image is image content;

[0128] The classification results are combined and represented as:

[0129] C audio →{Audio Category1,Audio Category2,…}

[0130] C text →{Text Category1,Text Category2,…}

[0131] C image→ {Image Category1, Image Category2, …}.

[0132] wherein the classification result of each category of content contains multiple categories.

[0133] This method can process and classify audio, text and image content separately, fully utilizing the strengths of each data type. For example, audio content is extracted through spectral features, text content is converted into vector representation through TF-IDF and deep learning models, and images are extracted through convolutional neural networks. This multi-modal classification processing method allows each type of content to be processed in the most appropriate way according to its characteristics, improving the accuracy and efficiency of classification.

[0134] By training separate deep learning models for each type of content and combining multiple feature extraction methods, the user's interest preferences for different types of content can be more accurately understood. By analyzing user behavior, user needs can be better matched, thereby improving the accuracy of content recommendation. This means that users can receive more personalized and relevant content, improving user experience and increasing user stickiness and platform activity.

[0135] Using deep learning models and spectral feature analysis, the system can continuously optimize classification results based on changes in content and user interests. Whether it is audio, text or image, the classification model can continuously learn and optimize as the amount of data increases, ensuring real-time updates and accurate matching of content classification. This dynamic adjustment ensures that the platform's content remains highly consistent with user needs, increasing the platform's long-term appeal and competitiveness.

[0136] Obtain user behavior data content on the new media platform, including user audio content, text content and image content, and split these behavior data into user audio content, user text content and user image content as follows:

[0137] If it is a voice message or voice comment, it is converted into text through voice recognition technology and classified as audio behavior data;

[0138] If it is an audio file, the metadata and audio of the audio file can be directly stored;

[0139] C audio = {Audio Content1, Audio Content2, …}

[0140] wherein Caudio represents the user's audio content, and AudioContenti is each piece of audio content published by the user.

[0141] extracting text content from the user's interaction records;

[0142] cleaning the extracted text content;

[0143] C text = {Text Content1, Text Content2, …}

[0144] where Ctext represents the user's text content, and TextContenti is each piece of text content posted by the user;

[0145] extracting image content from the user's upload records;

[0146] using a pre-trained deep learning model for image classification;

[0147] C image = {Image Content1, Image Content2, …}

[0148] where Cimage represents the user's image content, and ImageContenti is each image uploaded by the user;

[0149] Through the above steps, the user's behavior data is divided into audio, text, and image three categories, each category of data contains the content uploaded or interacted by the user, and each category of data is further stored and processed;

[0150] C user = {C audio , C text , C image}

[0151] where Cuser represents all the user's behavior data, including audio data Caudio, text data Ctext, and image data Cimage.

[0152] By dividing the user's behavior data into audio, text, and image three categories, each type of content can be processed and analyzed according to its characteristics. This classification makes the storage, management, and subsequent analysis of data more clear and efficient. For example, audio data can be classified as audio behavior data after being converted into text through speech recognition, text data can be cleaned and classified, and image data can be identified and analyzed through deep learning. This way, the confusion of different types of content can be avoided, and each type of data can be processed in the most suitable way.

[0153] Separate processing of different types of user behavior data can help the platform gain a deeper understanding of users' interests and needs. For example, through audio recognition and image classification, the platform can accurately capture the details of user interactions and analyze their preferences in different media. By further analyzing these behavior data, the platform can provide more personalized and accurate recommendations or services to users, thereby improving user experience and platform stickiness.

[0154] This solution enables the platform to conduct in-depth analysis and optimization for audio, text, and image content types respectively. Classification of audio, text, and image content helps better content recommendation, such as identifying user interests through uploaded audio data, improving text recommendation accuracy through semantic analysis of user text content, and more accurately understanding user visual needs through deep learning classification of image content. Ultimately, this classification optimization will greatly improve the accuracy of the platform's personalized recommendation system and services, providing users with more content that meets their interests and needs.

[0155] The method for classifying user behavior data content and further classifying user audio content, user text content, and user image content based on user interest preferences and historical behavior data is as follows:

[0156] Based on user historical behavior data, a model of user interest preferences is established;

[0157] P user = f(H audio , H text , H image )

[0158] Where Puser is the user's interest preference, Haudio, Htext, and Himage represent the user's historical behavior data in audio, text, and image respectively;

[0159] Audio feature extraction techniques are used to extract content features of audio, and deep learning models are used to classify audio based on these features;

[0160] C audio = {Music, Speech, Noise, …}

[0161] Text content is analyzed using natural language processing. First, word segmentation and word embedding are performed, and then machine learning models are used to classify text;

[0162] C text = {Positive, Negative, Neutral, …}

[0163] Convolutional neural networks are used for image classification, and deep learning models are used for feature extraction and classification of images;

[0164] C image = {Landscape, Portrait, Abstract,...}

[0165] Combine the user's interest preferences with the classification model to further refine the user's classification of audio, text, and image content.

[0166] C audio = f(P user , H audio )

[0167] C text = f(P user , H text )

[0168] C image = f(P user , H image )

[0169] Where Puser represents the user's interest preferences, Haudio, Htext, and Himage are the user's historical behavior data for audio, text, and image, respectively, and Caudio, Ctext, and Cimage are the classified audio, text, and image content categories.

[0170] By establishing a model of user interest preferences (Puser) based on historical behavior data, the system can understand the user's preferences for various content such as audio, text, and images. Combining these preferences with classified content (such as Caudio, Ctext, and Cimage) can provide more accurate personalized recommendations. For example, if a user prefers a specific type of audio (such as news), a specific style of text (such as popular science), or a specific theme of images (such as natural landscapes), the system can recommend content that meets their interests, thereby improving user satisfaction and activity.

[0171] This solution not only focuses on a single type of content (such as audio, text, or images), but also considers historical behavior across different content types for multi-modal data fusion. This approach can comprehensively reflect the user's interests and needs. Through deep learning model analysis of audio, text, and images, the system can better understand the user's interests in multiple areas, thereby providing more comprehensive content classification and recommendations. For example, a user's behavior in audio and images may reflect his interest in a specific field, and these fields can be combined through deep learning models to provide cross-domain personalized experiences for the user.

[0172] This solution employs audio feature extraction, natural language processing, and convolutional neural networks, among other deep learning techniques, to classify audio, text, and image content. These advanced techniques can effectively improve the accuracy of content classification, reduce the need for human intervention, and handle large amounts of complex data. Through deep learning, the system can extract effective features from massive data and automatically learn how to classify, improving the efficiency and effectiveness of the content recommendation system. Additionally, as user behavior data accumulates, the model can continuously optimize, providing more accurate classification and personalized recommendations.

[0173] The method for matching the operation audio content, operation text content, and operation image content with the user audio content, user text content, and user image content, and scoring the matching of each type of operation content with user behavior data content based on the matching degree is as follows:

[0174] The operation audio content, operation text content, and operation image content are converted into vector representations, as are the user audio content, user text content, and user image content.

[0175] The operation content and user content are compared, and their similarity is calculated using cosine similarity.

[0176]

[0177] where A·B is the dot product of two vectors, ||A|| and ||B|| are the norms of the vectors, and the cosine value is calculated to represent the angle between the two vectors. The closer the cosine value is to 1, the higher the similarity.

[0178] Different weights w are assigned to each content type. audio ,w text ,w image Then, a weighted comprehensive score is calculated based on the matching degree.

[0179] S total (U) = w audio × S audio audio(U) + w text × S text text(U) + w image × S image image(U)

[0180] where Saudio(U), Stext(U), and Simage(U) are the matching degree scores for audio, text, and image content with user behavior data, respectively, and w audio ,w text ,w image are weight coefficients used to adjust the relative importance of each content type.

[0181] Assuming the user prefers audio content of the music type, likes text content about technology, and tends to prefer images of landscapes. The operator can calculate the matching degree for each type of content according to these preferences and give a matching score.

[0182] Assuming the following are the calculation results:

[0183] Audio matching degree: 0.85 (high matching)

[0184] Text matching degree: 0.65 (medium matching)

[0185] Image matching degree: 0.90 (high matching)

[0186] If the audio content is very important to the user, set the weight to 0.5, the text to 0.3, and the image to 0.2, then the final comprehensive score is:

[0187] S total = 0.5 × 0.85 + 0.3 × 0.65 + 0.2 × 0.90 = 0.425 + 0.195 + 0.18 = 0.8

[0188] This comprehensive score (0.8) can be used as the matching degree of the operator's content and the user's preferences.

[0189] By converting the operator's audio, text, and image content and the user's audio, text, and image content into vector representations and using cosine similarity for matching, the similarity of each type of content to user behavior data can be accurately measured. Cosine similarity is a common vector space similarity measure that can effectively calculate the similarity of different types of content, ensuring that the content recommended to users is more in line with their interests and preferences, improving the accuracy of the recommendation system and user experience.

[0190] By assigning different weight coefficients (waudio, wtext, wimage) to audio, text, and image content, the system can flexibly adjust the importance of each type of content. This makes it possible to make personalized recommendations based on user preferences. For example, some users may prefer audio content, while others may be more interested in image content. By adjusting the weights, the system can optimize the recommendation effect so that each user receives content that is more in line with their specific needs and interests.

[0191] By calculating a weighted comprehensive score (Saudio(U), Stext(U), Simage(U)) based on the matching degree, this scheme can consider the matching of various types of content comprehensively and provide a more comprehensive recommendation score. This is not just a single type of content recommendation, but a comprehensive score considering multiple content forms, which helps to balance the recommendation of different types of content and ensure the diversity and richness of the recommendation results. At the same time, the system can prioritize different types of content, providing more accurate and personalized content recommendations.

[0192] The method for calculating the comprehensive score of each operation content is:

[0193] The formula for comprehensive matching degree score is as follows:

[0194]

[0195] where A audio ,A text ,A image are the vectors of operation audio, text and image content, respectively, U audio ,U text ,U image are the vectors of user audio, text and image content, respectively, w audio ,w text ,w image are the weights of each type of content.

[0196] This method considers different types of content such as audio, text and image, and can combine the features of multiple content forms through vectorization and cosine similarity calculation. Users may have different preferences in different content forms, and by combining the matching degrees of these content forms, more accurate and comprehensive recommendations can be provided to meet the diverse needs of users.

[0197] By introducing weight coefficients (waudio, wtext, wimage), the system can assign different importance to different types of content according to the needs of different users. For example, if a user prefers audio content, the weight of audio content can be increased. This flexible adjustment method can dynamically optimize according to the user's personalized preferences, thereby improving the user experience.

[0198] The calculation of the comprehensive score not only considers the matching of a certain type of content, but also combines the matching degrees of multiple content types such as audio, text and image. This multi-dimensional comprehensive scoring method helps to evaluate the fit between various types of content and user preferences as a whole, making the recommendation results more balanced and comprehensive. Through weighted calculation, the final score can be adjusted according to the importance of different content types, ensuring that the recommendation results are more in line with user interests.

[0199] The embodiment also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for recommending new media operation content based on big data when executing the computer program.

[0200] The embodiment also provides a computer readable storage medium, which stores a computer program, wherein the computer program is executable on a processor to implement the method for recommending new media operation content based on big data.

[0201] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the computer program can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0202] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the system is divided into different functional units or modules to complete all or part of the above-mentioned functions.

[0203] The above-mentioned embodiments of the present application are not a limitation on the protection scope of the present application, and the embodiments of the present application are not limited thereto. According to the above-mentioned content of the present application, other various forms of modifications, replacements or changes of the above-mentioned structure of the present application can be made according to ordinary technical knowledge and common means in the art without departing from the above-mentioned basic technical idea of the present application, and all of them shall fall within the protection scope of the present application.

Claims

1. A new media operation content recommendation system based on big data, characterized in that, Comprise: Operation content classification module: for splitting new media operation content into operation audio content, operation text content and operation image content, and classifying each type of content; User behavior data acquisition module: for obtaining user behavior data content on the new media platform, the behavior data including user audio content, text content and image content; User behavior classification module: for splitting the collected user behavior data content into user audio content, user text content and user image content, and classifying each type of user behavior data content; Content and behavior matching module: for matching operation audio content, operation text content and operation image content with user audio content, user text content and user image content, and scoring the matching of each operation content and user behavior data content according to the matching degree; Comprehensive score calculation module: for calculating the comprehensive score of each operation content according to the matching score result, and obtaining the operation content with the highest comprehensive score; Content recommendation module: according to the operation content with the highest comprehensive score, recommend it to the user. 2.The big data based new media operation content recommendation system according to claim 1, characterized in that, The classification of operation audio content, operation text content and operation image content is based on content type, theme and emotional color; the user behavior data acquisition module includes data types of user's playback record, click behavior, comment behavior, sharing behavior and collection behavior; the classification of user audio content, user text content and user image content is based on user's interest preference, historical behavior and social interaction data.

3. A big data-based new media operation content recommendation method, characterized in that, Comprise the following steps: Split the new media operation content into operation audio content, operation text content and operation image content, and classify each type of content; Obtain user behavior data content on the new media platform, including user audio content, text content and image content, and split these behavior data into user audio content, user text content and user image content; Classify user behavior data content, further classify user audio content, user text content and user image content according to user's interest preference and historical behavior data; Match operation audio content, operation text content and operation image content with user audio content, user text content and user image content, and score the matching of each operation content and user behavior data content based on the matching degree; Calculate the comprehensive score of each operation content, and obtain the operation content with the highest comprehensive score; Recommend the operation content with the highest comprehensive score to the user. 4.The big data-based new media operation content recommendation method of claim 3, wherein, The method for splitting the new media operation content into operation audio content, operation text content and operation image content, and classifying each type of content is as follows: Use audio signal processing method to extract the frequency spectrum features of audio; Train a deep learning model to classify audio, the specific formula is as follows: Audio Category=f audio (C audio ) wherein f audio is an audio classification function, C audio is audio content; Use TF-IDF to convert text into vector representation; Use deep learning model to classify text, the specific formula is as follows: Text Category = f text (C text ) wherein f text is a text classification function, C text is the text content; Use convolutional neural network to extract high-level features from image; Use deep learning model to classify image, the specific formula is as follows: Image Category = f image (C image ) wherein f image is an image classification function, C image is an image content; The comprehensive classification result is represented as: C audio → {Audio Category1, Audio Category2,...} C text → {Text Category1, Text Category2,...} C image → {Image Category1, Image Category2,...}. Wherein, the classification result of each type of content contains multiple categories. 5.The big data-based new media operation content recommendation method of claim 3, wherein, The method for obtaining user behavior data content on a new media platform includes user audio content, text content, and image content, and the behavior data is divided into user audio content, user text content, and user image content as follows: If it is a voice message or voice comment, it is converted into text through voice recognition technology and classified as audio behavior data; If it is an audio file, the metadata and audio of the audio file can be directly stored; C audio = {Audio Content1, Audio Content2,...} Wherein, Caudio represents the audio content of the user, and Audio Contenti is each piece of audio content published by the user; Text content is extracted from user interaction records; The extracted text content is cleaned; C text = {Text Content1, Text Content2,...} Wherein, Ctext represents the text content of the user, and Text Contenti is each piece of text content published by the user; Image content is extracted from user upload records; A pre-trained deep learning model is used for image classification; C image = {Image Content1, Image Content2,...} Wherein, Cimage represents the image content of the user, and Image Contenti is each image uploaded by the user; Through the above steps, the user's behavior data is divided into audio, text, and image, each type of data containing user uploaded or interactive content, and each type of data is further stored and processed; C user = {C audio , C text , C image} Wherein, Cuser represents all the behavior data of the user, including audio data Caudio, text data Ctext, and image data Cimage. 6.The big data-based new media operation content recommendation method of claim 3, wherein, Classify the user behavior data content, and further classify the user audio content, user text content, and user image content based on the user's interest preferences and historical behavior data as follows: Based on the user's historical behavior data, a model of the user's interest preferences is established; P user = f(H audio , H text , H image ) Wherein, Puser is the user's interest preference, and Haudio, Htext, and Himage represent the user's historical behavior data in audio, text, and image, respectively; Use audio feature extraction technology to extract the content features of the audio, and use a deep learning model based on these features to classify the audio; C audio = {Music, Speech, Noise,...} Perform natural language processing analysis on the text content. First, perform word segmentation and word embedding, and then use a machine learning model to classify the text; C text = {Positive, Negative, Neutral,...} Use a convolutional neural network to classify images, and use a deep learning model to extract features and classify images; C image = {Landscape, Portrait, Abstract,...} Combine the user's interest preferences with the classification model to further refine the user's classification of audio, text, and image content; C audio = f(P user , H audio ) C text = f(P user , H text ) C image = f(P user , H image ) Wherein, Puser represents the user's interest preference, and Haudio, Htext, and Himage are the user's historical behavior data in audio, text, and image, respectively, and Caudio, Ctext, and Cimage are the classified audio, text, and image content categories. 7.The big data-based new media operation content recommendation method of claim 3, wherein, Match the operation audio content, operation text content, and operation image content with the user audio content, user text content, and user image content, and score the matching of each operation content with the user behavior data content based on the matching degree as follows: Convert the operation audio content, operation text content, and operation image content into vector representations; The operation content is compared with the user content, and similarity between them is calculated by cosine similarity; Wherein, A.B is the dot product of two vectors, ‖A‖ and ‖B‖ are the modulus of vectors, the cosine value of the included angle of two vectors is calculated, and the closer the cosine value is to 1, the higher the similarity is; assigning different weights w to each content type audio ,w text ,w image and then calculating a weighted overall score based on the matching degree S total (U) = w audio x S andio (U) + w text x S text (U) + w image x S image (U) where Saudio(U), Stext(U), Simage(U) are the matching degree scores of audio, text and image content with user behavior data, respectively, and w audio ,w text ,w image are weight coefficients to adjust the relative importance of each content type. 8.The big data-based new media operation content recommendation method of claim 3, wherein, The comprehensive score of each operation content is calculated, and the method for obtaining the operation content with the highest comprehensive score is: The formula of the comprehensive matching degree score is as follows: where: A audio ,A text ,A image are vectors of operational audio, text and image content, respectively, U audio ,U text ,U image are vectors of user audio, text and image content, respectively, w audio ,w text ,w image are weights for each type of content.

9. An electronic device, comprising: The computer program stored in the memory and executable on the processor, when the processor executes the program, realizes the new media operation content recommendation method based on big data as described in claims 3-8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to realize the new media operation content recommendation method based on big data as described in claims 3-8.

Images