Generative models improve forum post retrieval using cross-platform data
Exploring Forum Post Retrieval with Generative Modeling
Information Retrieval
Summary
Finding relevant forum posts for users can be hard when there is little user data. The authors show that by training a large language model on related Facebook group data, and reusing semantic identifiers from other Facebook platforms, they can better match posts to users. This approach helps recommend posts in Facebook Forum without needing much specific Forum data. The paper also looks into which parts of their method make the biggest difference in performance.
What this means in practice
- •For social media platform engineers: Improve content recommendations in new social platform areas by transferring models and semantic IDs from established platforms.
- •For online community managers: Enhance user engagement by providing better forum post suggestions with generative models trained on broader community data.
Authors
Yang Li, Yaguang Liu, Heng Liu, Shengbo Guo, Samson Komo, Jane Kou, Yulian Zhou, Gang Yang, Shubhojeet Sarkar, Gaurav Chakravorty, Yujie Liu, Haipeng Chen, Yonghuan Yang, Deepti Chheda, Yamin Wang, Mike Plumpe, Rish Tandon
Abstract
Generative recommendation (GR) has emerged as an alternative to embedding-based retrieval, building on the success of generative models in language and vision. We are exploring GR on Facebook Forum, a standalone application for medium-to-heavy users of Facebook Groups. Because Forum is a new surface, its own interaction data are too sparse to train a GR model from scratch. We address this with transfer along two axes: we train on a broader corpus of Facebook Groups engagements rather than Forum sessions alone, and we reuse hierarchical, prefix-based semantic IDs (SIDs) learned from cross-platform Facebook Feed data instead of fitting a Forum-specific tokenizer. A 3B-parameter instruction-tuned language model is then supervised-fine-tuned to generate SIDs directly from user context. We systematically ablate the design choices that matter most in practice, including SID construction, the composition and length of user history, and the inclusion of user-profile features. Our results show that cross-platform SIDs transfer to a new recommendation surface, and offer practical guidance for teams deploying GR on real-world social platforms.