Tuning Large Language Models for Recommendation Tasks

[reachsumit]

https://blog.reachsumit.com/posts/2023/05/tuning-llm-for-recsys/

[My3VM]

This is really a great compilation that I have ever found about LLMs for Recommendation. I have a question though on the target candidates selection process from the repository with large datasets. Without an explicit search query how can one arrive with candidate lists in case of recommendations? Perhaps, the summary of user preferences are queried against the vector store to arrive at candidate list?

[reachsumit]

@My3VM This is really a great compilation that I have ever found about LLMs for Recommendation. I have a question though on the target candidates selection process from the repository with large datasets. Without an explicit search query how can one arrive with candidate lists in case of recommendations? Perhaps, the summary of user preferences are queried against the vector store to arrive at candidate list?

Depending on the nature of the recommendation task, you may have implicit or explicit user feedback that you can use as your ground truth. For example, past user ratings for a rating prediction task, and user-item interaction history for a sequential recommendation task can be used as the label.