Understanding User Preferences with AI
The ability to analyze user preferences based on their past behavior is essential for providing personalized recommendations. However, this task becomes more challenging when products do not have star ratings. Typically, past actions are interpreted as binary data, indicating whether a user has interacted with a certain object or not. To derive user preferences from this covert input, additional assumptions must be made.
The Problem with Assuming User Engagement
Assuming that users enjoy the content they engage with and dismiss the content that doesn’t grab their attention is not always accurate. Users may be unaware of certain products, leading to a lack of engagement. It is more plausible to assume that users ignore or don’t care about aspects they cannot interact with.
The Bayesian Personalized Ranking (BPR) Approach
Past studies have suggested that users tend to favor familiar products. This idea forms the basis of Bayesian Personalized Ranking (BPR), a technique for personalized recommendations. BPR transforms the data into a three-dimensional binary tensor called D, with the first dimension representing the users.
A recent Apple study introduced a variant of the BPR model called Sliced Anti-symmetric Decomposition (SAD). SAD is an implicit-feedback-based model for collaborative filtering, which adds an additional latent vector to each item. This new vector allows for the evaluation of relative preferences between items. By allowing the item vector’s values to exceed 1, SAD captures non-linear patterns in user preferences.
Parameter Estimation and Model Comparison
The research team developed a quick group coordinate descent method for parameter estimation in SAD. They demonstrated the efficacy of this method using a simulated study. Additionally, they compared SAD against seven alternative recommendation models using freely available resources. The updated SAD model incorporates previously ignored data and relationships between entities, resulting in more reliable and accurate recommendations.
While the researchers refer to collaborative filtering as implicit feedback, the applications of SAD are not limited to this type of data. Datasets with explicit ratings can also be integrated into the model, providing immediate information during model fitting.
For more information, refer to the research paper and the GitHub link. Credit goes to the researchers involved in this project. Don’t forget to join our ML SubReddit, Facebook community, Discord channel, and subscribe to our email newsletter for the latest AI research news and projects.
If you enjoy our work, you’ll love our newsletter. Subscribe now!
Dhanshree Shenwai is a Computer Science Engineer with experience in the FinTech industry. She has a keen interest in AI applications and is