ARES – Aesthetics-based Product Recommendation System for Web Shops.

In ARES we will develop technology for online shopping portals that recommend products to customers based on their individual aesthetic preferences. As small shopping systems do not possess a sufficiently large database for state of the art item-to-item or user-to-user recommendations, we will develop prediction models that do not require existing databases.

When visiting a furniture store such as IKEA we can observe an interesting phenomenon. Although we wanted to buy a single, specific product we also add items such as candles, an indoor palm tree, napkins, a wooden box, and a pillow. Within ARES we want to address the question of how such buying behaviour can be elicited in online shops.
The choice of additional items that we buy is not based on functionality or product categories, but on aesthetic preference. Recommendation systems of online shops (“costumers who bought this item also bought ….” or “… frequently bought together”) that fail to capture the aesthetics of a product will miss an opportunity of selling such additional products.

In ARES we will develop technology for online shopping portals that recommends products to customers based on their individual aesthetic preferences.

The major benefits of ARES are

• an individual product collection is presented to the customer

• the recommendation system works on the personal preference level and is not bound to product categories or functionality

• products can be grouped and visualized according to their aesthetic properties

• small online shops rarely own a large database to allow product specific recommendation systems – here more general prediction models can supplement the existing database.
• in contrast to existing item-to-item or user-to-user recommendations, new products can be integrated without additional efforts such as definition of meta-information or assignment to product categories into the recommendation system

The main result of ARES will be software libraries that are able to predict aesthetic properties of products based on images of the products. A predictive model that is calibrated through experimental procedures will use colour, texture and shape information to predict aesthetic properties.

The main benefit of ARES for SMEs (designers and operators of online shopping systems) is that customized product recommendations are available from the very first customer, without requiring an existing database of user preferences. Once the predictive model is calibrated for a particular context (e.g. furniture stores), it can be easily transferred between different online shops and refined based on collected user preferences.

online shop systems
recommendation systems
aesthetics
machine learning
content based retrieval / suggestion

RTD (coordinator): research center in Austria.
We received some EOIs but no additional partner is fixed as the partner search has currently started.