Personalizing Incremental Video Search with Hybrid Text and ID Embeddings
arXiv:2607.13493
The paper introduces a system for personalizing incremental video search on Apple TV by combining text‑based multilingual embeddings and ID‑based collaborative embeddings, using recent user watch history to improve ranking after each keystroke.
Abstract
Incremental video search requires high-quality ranking after each keystroke, where intent is often underspecified (e.g., 1-3 character prefixes). We present a personalization system for Apple TV search that combines complementary semantic and collaborative signals at ranking time. Our approach learns two item embedding spaces: (i) a text-based multilingual encoder (TextEmb) fine-tuned on co-engagement triplets via contrastive learning, and (ii) an ID-based collaborative embedding model (IdEmb) trained on interaction-derived positives. At serving time, we construct user representations from recent watch history and inject text- and ID-based user-item cosine similarities into a pairwise XGBoost ranker. We evaluate with temporally held-out offline datasets and a three-week online controlled experiment. Offline, for sessions with user history, the personalized ranker improves NDCG@10 by 2.99% and MRR by 3.30% over the non-personalized baseline. Slice analyses show that personalization is most needed in incremental search, where intent is still forming: on ambiguous prefix queries (1-3 characters), NDCG@10 lift is +8.63%, versus +1.46% on longer, fully specified queries. Longer-history users benefit more: NDCG lift rises from +2.13% for users with 1-5 history items to +4.37% for users with 51-100, even though baseline relevance is lower for these cohorts (NDCG@10 drops from 0.733 to 0.680), indicating that personalization adds the most value where default ranking underperforms. Online, treatment yields statistically significant gains of +1.14% tap-through rate and +1.23% conversion rate, with a 2.91% improvement in converted-item rank position. We further analyze coverage-precision trade-offs between semantic and collaborative embeddings via ablations isolating each signal, and evaluate embedding quality on a held-out corpus with LLM-judged similarity labels to reduce click/exposure bias.
Accepted to the Industry Track of the 20th ACM Conference on Recommender Systems (RecSys 2026)