Recommendation Model Updating with Local Parameter Sparsification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The rapid increase in data input for updating recommendation models leads to significant consumption of internal memory space, resulting in increased loading and prediction times.
Innovation Solution
A method and apparatus for updating recommendation models by acquiring real-time user and recommendation object data, generating model updating sample characteristics with multiple dimensions, and performing local sparsification on model parameters meeting a sparsification condition to reduce memory consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large amount of newly generated data is input into the recommendation model for updating, then the model accuracy and recommendation performance are improved, but the internal memory space consumption increases significantly
Solution Approach 1:
The patent extracts and processes only the essential features from the large amount of newly generated data through feature extraction and selection mechanisms. By identifying and retaining only the most relevant features that contribute to model accuracy, the system reduces the quantity of data stored in memory while maintaining recommendation performance.
Solution Approach 2:
The patent changes the parameters of data processing by transforming raw data into feature representations and adjusting the dimensionality and complexity of data structures. This involves converting high-dimensional raw data into lower-dimensional feature vectors and dynamically adjusting memory allocation based on feature importance, thereby reducing memory consumption while preserving model accuracy.
2Measurement precision
If the recommendation model is updated with more data, then the recommendation performance is improved, but the loading time and prediction time increase
Solution Approach 1:
The patent segments the data processing and model updating process into distinct stages: data ingestion, feature extraction, model updating, and prediction. By dividing the workflow and processing data in batches rather than loading all data at once, the system reduces the time required for model loading and prediction while maintaining recommendation performance.
Solution Approach 2:
The patent performs preliminary feature extraction and preprocessing actions on incoming data before it is used for model updating. By pre-processing and filtering data in advance to retain only essential features, the system reduces the computational burden during model loading and prediction, thereby decreasing loading time and prediction time while maintaining recommendation accuracy.
3Speed
If the model parameters are updated in real-time, then the recommendation timeliness is improved, but the model complexity and memory consumption increase rapidly
Solution Approach 1:
The patent applies partial updates to model parameters based on the importance and relevance of incoming features. Instead of updating all model parameters with every new data point, the system selectively updates only the most relevant parameters, thereby maintaining real-time update capability while preventing rapid increase in model complexity and memory consumption.
Solution Approach 2:
The patent implements dynamic model parameter management where the model structure and parameter count can change adaptively based on data characteristics. The system dynamically adjusts model complexity by adding or removing parameters based on feature importance and data distribution, allowing real-time updates while controlling model growth to prevent excessive complexity and memory consumption.
Data Source
AI summary
A method for updating a recommendation model based on artificial intelligence includes: acquiring sample user data and sample recommendation object data, the sample recommendation object data including data related to sample media objects; generating model updating sample characteristics according to the sample user data and the sample recommendation object data, the model updating sample characteristics including sub-characteristics of at least two characteristic dimensions; inputting the model updating sample characteristics into the recommendation model to update the recommendation model, the recommendation model updated performing recommendation on media objects, and performing local sparsification on model parameters meeting a model sparsification condition in response to determining that the corresponding model parameters of the sub-characteristics in the model updating sample characteristics meet the model sparsification condition in the process of updating the recommendation model.