Repair method improves ranking from compressed user preference states
Not All Is Lost: Repairing Lossy User Preference States of Personalization Encoders
Machine LearningInformation Retrieval
Summary
Personalization systems try to summarize a user’s likes and dislikes into a compact state to recommend items or generate text. But this compressed summary can miss some useful details stored in the system’s internal memory from past interactions. The authors introduce REPAIR, a technique that looks at these internal memories and fixes the compressed state by adding important missing information before making recommendations. This method improves recommendation accuracy across multiple datasets and works better than just adjusting the final ranking part. It also helps personalized text generation better reflect user preferences.
What this means in practice
- •For recommendation system engineers: Improve recommendation accuracy by correcting compressed user preference states using cached interaction representations without retraining the entire personalization model.
- •For personalized text generation developers: Enhance responsiveness of text generation systems to user preferences by integrating a repair module that supplements the compressed user state.
- •For advertising technology teams: Use state correction techniques to boost precision in targeting user preferences without needing to update recommendation models frequently.$Commercial implications: Enables more precise ad targeting models to be built and deployed with lower retraining costs, improving ad performance for advertisers.
Authors
Parthiv Chatterjee, Dhiraj Golhar, Ummesalma Diwan, Sourish Dasgupta, Manjunath Joshi, Tanmoy Chakraborty
Abstract
Personalization encoders compress evolving interaction histories into preference states used to rank items or condition text generation. A task head operating only on this state can miss useful evidence that remains in the frozen encoder's cached representations for individual timesteps. We study this recoverability gap and propose REPAIR, which compares cached representations with the current preference state in a compact learned coordinate space. It resolves corrective evidence over extended history, recent interactions, and localized bursts. It then selects which patterns at which timesteps contribute and adds their aggregate correction to the state before the task head. Encoder-host repair reuses representations from the existing forward computation without re-encoding the history. Across MovieLens, PENS, MIND, and Amazon Reviews 2023, training only REPAIR improves MRR and nDCG@10 for all twelve representative recommendation hosts while both encoder and task head remain frozen. Head-only finetuning of the same hosts yields smaller gains. For example, Mamba4Rec on MovieLens gains 3.96 MRR points, compared with 0.19 from head-only finetuning. Rank and temporal diagnostics support a compact, host-dependent corrective structure. In personalized generation, IMPerSumm improves the two reported weighted PerSEval variants, which assess responsiveness to user preference, by up to 25.23%. These results support post-compression state correction and distinguish the availability of preference evidence from its downstream use.