RWB (the united Wildberries & Russ company) is launching an AI filter test for reviews, a tool that automatically groups user comments by topic and allows you to quickly find opinions about specific product characteristics. The ML-based model analyzes the text of reviews, highlights key topics and forms neutral filters like "Size", "Smell", "Color", "Taste", combining both positive and negative ratings in one filter.
It is also important that the filters are based on the latest reviews from the last two years, and the list of topics is dynamic: new reviews are run through the model daily, and filters can be changed once a day. The feature is already available to most users of the mobile app in the Ratings and Reviews section, and will appear on the website later.
The key difference between this solution and the usual "sorting" is that it responds to the real pain of marketplaces: with a large volume of UGC, the buyer drowns in hundreds of similar messages and eventually makes a decision on price or rating without understanding the nuances. AI filters translate reviews from the "noise tape" into fact-based navigation: how the size grid behaves, whether the material smells, whether the color matches the photo, how loud the hardware is, etc. This improves the quality of choice and reduces the share of disappointments — and therefore, potentially, returns.
The key quote explaining the product logic came from Polina Ovchinnikova, head of the Product Reviews, Questions and Rating department.:
"Artificial intelligence helps us systematize a large number of product reviews and speed up routine processes for users. Now, instead of manually examining dozens or even hundreds of comments, the AI will instantly group and filter them. This allows users to quickly learn exactly about the characteristics of the product that are important to them, saving time and making the choice more conscious," commented Polina Ovchinnikova, head of the Product Reviews, Questions and Rating department.
For sellers, this has a double effect. On the one hand, "bad" themes (for example, unstable size or persistent odor) will become visible faster — filters make problematic patterns more noticeable. On the other hand, honest products win: the buyer finds confirmation of the necessary parameters faster and doubts less. As a result, the competition is shifting from "impression manipulation" to quality and content management: correct flashcards, accurate photos, clear size charts, and answers to questions.
There is also an indirect effect in logistics and foreign economic activity. The more accurate the buyer's expectations are, the lower the load on return logistics (returns / exchanges), less "sawing" of shipments in warehouses and less costs for sorting the return flow. This is especially important for the marketplace against the background of rising costs for last mile and warehouse processing.: AI tools are starting to work as a technology to reduce transaction costs by improving the quality of choice, and not just as an "interface feature."
Contextually, this is a continuation of Wildberries' line of implementing AI tools: previously, the company tested a neural network retelling of reviews ("Important from reviews") to speed up the reading of long comment threads.
