Recommendation Model Training With Data Augmentation for Cold Start
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Recommendation systems struggle to recommend virtual experiences to new users or for new experiences, often favoring established experiences over newer ones, leading to less frequent recommendations of newer experiences.
Innovation Solution
A method is employed to train a machine-learning model using data augmentation techniques, generating augmented training examples by modifying user or item features, and applying a loss function that minimizes similarity within examples and maximizes similarity between original and augmented examples, to enhance recommendations for new users or experiences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If recommendation systems prioritize established experiences, then recommendation reliability improves, but newer experiences become less frequently recommended
Solution Approach 1:
The patent applies data augmentation techniques that modify training data parameters (masking features, reordering sequences, cropping subsequences) to change how the model processes information. This allows the system to learn robust representations that work reliably for both established and new experiences, resolving the contradiction between recommendation reliability and diversity.
Solution Approach 2:
The patent performs preliminary data augmentation and representation learning before actual recommendation. By pre-training with augmented data that includes masked and reordered features, the model develops adaptive capabilities in advance, enabling it to handle both familiar and novel experiences effectively when making recommendations.
2Adaptability or versatility
If data augmentation is applied to training examples, then model adaptability improves, but training complexity increases
Solution Approach 1:
The patent segments the training process into distinct augmentation operations (masking, reordering, cropping) that can be applied independently to different features. This modular approach allows the system to achieve high adaptability through multiple simple transformations rather than one complex transformation, reducing overall training complexity while maintaining model versatility.
Data Source
AI summary
A computer-implemented method to train a machine-learning model to recommend virtual experiences to a user. The method includes receiving training data that includes original training examples corresponding to a set of virtual experiences, where individual training examples comprise user features and item features. The method further generating augmented training examples by modifying one or more of a user feature or an item feature from corresponding original training examples. The method further extracting respective representation embeddings from the original training examples and the augmented training examples. The method further includes determining a loss function such that: a first loss between representation embeddings of individual original training examples and representation embeddings of corresponding augmented training examples is minimized and a second loss between the representation embeddings of individual original training examples and representation embeddings of non-corresponding augmented training examples is maximized.