AI-based IPTV content recommendation method

Abstract

The invention provides an AI-based IPTV (Internet Protocol Television) content recommendation method, which comprises the following steps of: performing classified management according to a content library of an IPTV platform to obtain a content label database; preference information of a user is obtained and preprocessed; wherein the preference information comprises basic information and watching history information; a basic recommendation model is constructed, the basic recommendation model is trained through the preprocessed preference information, and a recommendation model is obtained after training is completed; generating personalized recommendation content through the recommendation model and the content label database; a trained recommendation system is deployed in a recommendation system of an IPTV platform, so that the IPTV platform can use a recommendation model to generate personalized recommendation content and design a user-friendly recommendation content display interface according to the watching history and preference of a user, and the visibility and attraction of the recommendation content are ensured; the problem that an operator spends a lot of manpower, material resources and financial resources to deploy the IPTV content which is virtualized and is not willing to use, and resource waste is caused is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of content recommendation, and particularly to an AI-based IPTV content recommendation method. Background Art

[0002] With the development of Internet technology, people's entertainment methods tend to be diversified. Among them, IPTV (Internet Protocol Television) and online video platforms are both popular entertainment methods related to video content nowadays. Now, in the current fast-paced life, people are more inclined to find their favorite video content on online platforms. As a result, the deployment of IPTV content by operators with a large amount of manpower, material resources, and financial resources is in vain, and no one is willing to use it, resulting in a waste of resources. The reason is that when the existing IPTV makes content recommendations, the recommended content is uniformly arranged by the operator and cannot be recommended according to each person's preferences, resulting in poor quality of the content recommended by IPTV and low matching degree with the user's interests. Summary of the Invention

[0003] The present invention provides an AI-based IPTV content recommendation method to solve the problem that IPTV cannot recommend content according to each person's preferences, resulting in poor quality of the content recommended by IPTV and low matching degree with the user's interests.

[0004] According to an embodiment of the present invention, an AI-based IPTV content recommendation method includes:

[0005] Obtaining a content tag database obtained from the content library of the IPTV platform;

[0006] Obtaining the preference information of the user and performing preprocessing; wherein, the preference information includes: basic information and viewing history information;

[0007] Constructing a basic recommendation model and training the basic recommendation model with the preprocessed preference information, and obtaining a recommendation model after the training is completed;

[0008] Generating personalized recommendation content through the recommendation model and the content tag database.

[0009] As an embodiment of the present invention, the basic recommendation model is any one of a collaborative filtering model, a DeepFM model, and a matrix factorization model.

[0010] As an embodiment of the present invention, it further includes:

[0011] Periodically obtaining the viewing history information of the user and calculating the click-through rate to optimize the recommendation model.

[0012] As an embodiment of the present invention, the method for optimizing a recommendation model by periodically obtaining the viewing history information of a user and calculating the click-through rate includes:

[0013] Obtain the viewing history information of the user, and calculate the click-through rate according to the viewing history information. The calculation formula of the click-through rate is as follows:

[0014]

[0015] where P is the click-through rate, k is a constant, N is the number of recommendations, and n is the number of clicks;

[0016] Determine whether the click-through rate is less than the preset click-through rate; if it is less, update the recommendation model according to the viewing history information and basic information of the user.

[0017] As an embodiment of the present invention, the method for optimizing a recommendation model by periodically obtaining the viewing history information of a user and calculating the click-through rate further includes:

[0018] Obtain the update time of the recommendation model since the last optimization, and determine whether the update time is greater than the preset update time; if the update time is greater than the preset update time, update the recommendation model according to the viewing history information and basic information of the user.

[0019] Compared with the prior art, the present invention has the following beneficial effects: By deploying the trained recommendation system into the recommendation system of the IPTV platform, the IPTV platform can generate personalized recommendation content using the recommendation model according to the viewing history and preferences of the user, and design a user-friendly recommendation content display interface to ensure the visibility and attractiveness of the recommendation content; it solves the problem that the deployment of IPTV content by operators, which costs a large amount of manpower, material resources and financial resources, is in vain and no one is willing to use it, resulting in a waste of resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is the overall architecture diagram of the embodiment of the present invention;

[0021] Figure 2 is the flowchart of another embodiment of the present invention;

[0022] Figure 3 is the architecture diagram for obtaining preference data of another embodiment of the present invention;

[0023] Figure 4 is the schematic diagram for preprocessing preference data and constructing a recommendation model of another embodiment of the present invention;

[0024] Figure 5 is the schematic diagram for training and deploying a recommendation model of another embodiment of the present invention;

[0025] Figure 6 This is the overall architecture diagram of another embodiment of the present invention. Specific implementation manner

[0026] The technical solutions in the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0027] As Figures 1 to 6 shown, an AI-based IPTV content recommendation method is proposed in an embodiment of the present invention, including:

[0028] Obtain a content tag database obtained from the content library of the IPTV platform;

[0029] Obtain the preference information of the user and perform preprocessing; wherein, the preference information includes: basic information and viewing history information;

[0030] Construct a basic recommendation model, and train the basic recommendation model with the preprocessed preference information. After the training is completed, a recommendation model is obtained;

[0031] Generate personalized recommended content through the recommendation model and the content tag database.

[0032] The working principle of the above technical solution: In the actual use process, by classifying and managing the content library of the IPTV platform, detailed tags are created for each content, such as director, actor, theme, type, duration, region, etc., and a content tag database is established; at the same time, when the user registers for the IPTV service, the basic information is filled in, and the basic information such as the user's gender, age, region, occupation, etc. is collected, and the user ID is used as the unique identifier to bind to the user's IPTV account; then, using the log system of the IPTV platform, the viewing history information of the user is recorded, including the program name, start time of viewing, end time of viewing, etc., and information such as viewing duration, viewing period, content type preference, user behavior information, etc. is extracted from it; wherein, the user behavior information includes search records and rating feedback on the content, etc.; then the basic recommendation model is trained with the basic information and the viewing history information, and a recommendation model is obtained after the training; recommended content is generated through the recommendation model and the content tag database;

[0033] Among them, the process of preprocessing the preference data is to perform data cleaning on the collected user data, including filling in missing values, detecting and processing outliers to remove invalid data; applying denoising techniques, such as wavelet transform or filters, to remove the noise in the data; formatting the data, converting it into data records in JSON format, ensuring that each data record contains unified fields and structures for subsequent processing;

[0034] Among them, the content tag database is obtained by classifying and managing the content library of the IPTV platform;

[0035] Beneficial effects of the above technical solution: Through the above technical solution, by deploying the trained recommendation system into the recommendation system of the IPTV platform, the IPTV platform can generate personalized recommendation content using the recommendation model based on the user's viewing history and preferences, and design a user-friendly recommendation content display interface to ensure the visibility and attractiveness of the recommendation content; it solves the problem that the deployment of IPTV content by operators, which costs a large amount of manpower, material resources, and financial resources, is in vain and no one is willing to use it, resulting in a waste of resources.

[0036] In one embodiment, the basic recommendation model is any one of a collaborative filtering model, a DeepFM model, and a matrix factorization model.

[0037] Working principle of the above technical solution: In the actual use process, the basic recommendation module includes but is not limited to any one of a collaborative filtering model, a DeepFM model, and a matrix factorization model. When the basic recommendation model is the DeepFM model, the process of constructing the DeepFM model is as follows: The first step is to select the model architecture: Design the DeepFM model architecture, which combines the advantages of the Factorization Machine (FM) and the Deep Neural Network (DNN) to process high-dimensional sparse data and capture complex interactions between features; The second step is to construct the feature input layer: According to the collected user information and content information, construct the feature input layer; This layer will receive the original one-dimensional feature vectors and convert them into a format that the model can process; The third step is to construct the embedding layer: For categorical features (such as gender, region, occupation, etc.), map them to dense vectors in a low-dimensional space through the embedding layer. Each feature has an embedding matrix, and the dimension of the matrix is determined by the number of features and the dimension of the embedding vector; The fourth step is to construct the FM component: In the FM component, first calculate the first-order linear combination of the features; Then calculate the vector inner product of the second-order cross combination of the feature pairs to capture the interaction between the features. Use the embedding vectors to represent the interaction of the feature pairs; The fifth step is to construct the DNN component: In the DNN component, use the concatenation of the embedding layer and the first-order linear features as the input, and learn the non-linear combination of the features through multiple hidden layers; Specifically, its detailed steps are as follows: Initialize the embedding matrix and the weights of the DNN network; Define the embedding layer to map each categorical feature to its corresponding embedding vector; In the FM part, calculate the first-order feature weights and the second-order feature interactions; Perform the second-order interaction calculation of FM. For example, for each feature pair (i, j), calculate the inner product S of their embedding vectors, and the calculation formula is as follows:

[0038]

[0039] Among them, a is the dimension of the embedding vector; in the DNN part, the embedding vector and the first-order linear features are concatenated as the input of the DNN; the DNN is constructed through multiple fully-connected layers (each layer followed by an activation function such as ReLU); Step 6, model output layer: The outputs of the FM component and the DNN component are concatenated and then input into the output layer, which is usually a single neuron, and a logistic function (such as Sigmoid) is used to output the final prediction result; Step 7, model optimization: Select a loss function such as binary cross-entropy loss for optimization during the training process; use an optimization algorithm such as Adam to update the model weights;

[0040] Meanwhile, after the model is constructed, the constructed model is trained with preference data, that is, the data set is divided into a training set, a validation set, and a test set to evaluate the generalization ability of the model; then, the DeepFM model is trained using the training set, and the model weights are updated through an optimization algorithm (such as Adam); the cross-validation method (such as k-fold cross-validation) is applied to evaluate the accuracy of the model on the validation set; finally, the model parameters are adjusted to obtain the best performance on the validation set.

[0041] Beneficial effects of the above technical solutions: By constructing a recommendation model through the above model, and then recommending the content of the IPTV platform through the recommendation model, the accuracy of the recommendation is improved: through deep learning algorithms, the user preferences can be understood more accurately, and the relevance of the recommended content can be improved; the user experience is enhanced: personalized recommendations can reduce the time for users to search through a large amount of content and enhance the user experience; the user stickiness is increased: by providing content that users are interested in, the viewing time of users and the platform stickiness are increased; the resource allocation is optimized: intelligent recommendations help content providers optimize resource allocation and improve the exposure rate and viewing rate of content; the platform revenue is increased: increasing the user playback duration and usage time can significantly increase the revenue of the platform's film and television memberships, advertisements, etc.

[0042] In one embodiment, it further includes:

[0043] Periodically obtain the user's viewing history information and calculate the click-through rate to optimize the recommendation model;

[0044] Periodically obtaining the user's viewing history information and calculating the click-through rate to optimize the recommendation model includes:

[0045] Obtain the user's viewing history information, calculate the click-through rate according to the viewing history information, and the calculation formula of the click-through rate is as follows:

[0046]

[0047] Among them, P is the click-through rate, k is a constant, N is the number of recommendations, and n is the number of clicks;

[0048] Determine whether the click-through rate is less than the preset click-through rate; if it is less, update the recommendation model according to the user's viewing history information and basic information;

[0049] Periodically obtain the user's viewing history information and calculate the click-through rate, and optimize the recommendation model, further including:

[0050] Obtain the update time of the recommendation model since the last optimization, and determine whether the update time is greater than the preset update time; if the update time is greater than the preset update time, update the recommendation model according to the user's viewing history information and basic information;

[0051] The working principle of the above technical solution: In the actual use process, real-time monitor and record the user's click, viewing, rating and other behavior data, and update the user's viewing history data through the behavior data; then periodically obtain the user's viewing history data, and calculate the click-through rate (CTR) according to the viewing history information, that is, the ratio of the number of clicks on the recommended content to the number of recommendations; then determine whether the click-through rate is less than the preset click-through rate; if it is less, update the recommendation model according to the user's viewing history information and basic information; at the same time, obtain the update time of the recommendation model since the last optimization, and determine whether the update time is greater than the preset update time; if the update time is greater than the preset update time, update the recommendation model according to the user's viewing history information and basic information; among them, updating the recommendation model is to retrain the recommendation model with the updated user viewing history data and basic information, and the training process is the same as the training process of the basic recommendation model;

[0052] The beneficial effects of the above technical solution: By integrating user feedback data into the feedback loop of the recommendation system for continuous training and optimization of the model; at the same time, regularly evaluate the CTR index, analyze the performance of the recommendation system, and make corresponding adjustments and optimizations according to the results; enable the recommendation model to be more personalized to the user for recommendation, thereby improving the data usage rate on the IPTV platform and avoiding resource waste.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An AI-based IPTV content recommendation method, characterized in that: include: Acquire a content tag database obtained from a content library of an IPTV platform; Obtaining user preference information and preprocessing it; wherein the preference information includes: basic information and viewing history information; Build a basic recommendation model and train it with the preprocessed preference information. After the training is completed, the recommendation model is obtained. Generate personalized recommendation content through recommendation model and content label database.

2. The AI-based IPTV content recommendation method according to claim 1, characterized in that: The basic recommendation model is any one of the collaborative filtering model, the DeepFM model and the matrix decomposition model.

3. The AI-based IPTV content recommendation method according to claim 1, characterized in that: Also includes: Periodically obtain users' viewing history information and calculate click-through rates to optimize the recommendation model.

4. The AI-based IPTV content recommendation method according to claim 3, characterized in that: Periodically obtain the user's viewing history information and calculate the click-through rate to optimize the recommendation model, including: Get the user's viewing history information and calculate the click-through rate based on the viewing history information. The calculation formula for the click-through rate is as follows: Among them, P is the click-through rate, k is a constant, N is the number of recommendations, and n is the number of clicks; Determine whether the click-through rate is less than a preset click-through rate; if so, update the recommendation model based on the user's viewing history information and basic information.

5. The AI-based IPTV content recommendation method according to claim 3, characterized in that: Periodically obtain the user's viewing history information and calculate the click-through rate to optimize the recommendation model, including: Obtain the update time of the recommendation model from the last optimization, and determine whether the update time is greater than the preset update time; if the update time is greater than the preset update time, update the recommendation model according to the user's viewing history information and basic information.

Images