RecFormer
"Text is all you need" — a Longformer that does sequential recommendation with zero item IDs.
About
RecFormer takes UniSRec's text-based idea to its logical endpoint: throw away item IDs entirely. Published at KDD 2023 by Jiacheng Li and collaborators (including Julian McAuley), it flattens each item's attributes — title, category, brand — into plain text with attribute keys, concatenates a user's history into one long sequence, and trains a Longformer (149M parameters) on it directly with bidirectional masking.
No ID embeddings, no separate text encoder bolted onto an ID model — the language model is the recommender. The paper reported that this pure-text approach beat ID-based sequential models on Amazon datasets and, more importantly, transferred zero-shot to new domains, since a new catalog is just new text. It is also refreshingly honest about the tradeoff: you pay Longformer-scale compute for what an ID lookup table used to do cheaply.
RecFormer sits at the junction of two trends — text-based recsys and LLM-for-recommendation — and it is one of the clearest demonstrations that a pretrained language model can be a recommender without any ID machinery at all.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Paper-reported (KDD 2023, arXiv 2305.13731): outperformed ID-based sequential baselines on Amazon Sports/Beauty/Toys datasets and showed zero-shot transfer to new domains
- No independent published benchmarks; no third-party leaderboard numbers found
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (TIER verified, metrics-v1). Evidence: peer-reviewed KDD 2023 paper ("Text Is All You Need", arXiv 2305.13731) with a clearly specified backbone (LongFormer 149M) and reported gains over ID-based baselines on Amazon datasets; author code release at github.com/JiachengLi1995/RecFormer. Strengths: clean, radical design with real zero-shot transfer results; influential in the LLM4Rec literature. Docked for: (1) all benchmark claims are author-reported with no independent replication found; (2) compute cost is high relative to ID-based models for comparable catalogs. License: open research code.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · KDD 2023 paper ('Text Is All You Need', arXiv 2305.13731) from UC San Diego: LongFormer (149M) doing sequential recommendation purely from item text, zero item IDs.
- · Reported gains over ID-based baselines on Amazon datasets plus zero-shot cross-domain transfer; author code/checkpoints released.
- · No independent replication found; compute cost high vs ID-based models for comparable catalogs.