content platform developers

Abstract: A model trained on readers' highlights can show a newcomer where earlier readers converged on a page, even on pages with few readers. A year of stronger language models suggests two upgrades: use them directly, or use them as teachers. We compare both with a third, more reader data, on one social highlighting platform, with the comparisons, the selection rules and a sealed set of fresh pages fixed in advance. On 215 development pages, seven language models released by October 2026 rank sentences zero-shot at an average precision of at most 0.3744 against the sentences half the readers marked most, and 0.3931 when fused with reading position, while a 3-billion-parameter model trained on reader highlights alone reaches 0.4346. Used as the teacher that breaks ties among readers' counts, the stronger fusion changes a trained model by +0.0013 [-0.0012, +0.0040], whereas going from half to all of its training pages adds +0.0242 [+0.0179, +0.0301]. A 27-billion-parameter backbone on the reader-only recipe adds +0.0225 on the development pages; on 131 fresh pages, read once and too few for a confirmatory comparison, its difference from the 3-billion model is unresolved, both score above the fusion, and in an exploratory split their lead over the previous model sits on pages from domains that their fresh training pages also cover. In this recipe, more reader data improves a trained model where a stronger tie-breaking teacher yields no material gain, and frontier models used as rankers still trail a small model trained on readers.

Abstract: How can we personalize a shared expert library from a user's pairwise choices? Prior work can realize different reward trade-offs by merging reward-specialized experts, given a vector of trade-off weights. In practice, users can more naturally choose between outputs than specify numerical weights. The challenge is therefore to turn these choices into the coefficients required for merging, while accounting for ambiguity when feedback is limited. Our key idea is to treat the unknown reward weights as latent variables: infer a posterior over them from pairwise choices and reward-score differences, and use its mean directly as the merge coefficients. We instantiate this idea as Bayesian Pairwise Merging (BPM), whose posterior also characterizes which reward trade-offs remain plausible given the feedback. We evaluate BPM on radiology summarization, image captioning, and story generation, spanning text-to-text and image-to-text generation. With 100 feedback per simulated persona, BPM achieves macro decided win rates of 91.7%, 77.1%, and 64.3% against uniform merge. For six pairs of simulated personas, each prefers the model fitted to its own feedback, a pattern also observed in a human proof-of-concept. In simulations under BPM's model and prior, its nominal 90% intervals for temperature-scaled reward weights achieve task-averaged marginal coverage of 88.9% and 89.2% with only 10 and 25 comparisons, respectively. BPM thus enables personalization from pairwise feedback without per-user policy training, while characterizing the coefficient ambiguity left by limited feedback.

Abstract: Recent advances in generative modeling have reshaped recommender systems by formulating recommendation as a next-item generation problem. Existing retrieval approaches primarily follow two paradigms: user-to-item (U2I) methods represent user context using one or a few deterministic embeddings, which limits the ability to capture diverse and multi-mode interests, while semantic-ID-based autoregressive (SID-AR) methods model more expressive distributions but suffer from quantization errors and the low throughput of sequential decoding. To address these limitations, we propose X-Rec to directly learn the recommendation distribution in the continuous item embedding space through flow matching and generate embedding triggers for approximate nearest neighbor retrieval. X-Rec incorporates three key designs to make this formulation effective and efficient. First, we introduce anchor conditioning to decompose generation into coarse semantic-region selection and fine-grained refinement. Second, we adopt Riemannian flow matching to align generative trajectories with the hyperspherical geometry of item embeddings. Third, we design a late-interaction diffusion Transformer that restricts repeated velocity-field estimation to the final Transformer layer. On a streaming benchmark, X-Rec substantially outperforms U2I baselines, matches the retrieval quality of SID-AR methods, and delivers 3.46x higher inference throughput than SID-AR. X-Rec has also been deployed as a new retrieval source for a specific vertical content on TikTok, where two consecutive launches have yielded significant improvements in both vertical engagement (+4.1484%) and general engagement (+0.0111%).