On-demand content clustering method and apparatus, electronic device, and storage medium

Abstract

Disclosed are an on-demand content clustering method and apparatus, an electronic device, and a storage medium. In the present invention, a media asset clustering algorithm for coordinating user behavior is optimized for data characteristics in an IPTV service environment. By combining media asset content with user data, the accuracy of clustering can be effectively improved. In addition, in the present invention, feedback information from a large model is innovatively introduced to correct an IPTV recommendation model. By constructing paired tasks and querying the large model about whether the two are similar, in the present invention, the weights or parameters of the recommendation model can be adjusted according to the feedback from the large model, thereby further optimizing a clustering effect. According to the present invention, significant innovation and improvement on a multi-dimensional fine-granularity classification method for IPTV are achieved, providing advantages and effects such as efficient resource utilization, multi-dimensional classification, innovation in vectorization strategies, efficient model training and optimization, and promotion of business innovation and development, so that the present invention can be widely applied to the technical field of data processing.

Description

A method, apparatus, electronic device, and storage medium for clustering on-demand content. Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, electronic device, and storage medium for clustering on-demand content. Background Technology

[0002] With the rapid development of mobile internet technology, although smartphones and tablets occupy an important position in entertainment consumption, large-screen TVs still maintain an irreplaceable position in the film and television entertainment field due to their immersive viewing experience. IPTV, as an important component of large-screen TV services, transmits television signals via internet protocols, providing users with abundant film and television resources and a high-quality viewing experience, making it one of the mainstream viewing methods today.

[0003] However, compared to internet applications with massive user feedback data, such as social media apps, IPTV lags behind in content categorization. Its categorization is coarser and less detailed than that of social media platforms, hindering its recommendation and search functions. This is primarily because IPTV's media assets are unique, heavily reliant on information about the media itself (such as titles, genres, and cast). Furthermore, due to broadcast control restrictions on large-screen content, it lacks the strong interactivity of social media platforms. IPTV also lacks sufficient user behavior data (such as viewer ratings and program comments) to support advanced clustering algorithms. Traditional clustering methods, limited by data richness, struggle to achieve the clustering results of advanced algorithms in the internet field. Summary of the Invention

[0004] This invention proposes a method, apparatus, electronic device, and storage medium for clustering on-demand content, aiming to at least partially solve one of the technical problems in related technologies. The embodiments of this invention can effectively realize the clustering of on-demand content for IPTV.

[0005] On one hand, embodiments of the present invention provide a clustering method for on-demand content, including:

[0006] Obtain all media asset content from the internet TV set, and obtain the user attributes corresponding to each media asset content based on the playback data of the media asset content;

[0007] The media asset content is first vectorized to obtain the media asset vector corresponding to each media asset content; the user attributes are second vectorized to obtain the user vector corresponding to each user attribute.

[0008] Content data is obtained by organizing media asset content and user attributes according to the corresponding relationship; the media asset vector and user vector of the corresponding relationship are combined and concatenated to obtain the vector data corresponding to each content data;

[0009] The vector data is input into a preset classification model for classification processing to obtain the probability distribution corresponding to the vector data;

[0010] The classification model includes convolutional layers, pooling layers, fully connected layers, and an output layer; the probability distribution includes the class probability of each class, the sum of all class probabilities in the probability distribution is 1, and the highest class probability is the predicted classification of the vector data;

[0011] The information entropy corresponding to each content data is obtained by processing based on the probability distribution. The information entropy is then used to construct content pairs based on the content data. Each content pair consists of two content data.

[0012] The content pairs are input into a pre-defined large model for classification and discrimination, and the discrimination results are obtained. The discrimination results indicate whether the two content data in the content pair are of the same type.

[0013] The prediction accuracy of the classification model is obtained based on the discrimination results of all content pairs and the probability distribution of all vector data.

[0014] When the prediction accuracy is less than the preset threshold, the model parameters of the classification model are adjusted using the loss function. Then, the process of inputting the vector data into the preset classification model for classification processing and obtaining the probability distribution corresponding to the vector data is repeated until the prediction accuracy is greater than or equal to the preset threshold. The adjusted classification model is then used as the target model.

[0015] The target model is used to classify the media content of IPTV sets, so that IPTV sets can cluster the media content based on the classification results.

[0016] Optionally, the media asset content includes the content name and description, publication year, and cast and crew list; the media asset content undergoes a first vectorization process to obtain the media asset vector corresponding to each media asset content, including the following steps:

[0017] The content name and description are input into a pre-trained language representation model to obtain the first vectorized result;

[0018] The publication year determines the time range in which it falls, and a second vectorization result corresponding to the time range is obtained based on a preset first mapping table; wherein, the first mapping table includes the vector category result corresponding to each time range;

[0019] All cast and crew members in the cast and crew list are binary encoded to obtain a third vectorized result;

[0020] The media asset vector includes the first vectorization result, the second vectorization result, and the third vectorization result.

[0021] Optionally, user attributes include gender, age, geographic location, and user habit data; the user attributes undergo a second vectorization process to obtain a user vector corresponding to each user attribute, including the following steps:

[0022] The fourth vectorized result corresponding to the gender parameter is obtained based on the preset second mapping table; wherein, the second mapping table includes vector gender results corresponding to different gender parameters;

[0023] The age parameter is encoded in binary to obtain the fifth vectorized result;

[0024] The straight-line distance between the region and the target region is determined based on the regional parameters. The straight-line distance is then converted into a binary number to obtain the sixth vectorized result.

[0025] The user habit data is input into a pre-trained language representation model to obtain the seventh vectorization result; the user habit data includes viewing rate, viewing duration, and number of views.

[0026] The user vector includes the fourth vectorization result, the fifth vectorization result, the sixth vectorization result, and the seventh vectorization result.

[0027] Optionally, user attributes include the number of times a user has viewed media asset content; combining and concatenating the corresponding media asset vector and user vector includes the following steps:

[0028] Get the binary representation of the number of views;

[0029] Based on a preset length, the media asset vector, user vector, and binary representation are padded with the value 0;

[0030] The media asset vector, user vector, and binary representation after padding are sequentially concatenated.

[0031] Optionally, the vector data is input into a preset classification model for classification processing to obtain the probability distribution corresponding to the vector data, including the following steps:

[0032] Feature data is obtained by extracting features from vector data through convolutional layers;

[0033] Feature information is obtained by reducing the dimensionality of feature data through pooling layers;

[0034] The feature information is integrated and mapped through multiple fully connected layers to obtain the classification result;

[0035] The output layer contains a pre-defined classification function.

[0036] The classification result is transformed into a probability distribution through the classification function of the output layer.

[0037] Optionally, the information entropy corresponding to each content data is obtained based on the probability distribution, including the following steps:

[0038] The information entropy corresponding to the content data is obtained by performing a preset accumulation operation on the probabilities of all categories in the probability distribution;

[0039] The expression for information entropy is:

[0040] In the formula, H(X) represents the information entropy corresponding to content data X; n represents the total number of categories; i represents the i-th category; p(x) i ) represents the category probability of the i-th category.

[0041] Optionally, content pairs can be constructed based on content data using information entropy, including the following steps:

[0042] Based on all content data, initial content pairs for all possible combinations are constructed by grouping them into pairs;

[0043] The entropy value of each initial content pair is determined by the sum of the information entropy values ​​corresponding to the two content data in the initial content pair;

[0044] Use the initial content with the largest preset entropy value as the final content pair to be constructed;

[0045] Content pairs include pairs of similar content and pairs of different content.

[0046] On the other hand, embodiments of the present invention provide a clustering device for on-demand content, comprising:

[0047] The first module is used to obtain all media asset content of the network TV and obtain the user attributes corresponding to each media asset content based on the broadcast data of the media asset content.

[0048] The second module is used to perform a first vectorization process on the media asset content to obtain a media asset vector corresponding to each media asset content; and to perform a second vectorization process on the user attributes to obtain a user vector corresponding to each user attribute.

[0049] The third module is used to organize content data based on the corresponding media asset content and user attributes; it combines and concatenates the corresponding media asset vectors and user vectors to obtain the vector data corresponding to each content data.

[0050] The fourth module is used to input vector data into a preset classification model for classification processing to obtain the probability distribution corresponding to the vector data;

[0051] The classification model includes convolutional layers, pooling layers, fully connected layers, and an output layer; the probability distribution includes the class probability of each class, the sum of all class probabilities in the probability distribution is 1, and the highest class probability is the predicted classification of the vector data;

[0052] The fifth module is used to obtain the information entropy corresponding to each content data based on the probability distribution, and to construct content pairs based on the content data using the information entropy; each content pair includes two content data.

[0053] The sixth module is used to classify and discriminate the content pairs input to the pre-defined large model, and obtain the discrimination results; the discrimination results indicate whether the two content data in the content pair are of the same type;

[0054] The seventh module is used to obtain the prediction accuracy of the classification model based on the discrimination results of all content pairs and the probability distribution of all vector data;

[0055] The eighth module is used to adjust the model parameters of the classification model using the loss function when the prediction accuracy is less than the preset threshold. It then returns to the previous step of inputting the vector data into the preset classification model for classification processing to obtain the probability distribution corresponding to the vector data, until the prediction accuracy is greater than or equal to the preset threshold. The adjusted classification model is then used as the target model.

[0056] The target model is used to classify the media content of IPTV sets, so that IPTV sets can cluster the media content based on the classification results.

[0057] On the other hand, embodiments of the present invention provide an electronic device, including: a processor and a memory; the memory is used to store a program; the processor executes the program to implement the above-mentioned clustering method for on-demand content.

[0058] On the other hand, embodiments of the present invention provide a computer storage medium storing a processor-executable program, which, when executed by a processor, is used to implement the above-mentioned clustering method for on-demand content.

[0059] This invention embodiment acquires all media asset content from a network television set, and obtains user attributes corresponding to each media asset content based on the playback data of the media asset content; performs a first vectorization process on the media asset content to obtain a media asset vector corresponding to each media asset content; performs a second vectorization process on the user attributes to obtain a user vector corresponding to each user attribute; organizes the corresponding media asset content and user attributes to obtain content data; combines and concatenates the corresponding media asset vectors and user vectors to obtain vector data corresponding to each content data; inputs the vector data into a preset classification model for classification processing to obtain the probability distribution corresponding to the vector data; wherein, the classification model includes a convolutional layer, a pooling layer, a fully connected layer, and an output layer; the probability distribution includes the class probability of each category, the sum of all class probabilities in the probability distribution is 1, and the highest class probability is the predicted classification of the vector data; based on the probability distribution processing, the following is obtained: The information entropy corresponding to each content data point is used to construct content pairs based on the content data. Each content pair consists of two content data points. The content pairs are input into a pre-defined large model for classification and discrimination, yielding a discrimination result. The discrimination result indicates whether the two content data points in the content pair belong to the same category. Based on the discrimination results of all content pairs and the probability distribution of all vector data, the prediction accuracy of the classification model is obtained. When the prediction accuracy is less than a pre-defined threshold, the model parameters of the classification model are adjusted using a loss function. The process is repeated, inputting the vector data into the pre-defined classification model for classification processing to obtain the probability distribution corresponding to the vector data, until the prediction accuracy is greater than or equal to the pre-defined threshold. The adjusted classification model is then used as the target model. The target model is used to classify the media asset content of IPTVs by type, enabling IPTVs to cluster the media asset content based on the type classification results. This invention addresses the unique media asset content and limited user attributes of IPTVs. Without significantly increasing data collection costs, it fully utilizes existing data resources and uses the results of a large model to assist the classification model in achieving more accurate film and television content classification, thereby promoting the innovative development of IPTV services in the context of big data and artificial intelligence technologies. The embodiments of the present invention can effectively achieve the clustering of on-demand content. Attached Figure Description

[0060] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0061] Figure 1 is a schematic diagram of an implementation environment for clustering on-demand content provided in an embodiment of the present invention;

[0062] Figure 2 is a flowchart illustrating a clustering method for on-demand content provided in an embodiment of the present invention;

[0063] Figure 3 is a schematic diagram of the expansion process of the first vectorization process provided in an embodiment of the present invention;

[0064] Figure 4 is a schematic diagram of the expansion process of the second vectorization process provided in an embodiment of the present invention;

[0065] Figure 5 is a schematic diagram of the unfolding process of step S300 provided in an embodiment of the present invention;

[0066] Figure 6 is a schematic diagram of the overall process of the clustering method for on-demand content provided in an embodiment of the present invention;

[0067] Figure 7 is a schematic diagram illustrating the application of the large model for determining similar media assets provided in an embodiment of the present invention.

[0068] Figure 8 is a schematic diagram illustrating the application of the large model in determining different types of media assets according to an embodiment of the present invention.

[0069] Figure 9 is a schematic diagram of the structure of the on-demand content clustering device provided in an embodiment of the present invention;

[0070] Figure 10 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0072] It should be noted that although functional modules are divided in the system diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the system or the order in the flowchart. The terms "first / S100," "second / S200," etc., in the specification, claims, and the aforementioned figures are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0073] In this invention, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this invention can be combined with other embodiments.

[0074] To facilitate understanding of the technical solution of this invention, the proprietary names of the technical features that may appear in the technical solution of this invention are explained below:

[0075] IPTV (Interactive Personality TV): A network television set that uses a television, personal computer, or handheld device as a display terminal. It connects to a broadband network via a set-top box or computer to provide services such as digital television, time-shifted television, and interactive television. The emergence of IPTV has brought people a completely new way of watching television. It has changed the previous passive television viewing mode and enabled on-demand television viewing, allowing viewers to stop watching whenever they want.

[0076] BERT (Bidirectional Encoder Representation from Transformers): BERT is a pre-trained language representation model. Its core principle is to use a masked language model (MLM) to pre-train a bidirectional language model, thereby generating deep bidirectional language representations.

[0077] Figure 1 illustrates an implementation environment according to an embodiment of the present invention. Referring to Figure 1, this implementation environment includes at least one terminal 102 and a server 101. The terminal 102 and the server 101 can be connected via a network, either wirelessly or via a wired connection, to complete data transmission and exchange.

[0078] Server 101 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0079] Additionally, server 101 can also be a node server in a blockchain network. Blockchain is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms.

[0080] Terminal 102 can be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. Terminal 102 and server 101 can be directly or indirectly connected via wired or wireless communication, and this embodiment of the invention does not impose any limitations.

[0081] Based on the implementation environment shown in Figure 1, this embodiment of the invention provides a clustering method for on-demand content. The following description uses the application of this on-demand content clustering method in terminal 102 as an example. It can be understood that this on-demand content clustering method can also be applied to server 101.

[0082] Referring to Figure 2, which is a flowchart of a clustering method for on-demand content applied to a terminal according to an embodiment of the present invention, the executing entity of this on-demand content clustering method can be any of the aforementioned computer devices (including servers or terminals). Referring to Figure 2, the method includes the following steps:

[0083] S100: Obtain all media asset content from the network TV and obtain the user attributes corresponding to each media asset content based on the playback data of the media asset content;

[0084] Exemplary examples include some specific implementations:

[0085] 1. Access to media content on internet TV sets:

[0086] The media content of internet TV sets comes from a wide range of sources, which may include digital versions of traditional TV programs, online dramas, short videos, documentaries, movies, variety shows, etc.

[0087] These media assets are typically uploaded to online TV platforms by content providers for users to watch. Content providers may include television stations, film and television production companies, and individual creators.

[0088] 2. Collection of broadcast data:

[0089] Broadcast data refers to various data generated when users watch media content, such as viewing duration, number of views, playback position, number of pauses, and number of fast forwards and rewinds.

[0090] This data can be collected in real time through the back-end system of the internet TV and used to analyze users' viewing behavior and preferences.

[0091] 3. Obtaining user attributes:

[0092] Based on broadcast data, data analysis techniques can be used to infer user attributes. User attributes may include age, gender, region, occupation, interests, etc.

[0093] For example, by analyzing the types of programs users watch during a specific time period, we can infer their age and interests; by analyzing the sources of the programs users watch (such as domestic TV stations, foreign TV stations, online original content, etc.), we can infer their geographical and cultural background.

[0094] Using broadcast data of online TV media content to obtain user attributes is an effective data analysis method. It helps platforms to gain a deeper understanding of user needs and behavioral characteristics, thereby providing more personalized services and content recommendations.

[0095] S200. Perform a first vectorization process on the media asset content to obtain the media asset vector corresponding to each media asset content; perform a second vectorization process on the user attributes to obtain the user vector corresponding to each user attribute;

[0096] It should be noted that the media asset content includes the content name and description, publication year, and cast and crew list. In some embodiments, as shown in Figure 3, the media asset content undergoes a first vectorization process to obtain a media asset vector corresponding to each media asset content. This may include the following steps: S211, inputting the content name and description into a pre-trained language representation model to obtain a first vectorization result; S212, determining the era range based on the publication year, and obtaining a second vectorization result corresponding to the era range based on a preset first mapping table; wherein, the first mapping table includes vector category results corresponding to each era range; S213, performing binary encoding on all actors and staff in the cast and crew list to obtain a third vectorization result; wherein, the media asset vector includes the first vectorization result, the second vectorization result, and the third vectorization result.

[0097] For example, in some specific implementations, the vectorization of media asset content can be achieved as follows:

[0098] Media assets are primarily determined by the following attributes: content title, director, actors, release date, publication region, and description. Different vectorization strategies need to be applied to these different attributes.

[0099] 1. Content name and description: Use a pre-trained BERT model.

[0100] 2. The publication year is always 4 digits long, but it can be divided into stages based on the years. For example, the period before the 1990s is divided into one category, 1990 to 2000 into another, 2000 to 2010 into another, 2010 to 2020 into another, and 2020 to the present into another.

[0101] 3. Directors and actors are all names and do not carry semantic information. First, extract all the name information and encode all the cast and crew members in binary to represent the personnel. If the length is insufficient, pad with 0.

[0102] It should be noted that user attributes include gender parameters, age parameters, geographic parameters, and user habit data. In some embodiments, as shown in Figure 4, the user attributes undergo a second vectorization process to obtain a user vector corresponding to each user attribute. This may include the following steps: S221, obtaining the fourth vectorization result corresponding to the gender parameter based on a preset second mapping table; wherein, the second mapping table includes vector gender results corresponding to different gender parameters; S222, performing binary encoding on the age parameter to obtain the fifth vectorization result; S223, determining the straight-line distance between the user's geographic region and the target geographic region based on the geographic parameter, converting the straight-line distance into a binary number to obtain the sixth vectorization result; S224, inputting the user habit data into a pre-trained language representation model for processing to obtain the seventh vectorization result; the user habit data includes viewing rate, viewing duration, and number of views; wherein, the user vector includes the fourth, fifth, sixth, and seventh vectorization results.

[0103] For example, in some specific implementations, the initial vectorization rules for user-related attributes are as follows: gender parameter uses a vector of all 0s or all 1s to represent male and female, age is represented in binary, and if the length is insufficient, it is padded with 0s, region is represented by the binary number of the straight-line distance from Beijing, and user habit data uses the BERT model word vector model. The above is just a simple example of user attributes. In fact, users can be described in more dimensions according to business needs, such as viewing rate, viewing duration, and number of views, etc.

[0104] S300. Based on the corresponding media asset content and user attributes, organize the content data; combine and concatenate the corresponding media asset vector and user vector to obtain the vector data corresponding to each content data.

[0105] It should be noted that the user attributes include the number of times the user has viewed the media asset content; in some embodiments, as shown in Figure 5, combining and concatenating the corresponding media asset vector and user vector may include the following steps: S301, obtaining the binary representation corresponding to the number of views; S302, filling the media asset vector, user vector and binary representation with the value 0 based on a preset length; S303, concatenating the filled media asset vector, user vector and binary representation in sequence.

[0106] For example, in some specific implementations, the media asset vector and the user vector are combined in the following way: media asset vector + user vector that viewed the media asset content + binary representation of the number of times the user viewed the media asset. The length is set according to the media asset content vector and the user vector, and zeros are padded if necessary. In other words, the media asset content vector is now not only described based on the media asset content itself, but also incorporates user behavior information.

[0107] S400. Input the vector data into the preset classification model for classification processing to obtain the probability distribution corresponding to the vector data;

[0108] The classification model includes convolutional layers, pooling layers, fully connected layers, and an output layer; the probability distribution includes the class probability of each class, the sum of all class probabilities in the probability distribution is 1, and the highest class probability is the predicted classification of the vector data;

[0109] It should be noted that in some embodiments, inputting vector data into a preset classification model for classification processing to obtain the probability distribution corresponding to the vector data may include the following steps: extracting features from the vector data through a convolutional layer to obtain feature data; reducing the dimensionality of the feature data through a pooling layer to obtain feature information; integrating and mapping the feature information through multiple fully connected layers to obtain the classification result; wherein a preset classification function is embedded in the output layer; and the classification result is transformed into a probability distribution through the classification function of the output layer.

[0110] For example, in some specific embodiments, the present invention can use a convolutional neural network (CNN) as the main classification model architecture, which can effectively extract features from the input data. The vectorized content attribute data is used as input to the CNN, and features at different levels are extracted step by step through multiple convolutional and pooling layers.

[0111] Its main layers are as follows:

[0112] Convolutional layers: Convolutional kernels of different sizes are used to capture features at different scales. For example, for vectors containing content names and descriptions, smaller convolutional kernels can be used to extract local semantic features; for binary-coded director and actor information, larger convolutional kernels can be used to capture the overall combination of personnel features.

[0113] Pooling layer: Employs max pooling or average pooling operations to reduce feature dimensionality while retaining important feature information.

[0114] Fully connected layers: After convolutional and pooling layers, several fully connected layers are connected to further integrate and map the extracted features to obtain the final classification result.

[0115] In the output layer of the model, the softmax function can be used as the classification function. The softmax function transforms the model's output values ​​into a probability distribution, such that each output value is between 0 and 1, and the sum of all output values ​​is 1. Thus, for a given on-demand content, the model can output the probability of it belonging to each category, and the category with the highest probability is the predicted classification of that content.

[0116] S500: Based on probability distribution processing, obtain the information entropy corresponding to each content data, and use the information entropy to construct content pairs based on the content data; each content pair includes two content data.

[0117] It should be noted that, in some embodiments, obtaining the information entropy corresponding to each content data based on probability distribution processing may include the following steps:

[0118] The information entropy corresponding to the content data is obtained by performing a preset accumulation operation on the probabilities of all categories in the probability distribution;

[0119] The expression for information entropy is:

[0120] In the formula, H(X) represents the information entropy corresponding to content data X; n represents the total number of categories; i represents the i-th category; p(x) i ) represents the category probability of the i-th category.

[0121] It should be noted that, in some embodiments, constructing content pairs based on content data using information entropy may include the following steps: constructing initial content pairs for all combinations based on all content data, grouping them into pairs; determining the entropy value of each initial content pair based on the sum of the information entropy values ​​of the two content data in the initial content pair; and using the initial content with the largest preset number of entropy values ​​as the final constructed content pair; wherein, the content pairs include content pairs of the same type and content pairs of different types.

[0122] For example, in some implementations, randomly constructing content pairs to query a large model can be very inefficient because many content pairs may contain no information. To efficiently utilize a large model, content pairs can be constructed based on information entropy.

[0123] The formula for information entropy is:

[0124] In the formula, H(X) represents the information entropy corresponding to content data X; n represents the total number of categories; i represents the i-th category; p(x) i ) represents the category probability of the i-th category.

[0125] When constructing content pairs, use the content pairs with the higher entropy values ​​as input to the large model.

[0126] S600: Classify and discriminate the content against the input pre-defined large model to obtain the discrimination result;

[0127] The discrimination result indicates whether the two content data in the content pair are of the same type;

[0128] For example, in some specific implementations, content pairs can be organized according to a preset text template and input into a large model. This invention introduces feedback from the large model to the small model. This invention proposes using the feedback information from the large model to continuously correct the media asset similarity calculation results of the small model. This is mainly achieved through the following three stages:

[0129] The first stage is the similarity asset discrimination task, which aims to optimize the small model's ability to identify similar media assets. Information about two media assets, <Media Asset 1, Media Asset 2 similar to Media Asset 1>, is input into the large model, which then determines whether the two media assets belong to the same category. A specific example is provided below:

[0130] [Please answer whether the media assets “Content: Palace Drama 1, User A attributes: 20 years old, female, Shandong, active time period: 8 PM to 1 AM, User A has watched it 100 times, ......” and “Content: Palace Drama 2, User B attributes: 23 years old, female, Shandong, active time period: 8 PM to 1 AM, User B has watched it 100 times, ......” can be classified as similar content. Output 1 if correct, 0 if not.]

[0131] The second stage is a dissimilarity discrimination task, <Media Asset 1, Media Asset 3 which is dissimilar to Media Asset 1>, with the aim of optimizing the small model's judgment of media assets that are not of the same class. Here is a specific example:

[0132] [Please answer whether the media assets “Content: Palace Drama, User A: 20 years old, female, Shandong, active time period: 8 PM to 1 AM, User A has watched it 100 times, ......” and “Content: War Drama, User C: 73 years old, female, Shandong, active time period: 7 PM to 9 PM, User C has watched it 30 times” are the same type of media asset. You can output 1, not 0.]

[0133] The above task questions are input into the large model. The large model answers the questions based on its own perspective or by combining information from the internet. After receiving the answers, it returns the results to the small model, allowing the small model to continuously refine and optimize its performance. Each round can be configured to distinguish between 100 pairs of similar content and 100 pairs of different content. Training can stop when the error between the model's output and the large model's judgment results is less than 5% after N consecutive rounds, indicating that the recommendation model has been successfully trained. N can be determined based on the size of the training set.

[0134] S700. Based on the discrimination results of all content pairs and the probability distribution of all vector data, the prediction accuracy of the classification model is obtained;

[0135] For example, in some specific application scenarios, such as when a large model determines that the content data corresponding to the first media asset and the content data corresponding to the second media asset are the same type of content, but the classification model makes different predictions for the first media asset and the second media asset, the classification results of the classification model for the first media asset and the second media asset can be marked as inaccurate, and then all results can be counted to achieve the prediction accuracy.

[0136] S800. When the prediction accuracy is less than the preset threshold, the model parameters of the classification model are adjusted using the loss function. Then, the process of inputting the vector data into the preset classification model for classification processing and obtaining the probability distribution corresponding to the vector data is repeated until the prediction accuracy is greater than or equal to the preset threshold. The adjusted classification model is then used as the target model.

[0137] The target model is used to classify the media content of IPTV sets, so that IPTV sets can cluster the media content based on the classification results.

[0138] For example, in some specific implementations, the questions constructed based on the aforementioned content pairs can be input into a large model. The large model will answer the questions based on its own perspective or by combining information from the internet. After receiving the answers, it returns the results to a small model, allowing the small model to continuously refine and optimize its performance. Each round can be configured with 100 pairs of similar content and 100 pairs of dissimilar content. Training can stop when the error between the model's output and the large model's judgment results is less than 5% after N consecutive rounds, indicating that the recommendation model has been successfully trained. N can be determined based on the size of the training set.

[0139] In some specific application scenarios, the cross-entropy loss function can be used. Cross-entropy loss is commonly used in classification problems to measure the difference between the probability distribution predicted by the model and the true labels. For IPTV on-demand content classification tasks, each piece of content has a specific category label, and the model's output is the predicted probability for each category. The cross-entropy loss function prompts the model to continuously adjust its parameters, making the predicted probability distribution as close as possible to the true category label distribution.

[0140] To explain in detail the principle of the technical solution of the present invention, the overall process of the present invention will be described below with reference to some specific embodiments. It is easy to understand that the following is an explanation of the technical principle of the present invention and should not be regarded as a limitation of the present invention.

[0141] First, it's important to clarify that, given the limitations of existing technologies, designing and implementing a method that adapts to the characteristics of IPTV data while effectively improving the granularity and increasing the dimensions of classification to better support the development of IPTV large-screen video services has become a pressing issue in the current development of IPTV services. This invention is proposed against this backdrop, aiming to provide IPTV platforms with a more efficient and accurate media asset clustering scheme by combining diverse media asset data with user behavior data, thereby better supporting IPTV's search and recommendation functions.

[0142] The development of large-scale models is currently very rapid. Considering the excellent performance they have demonstrated in various fields, this invention did initially consider using large-scale models. However, it's impossible to completely transmit users' historical behavior data to large-scale models for analysis. Furthermore, the services currently provided by large-scale models are all implemented using APIs. Even if all users' historical behavior data were input into the large-scale model, it would exceed the token limits of the model. Therefore, large-scale models cannot directly serve as "decision-makers" in recommendation systems. Therefore, this invention considers leveraging the capabilities of large-scale models to participate in the training of smaller models, assisting in the optimization of the smaller models' training.

[0143] As shown in Figure 6, the technical solution of the present invention can be achieved as follows:

[0144] Media asset data is a very rich existing data resource. The first step of this invention is the vectorization of media asset content. Media assets mainly have the following attributes for judgment: content name, director, actors, release date, publication region, description, etc. Different vectorization strategies need to be adopted for these different attributes.

[0145] 1. Content name and description: Use a pre-trained BERT model.

[0146] 2. The publication year is always 4 digits long, but it can be divided into stages based on the years. For example, the period before the 1990s is divided into one category, 1990 to 2000 into another, 2000 to 2010 into another, 2010 to 2020 into another, and 2020 to the present into another.

[0147] 3. Directors and actors are all names and do not carry semantic information. First, extract all the name information and encode all the cast and crew members in binary to represent the personnel. If the length is insufficient, pad with 0.

[0148] The initial vectorization rules for user attributes are as follows: gender is represented by a vector of all zeros or all ones, age is represented in binary and padded with zeros if the length is insufficient, region is represented by a binary number representing the straight-line distance from Beijing, and user habit data is represented by a BERT model word vector model. The above is just a simple example of user attributes. In fact, users can be described in more dimensions according to business needs, such as viewing rate, viewing duration, and number of views.

[0149] Next, the media asset vector and user vector are combined in the following way: media asset vector + user vector that viewed the media asset + binary representation of the number of times the user viewed the media asset. The length is set according to the media asset vector and user vector, and zeros are added if necessary. In other words, the media asset vector is now not only described by the media asset content itself, but also incorporates user behavior information.

[0150] The next step is the network training phase, which this invention implements using a Convolutional Neural Network (CNN). As the main classification model architecture, CNN can effectively extract features from the input data. Vectorized content attribute data is used as input to the CNN, and features at different levels are extracted progressively through multiple convolutional and pooling layers.

[0151] Its main layers are as follows:

[0152] Convolutional layers: Convolutional kernels of different sizes are used to capture features at different scales. For example, for vectors containing content names and descriptions, smaller convolutional kernels can be used to extract local semantic features; for binary-coded director and actor information, larger convolutional kernels can be used to capture the overall combination of personnel features.

[0153] Pooling layer: Employs max pooling or average pooling operations to reduce feature dimensionality while retaining important feature information.

[0154] Fully connected layers: After convolutional and pooling layers, several fully connected layers are connected to further integrate and map the extracted features to obtain the final classification result.

[0155] In the output layer of the model, the softmax function can be used as the classification function. The softmax function transforms the model's output values ​​into a probability distribution, such that each output value is between 0 and 1, and the sum of all output values ​​is 1. Thus, for a given on-demand content, the model can output the probability of it belonging to each category, and the category with the highest probability is the predicted classification of that content.

[0156] The loss function can be the cross-entropy loss function. Cross-entropy loss is commonly used in classification problems to measure the difference between the probability distribution predicted by the model and the true labels. For the classification task of IPTV on-demand content, each piece of content has a specific class label, and the model's output is the predicted probability for each class. The cross-entropy loss function prompts the model to continuously adjust its parameters, making the predicted probability distribution as close as possible to the true class label distribution.

[0157] This invention introduces a large model to provide feedback to a small model. It proposes using the feedback information from the large model to continuously correct the media asset similarity calculation results of the small model. This is mainly achieved through the following three stages:

[0158] The first stage is the similarity asset discrimination task, which aims to optimize the small model's ability to identify similar media assets. Information about two media assets, <Media Asset 1, Media Asset 2 similar to Media Asset 1>, is input into the large model, which then determines whether the two media assets belong to the same category. As shown in Figure 7, the following is a specific example:

[0159] [Please answer whether the media assets “Content: Palace Drama 1, User A attributes: 20 years old, female, Shandong, active time period: 8 PM to 1 AM, User A has watched it 100 times, ......” and “Content: Palace Drama 2, User B attributes: 23 years old, female, Shandong, active time period: 8 PM to 1 AM, User B has watched it 100 times, ......” can be classified as similar content. Output 1 if correct, 0 if not.]

[0160] The second stage is the dissimilarity discrimination task, <media asset 1, media asset 3 which is dissimilar to media asset 1>, with the aim of optimizing the small model's judgment of media assets that are not of the same class. As shown in Figure 8, the following is a specific example:

[0161] [Please answer whether the media assets “Content: Palace Drama x, User A: 20 years old, female, Shandong, active time period: 8 PM to 1 AM, User A has watched it 100 times, ......” and “Content: War Drama y, User C: 73 years old, female, Shandong, active time period: 7 PM to 9 PM, User C has watched it 30 times” are the same type of media asset. You can output 1, not 0.]

[0162] Randomly constructing content pairs to query a large model would be very inefficient, because many content pairs might lack information (for example, content 1 and content 3 belong to completely different categories). To efficiently utilize the large model, we construct them based on information entropy.

[0163] The formula for information entropy is:

[0164] Where: H(X) represents the information entropy corresponding to content data X; n represents the total number of categories; i represents the i-th category; p(x) i ) represents the category probability of the i-th category.

[0165] When constructing content pairs, use the content pairs with the higher entropy values ​​as input to the large model.

[0166] The above task questions are input into the large model. The large model answers the questions based on its own perspective or by combining information from the internet. After receiving the answers, it returns the results to the small model, allowing the small model to continuously refine and optimize its performance. Each round sets 100 pairs of similar content and 100 pairs of different content to distinguish. Training can stop when the error between the model's output and the large model's judgment results is less than 5% after N consecutive rounds, indicating that the recommendation model has been successfully trained. N can be determined based on the size of the training set.

[0167] In some specific implementations, the model training process of this invention can be implemented as follows:

[0168] Step 1: Data Preprocessing

[0169] First, the media asset data and user data are structured.

[0170] Step 2: Attribute Vectorization:

[0171] Based on the preprocessed data, media asset content data and user data are vectorized.

[0172] Step 3: Training the convolutional neural network for clustering:

[0173] The data from step two is used as input to a convolutional neural network, which then performs convolution to extract features and perform clustering.

[0174] Step 4: Analyze the framework content and input the large model to receive feedback:

[0175] Phase 1: Construct a binary task <Media Asset 1, Media Asset 2 which is similar to Media Asset 1>, query the large model to see if the two belong to the same class, and adjust the weights or parameters of the trained clustering model based on the feedback from the large model.

[0176] Phase 2: Construct the binary task <Media Asset 1, Media Asset 3 which is dissimilar to Media Asset 1 and of a different class>, as above, but Media Asset 3 is not of the same class as Media Asset 1.

[0177] Step Six: Iterate through the above steps:

[0178] Repeat the above steps until the accuracy rate reaches 95% in N rounds.

[0179] In summary, the main objective of this invention is to design and implement a multi-dimensional, fine-grained classification method for IPTV services, taking into account its unique media asset data and limited user behavior data. This method is not limited to classification by age (e.g., preferences of the elderly vs. children), by region (e.g., preferences of northerners vs. southerners), or by activity time (e.g., preferences of late-night viewers vs. early-morning viewers). This algorithm aims to improve the tagging capabilities of IPTV media assets, thereby enabling more refined searches during user searches and recommending more suitable content, thus increasing click-through rates. Through the implementation of this invention, it is hoped that without significantly increasing data collection costs, existing data resources can be fully utilized to achieve more accurate film and television content classification, promoting the innovative development of IPTV services in the context of big data and artificial intelligence technologies.

[0180] Specifically, this invention optimizes a media asset clustering algorithm based on collaborative user behavior, taking into account the data characteristics of the IPTV service environment. By combining media asset content and user data, the accuracy of clustering can be effectively improved. Furthermore, this invention innovatively introduces feedback information from a large model to correct the IPTV recommendation model. By constructing a binary task and querying the large model to determine whether the two are similar, this invention can adjust the weights or parameters of the recommendation model based on the feedback from the large model, thereby further optimizing the clustering effect.

[0181] Compared with the prior art, the present invention has at least the following beneficial effects:

[0182] 1. Efficiently utilize the combination of large and small models:

[0183] This invention cleverly integrates the capabilities of a large model into the training of a small model. By leveraging the feedback from the large model on media asset similarity, the classification performance of the small model is optimized. This approach not only avoids the high costs and resource consumption of directly using a large model for real-time recommendations but also fully utilizes the advantages of large models in feature extraction and semantic understanding.

[0184] Compared with traditional methods, the solution of this invention significantly improves the accuracy and refinement of classification while maintaining low data collection costs.

[0185] 2. Classify IPTV content in a multi-dimensional, fine-grained manner based on user behavior:

[0186] This invention breaks through the limitations of traditional classification based on media asset content itself, and realizes fine-grained classification based on multiple dimensions such as media asset content and user behavior data.

[0187] This multi-dimensional classification method can more accurately capture the subtle differences between media content and user needs, thereby improving the accuracy of the recommendation system and user satisfaction.

[0188] 3. Innovation in vectorization strategies:

[0189] For media asset data and user data with different attributes, this invention adopts different vectorization strategies, such as using a pre-trained BERT model to process content names and descriptions, and binary encoding of directors and actors.

[0190] These innovative vectorization strategies not only improve the representation of data, but also provide high-quality input for subsequent training of convolutional neural networks.

[0191] 4. Efficient model training and optimization:

[0192] This invention uses a convolutional neural network (CNN) as the main classification model architecture, and extracts features step by step through multiple convolutional layers and pooling layers to achieve accurate classification of media asset content.

[0193] During model training, this invention introduces the concept of information entropy, constructing high-information content as input to a large model for discrimination, thereby improving the efficiency of model training.

[0194] By continuously refining the media asset similarity calculation results of the small model, this invention ultimately achieves a high degree of consistency with the discrimination results of the large model, ensuring the accuracy and reliability of the recommendation system.

[0195] 5. Promote the innovative development of IPTV services:

[0196] The implementation of this invention will significantly enhance the user experience and competitiveness of IPTV services. Through more accurate film and television content categorization and recommendation, users can find their favorite content more quickly, improving viewing satisfaction and loyalty.

[0197] At the same time, this invention also provides strong technical support and a demonstration effect for the innovative development of IPTV services in the context of big data and artificial intelligence technologies.

[0198] In summary, this invention achieves significant innovation and improvement in IPTV multi-dimensional fine-grained classification methods, and has advantages and effects such as efficient resource utilization, multi-dimensional classification, vectorization strategy innovation, efficient model training and optimization, and promotion of business innovation and development.

[0199] On the other hand, as shown in Figure 9, an embodiment of the present invention provides a clustering device 900 for on-demand content, including:

[0200] The first module 901 is used to obtain all media asset content of the network TV and obtain the user attributes corresponding to each media asset content based on the playback data of the media asset content.

[0201] The second module 902 is used to perform a first vectorization process on the media asset content to obtain a media asset vector corresponding to each media asset content; and to perform a second vectorization process on the user attributes to obtain a user vector corresponding to each user attribute.

[0202] The third module 903 is used to organize content data based on the corresponding media asset content and user attributes; and to combine and concatenate the corresponding media asset vector and user vector to obtain the vector data corresponding to each content data.

[0203] The fourth module 904 is used to input vector data into a preset classification model for classification processing to obtain the probability distribution corresponding to the vector data;

[0204] The classification model includes convolutional layers, pooling layers, fully connected layers, and an output layer; the probability distribution includes the class probability of each class, the sum of all class probabilities in the probability distribution is 1, and the highest class probability is the predicted classification of the vector data;

[0205] Module 5, 905, is used to obtain the information entropy corresponding to each content data based on probability distribution processing, and to construct content pairs based on the content data using the information entropy; each content pair includes two content data.

[0206] Module 6, 906, is used to classify and discriminate the content pairs input to the pre-defined large model, and obtain the discrimination results; the discrimination results indicate whether the two content data in the content pair are of the same type;

[0207] Module 7, 907, is used to obtain the prediction accuracy of the classification model based on the discrimination results of all content pairs and the probability distribution of all vector data.

[0208] The eighth module 908 is used to adjust the model parameters of the classification model using the loss function when the prediction accuracy is less than the preset threshold. It then returns to the step of inputting the vector data into the preset classification model for classification processing to obtain the probability distribution corresponding to the vector data, until the prediction accuracy is greater than or equal to the preset threshold. The adjusted classification model is then used as the target model.

[0209] The target model is used to classify the media content of IPTV sets, so that IPTV sets can cluster the media content based on the classification results.

[0210] The content of the method embodiments of the present invention is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above methods.

[0211] On the other hand, embodiments of the present invention also provide an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned sensitive information method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0212] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0213] As shown in Figure 10, Figure 10 illustrates a specific example of the hardware structure of an electronic device 1000 according to an embodiment. The electronic device 1000 includes:

[0214] The processor 1001 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0215] The memory 1002 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1002 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1002 and is called and executed by the processor 1001 to execute the network node population optimization method of the embodiments of this invention.

[0216] Input / output interface 1003 is used to implement information input and output;

[0217] The communication interface 1004 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0218] Bus 1005 transmits information between various components of the device (e.g., processor 1001, memory 1002, input / output interface 1003, and communication interface 1004);

[0219] The processor 1001, memory 1002, input / output interface 1003 and communication interface 1004 are connected to each other within the device via bus 1005.

[0220] The electronic device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0221] The content of the method embodiments of the present invention is applicable to the embodiments of the present electronic device. The specific functions implemented by the embodiments of the present electronic device are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above methods.

[0222] Another aspect of this invention provides a computer-readable storage medium storing a program that is executed by a processor to implement the aforementioned method.

[0223] It should be noted that the computer-readable medium shown in the embodiments of the present invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD to ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, wherein computer-readable program code is carried. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0224] The content of the method embodiments of the present invention is applicable to the computer-readable storage medium embodiments. The specific functions implemented by the computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above methods.

[0225] This invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the aforementioned method.

[0226] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0227] It should be noted that although several modules for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0228] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, portable hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the method according to the embodiments of the present invention.

[0229] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented in this invention. Alternative embodiments are contemplated, in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0230] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0231] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0232] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution means, apparatus, or device (such as a computer-based device, a processor-including device, or other means that can fetch and execute instructions from, or in conjunction with, an instruction execution means, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution means, apparatus, or device.

[0233] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0234] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution device. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0235] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0236] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0237] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A clustering method for on-demand content, characterized in that, Includes the following steps: Obtain all media asset content of the network TV, and obtain the user attributes corresponding to each media asset content based on the playback data of the media asset content; The media asset content is subjected to a first vectorization process to obtain a media asset vector corresponding to each media asset content; the user attribute is subjected to a second vectorization process to obtain a user vector corresponding to each user attribute; Content data is obtained by organizing the media asset content and user attributes according to the corresponding relationship; the media asset vector and user vector of the corresponding relationship are combined and concatenated to obtain vector data corresponding to each content data; The vector data is input into a preset classification model for classification processing to obtain the probability distribution corresponding to the vector data; The classification model includes convolutional layers, pooling layers, fully connected layers, and an output layer; the probability distribution includes the class probability of each class, the sum of all class probabilities in the probability distribution is 1, and the largest class probability is the predicted classification of the vector data; Based on the probability distribution, the information entropy corresponding to each content data is obtained, and the information entropy is used to construct a content pair based on the content data; the content pair includes two content data. The content is input to a pre-defined large model for classification and discrimination to obtain a discrimination result; the discrimination result indicates whether the two content data in the content pair are of the same type; The prediction accuracy of the classification model is obtained based on the discrimination results of all the content pairs and the probability distribution of all the vector data. When the prediction accuracy is less than a preset threshold, the model parameters of the classification model are adjusted using a loss function, and the process of inputting the vector data into the preset classification model for classification processing to obtain the probability distribution corresponding to the vector data is repeated until the prediction accuracy is greater than or equal to the preset threshold. The adjusted classification model is then used as the target model. The target model is used to classify the media content of the network TV, so that the network TV can perform content clustering of the media content based on the results of the classification.

2. The clustering method for on-demand content according to claim 1, characterized in that, The media asset content includes content name and description, publication year, and cast / responsibility list; the first vectorization process for the media asset content to obtain the media asset vector corresponding to each media asset content includes the following steps: The content name and description are input into a pre-trained language representation model to obtain the first vectorization result; The publication year determines the time range in which the publication falls, and a second vectorization result corresponding to the time range is obtained based on a preset first mapping table; wherein, the first mapping table includes vector category results corresponding to each time range; All actors and staff members in the cast list are binary encoded to obtain a third vectorized result; The media asset vector includes the first vectorization result, the second vectorization result, and the third vectorization result.

3. The clustering method for on-demand content according to claim 1, characterized in that, The user attributes include gender parameters, age parameters, region parameters, and user habit data; the second vectorization processing of the user attributes to obtain a user vector corresponding to each user attribute includes the following steps: The fourth vectorized result corresponding to the gender parameter is obtained based on the preset second mapping table; wherein, the second mapping table includes vector gender results corresponding to different gender parameters; The age parameter is binary encoded to obtain the fifth vectorized result; The straight-line distance between the region and the target region is determined based on the regional parameters. The straight-line distance is then converted into a binary number to obtain the sixth vectorization result. The user habit data is input into a pre-trained language representation model for processing to obtain the seventh vectorization result; the user habit data includes viewing rate, viewing duration, and number of views. The user vector includes the fourth vectorization result, the fifth vectorization result, the sixth vectorization result, and the seventh vectorization result.

4. The clustering method for on-demand content according to claim 1, characterized in that, The user attributes include the number of times the user has viewed the media asset content; the step of combining and concatenating the corresponding media asset vector and user vector includes the following steps: Obtain the binary representation of the number of views; Based on a preset length, the media asset vector, the user vector, and the binary representation are padded with the value 0; The media asset vector, the user vector, and the binary representation after the padding process are sequentially concatenated.

5. The clustering method for on-demand content according to claim 1, characterized in that, The step of inputting the vector data into a preset classification model for classification processing to obtain the probability distribution corresponding to the vector data includes the following steps: Feature data is obtained by extracting features from the vector data through the convolutional layer; The feature data is reduced in dimensionality using the pooling layer to obtain feature information; The feature information is integrated and mapped through multiple fully connected layers to obtain the classification result; The output layer contains a pre-defined classification function. The classification result is transformed into the probability distribution through the classification function of the output layer.

6. The clustering method for on-demand content according to claim 1, characterized in that, The process of obtaining the information entropy corresponding to each content data based on the probability distribution includes the following steps: The information entropy corresponding to the content data is obtained by performing a preset accumulation operation on all the category probabilities in the probability distribution; The expression for the information entropy is: In the formula, H(X) represents the information entropy corresponding to content data X; n represents the total number of categories; i represents the i-th category; p(x) i ) represents the category probability of the i-th category.

7. The clustering method for on-demand content according to claim 1, characterized in that, The process of constructing content pairs based on the content data using the information entropy includes the following steps: Based on all the content data, initial content pairs for all combinations are constructed in pairs. The entropy value of each initial content pair is determined based on the sum of the information entropies corresponding to the two content data in the initial content pair; The initial content with the largest entropy value is used as the final content pair obtained by construction. The content pairs include pairs of similar content and pairs of different content.

8. A clustering device for on-demand content, characterized in that, include: The first module is used to obtain all media asset content of the network TV and obtain the user attributes corresponding to each media asset content based on the playback data of the media asset content; The second module is used to perform a first vectorization process on the media asset content to obtain a media asset vector corresponding to each media asset content; and to perform a second vectorization process on the user attributes to obtain a user vector corresponding to each user attribute. The third module is used to organize content data according to the media asset content and user attributes of the corresponding relationship; and to combine and concatenate the media asset vector and user vector of the corresponding relationship to obtain vector data corresponding to each content data. The fourth module is used to input the vector data into a preset classification model for classification processing to obtain the probability distribution corresponding to the vector data; The classification model includes convolutional layers, pooling layers, fully connected layers, and an output layer; the probability distribution includes the class probability of each class, the sum of all class probabilities in the probability distribution is 1, and the largest class probability is the predicted classification of the vector data; The fifth module is used to process the probability distribution to obtain the information entropy corresponding to each content data, and to construct a content pair based on the content data using the information entropy; the content pair includes two content data. The sixth module is used to classify and discriminate the content pairs into a preset large model to obtain a discrimination result; the discrimination result indicates whether the two content data in the content pair are of the same type. The seventh module is used to obtain the prediction accuracy of the classification model based on the discrimination results of all the content pairs and the probability distribution of all the vector data; The eighth module is used to adjust the model parameters of the classification model using a loss function when the prediction accuracy is less than a preset threshold, and then return to the step of inputting the vector data into the preset classification model for classification processing to obtain the probability distribution corresponding to the vector data, until the prediction accuracy is greater than or equal to the preset threshold, and then use the adjusted classification model as the target model. The target model is used to classify the media content of the network TV, so that the network TV can perform content clustering of the media content based on the results of the classification.

9. An electronic device, characterized in that, Including the processor and memory; The memory is used to store programs; The processor executes the program to implement the method as described in any one of claims 1 to 7.

10. A computer storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to implement the method as described in any one of claims 1 to 7.

Images