LLM-SCRec: LLM-Powered Style-Context Insights for Fashion Recommendation
Bibliographic record
Abstract
Fashion recommendation is increasingly shaped by platform-native narratives such as titles, captions, tags, and reviews, where users implicitly reveal their style preferences and the contexts in which items are chosen. However, existing systems often simplify such text into predefined labels or treat it as auxiliary metadata, which leads to semantic loss and limits the ability to model fine-grained, context-dependent preferences. This gap prevents recommendation models from fully exploiting the expressive power of natural language to capture style intent. To address these challenges, we propose LLM-SCRec, an interaction-centric framework that directly leverages large language models to extract <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">style–context insights</i> from raw user–item interaction text. Specifically, (1) a frozen LLM reads platform-native descriptions and generates high-dimensional interaction-level insights; (2) a lightweight encoder projects these insights into a compact embedding space and aligns them with collaborative signals through contrastive objectives; and (3) aggregated user/item representations are seamlessly integrated with existing ID-based recommenders through simple concatenation. Extensive experiments across multiple benchmarks show that this framework consistently improves performance on standard Top-<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">K</i> metrics relative to strong baselines. Beyond accuracy, its plug-and-play nature makes the method practical: because it does not alter the backbone recommenders and requires only offline LLM processing, it offers an efficient and easily deployable solution.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".