Exploring the Upper Limits of Text-Based Collaborative Filtering Using Large Language Models: Discoveries and Insights
Primary Area: transfer learning, meta learning, and lifelong learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Recommendation sytem, Large language models, Text-based collaborative filtering, GPT-3
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Text-based collaborative filtering (TCF) has become the mainstream approach for text and news recommendation, utilizing text encoders, commonly referred to as language models (LMs), to represent items.
However, the current landscape of TCF models predominantly revolves around the utilization of small or medium-sized LMs. It remains uncertain what impact replacing the item encoder with one of the largest and most potent LMs, such as the 175-billion parameter GPT-3 model, would have on recommendation performance. Can we expect unprecedented results?
To this end, we conduct an extensive series of experiments aimed at exploring the performance limits of the TCF paradigm. Specifically, we progressively increase the size of item encoders from one hundred million to one hundred billion, revealing the scaling limits of the TCF paradigm. Furthermore, we investigate whether these extremely large LMs can enable a universal item representation for the recommendation task and revolutionize the traditional ID paradigm, which is considered a significant obstacle to developing transferable “one model fits all” recommendation models. Our study not only demonstrates positive results but also uncovers unexpected negative outcomes, illuminating the current state of the TCF paradigm within the community. These findings will evoke deep reflection and inspire further research on text-based recommendation models. Our code and datasets will be provided upon acceptance.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1750
Loading