Assess applicability of FACE beyond collaborative filtering

Investigate the applicability of FACE to recommendation paradigms beyond collaborative filtering, including sequential recommendation, and determine whether mapping model-specific representations into pretrained large language model token descriptors preserves performance and interpretability in these settings.

Background

The current instantiation of FACE targets collaborative filtering models by mapping their embeddings into LLM token space and aligning them semantically. The authors have not tested the framework in other recommendation scenarios such as sequential recommendation.

Demonstrating applicability beyond collaborative filtering would establish the broader utility and generality of descriptor-based alignment with LLM token spaces across diverse recommendation tasks.

References

Besides, our framework is currently limited to the collaborative filtering domain, and its applicability to other recommendation scenarios (e.g., sequence recommendation) remains unexplored.

— FACE: A General Framework for Mapping Collaborative Filtering Embeddings into LLM Tokens  (2510.15729 - Wang et al., 17 Oct 2025) in Appendix: Limitations

Collaborative filtering signals are a well-established driver of recommendation quality , yet integrating them into prompt-based MLLM frameworks remains an open problem .

— Multimodal Graph Retrieval-Augmented Sequential Recommendation via Collaborative Filtering Paths  (2610.11228 - Setiadi et al., 8 Oct 2026) in Section 1, Introduction