Abstract

Fashion companies offer a large variety if products to their consumers with the objective to meet the fashion preferences as best as possible. Consumer behavior research indicates that large product offerings may lead to choice overload at the consumer resulting in losses in sale and an overall negative consumer experience. To mitigate the choice overload predictive models for product recommendation are applied to individually filter the product range according to the consumers estimated preferences. The fashion industry is one of the largest global consumer markets. Nevertheless only few research is available how to recommend especially fashion products in a highly accurate manner. Traditional recommender systems assume that the individual consumer preferences are static. On the opposite consumer behavior research shows that during the consumer purchase process contextual information have a significant influence on the final purchase. Therefore context-aware recommender systems emerged as a new subfield within the recommender research community having the goal to integrate relevant consumer context in the product preference predictions.

Research Picture

Research Objective

The overall research objective is to assess the impact of user and product profile attributes on the predictive performance of state-of-the-art algorithms for creating predictive models for product recommendation.

Therefore first relevant static and dynamic user profile information like personality, emotion, or appearance are identified which are relevant within the fashion purchase process. Beside this the visual appearance of the fashion products are integrated as well in the prediction model by extraction color and pattern information from the product images. The predictive power of these context and product attributes are then empirically evaluated by using factorization methods, especially tensor factorization methods.

Company contacts

• adidas AG

Related Student Reports (excerpt)