The challenge
Making the search easy and convenient
Home apparel goods need to combine several functions, including comfort, purpose and looks. The last one poses a challenge, as it is not obvious if a particular piece of furniture fits the overall style of the room or apartment or how to compose a whole room in a consistent style. From the seller’s perspective, it may be extremely challenging to describe and communicate the features of the products, fit current trends or reach people potentially interested in buying this particular piece.
The approach
Word2vec in the world of furniture
Solvd team built a search engine that scrapes through a product database in one of the leading furniture retailer, finding the items that fit the prompt provided by the user. The system was trained on dataset consisting of images of products and already designed rooms. With that, the system was capable of spotting the patterns of context, colors and arrangements where a particular piece arrives.
The AI system behind the search engine was inspired by word2vec. Every item got its unique vector representation, so the neural network was capable of analyzing the images of whole rooms as a combination of vector representations of items.
Using these combinations, the system could deliver cloud representations of meanings, with overlapping areas, where an item can fit into different rooms or styles. For example, a white table can be used in the kitchen, in a dining room and in a living room.
The outcome
Spotting the perfect product
The end product allows the user to either upload the image of a particular piece of furniture or describe it in the natural language. The search engine returns the best fitting products, including styles, placement and anything else the user wishes to put in the prompt.
About client
Furniture retailer
The project was delivered to a large e-commerce retailer selling furniture and home apparel goods.
Industry
Retail & consumer goods
