Recommendation Model Training Using Corrected Preference Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing item recommendation models based on big data require extensive data processing, leading to long training times and inaccurate predictions due to infrequent updates when user preferences change.
Innovation Solution
A method for training a recommendation model that involves acquiring and correcting preference information using user and object features, generating training data based on these features, and performing iterative training to obtain a trained model that meets current user preferences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If extensive big data processing is performed to construct training data, then the recommendation model can capture user preferences, but the training time becomes relatively long
Solution Approach 1:
The patent segments the training data generation process into two stages: offline construction of user preference models using historical data, and online generation of training data using the learned models. This segmentation allows the computationally intensive part to be performed offline, while online training only requires lightweight data generation, significantly reducing training time without sacrificing prediction accuracy.
Solution Approach 2:
The patent performs preliminary action by pre-learning user preference models from historical data before the actual recommendation task. The user preference model captures user preferences in advance, so that during online training, the system only needs to generate training data based on the pre-learned model, avoiding the need to process all historical data again and reducing training time.
2Stability of the object's composition
If the recommendation model parameters are not updated frequently, then the model structure remains stable, but the prediction accuracy decreases when user preferences change over time
Solution Approach 1:
The patent implements dynamics by enabling the recommendation model to be updated frequently with new user behavior data. The online training process allows the model parameters to be dynamically adjusted based on current user preferences, while the pre-learned user preference models provide a stable foundation. This dynamic update mechanism ensures the model adapts to changing user preferences without compromising overall stability.
Solution Approach 2:
The patent incorporates feedback by using the output of the recommendation model to generate training data that feeds back into the model updating process. The trained model's predictions are used to generate new training data, which is then used to further train and refine the model, creating a feedback loop that continuously improves prediction accuracy while maintaining model stability through the pre-learned preference models.
3Quantity of substance
If a huge amount of data is processed for constructing training data, then the training data is comprehensive, but the complexity of data processing increases
Solution Approach 1:
The patent extracts the complex data processing tasks from the online training process and relocates them to the offline phase. The user preference models are learned from comprehensive historical data offline, while the online process only extracts and generates training data based on the pre-learned models. This extraction significantly reduces the complexity of online data processing while maintaining comprehensive training data coverage.
Solution Approach 2:
The patent introduces user preference models as intermediaries between the comprehensive historical data and the training data generation process. These intermediary models simplify the data processing by pre-processing and organizing the complex historical data into structured preference representations, making the subsequent training data generation much simpler and less complex.
Data Source
AI summary
A recommendation model training method includes: acquiring first preference information corresponding to users for a first object, the first object being an object on which the users have performed a designated operation; determining second preference information corresponding to users for a second object, the second object being an object on which the users have not performed a designated operation; correcting the first and second preference information based on preference features of users for an object and correlation features between objects and the users, to obtain third preference information of the users for the objects; generating training data based on the third preference information, a user attribute of a target user, an object attribute of a target object, and operation information of the target user for the target object; and performing iterative training on a target recommendation model based on the training data until a training condition is satisfied.