100,000 Reviews—Are You Reading Them All? Introducing Our AI Review Summary Feature

Musinsa introduced an AI review summary feature to solve the problem of customers struggling to find the information they want amid an overwhelming number of product reviews. For the clothing category, it provides keyword summaries using 12 keywords covering size, fit, material, and more, while other categories receive pros and cons summaries, delivering optimal information tailored to each user's shopping style.

Considering that new products may lack sufficient review data, the team designed a Priority Fallback structure that leverages reviews of other colors of the same product, expanding the coverage of the summary service.

On the technical side, the Qwen3-VL-8B-Instruct open-source model was chosen for cost efficiency. To address the sentence-copying issue that occurs with smaller models, the team designed prompts based on abstracted templates instead of concrete examples, and built a 9-stage post-processing pipeline to ensure quality and improve data accuracy.

In addition, cushion language was introduced to ease partner companies' concerns about exposing product shortcomings, and a UX/UI was designed to let users intuitively cross-check AI summaries against original reviews, securing the service's reliability.