| dc.description.abstract | Item recommendation is an increasingly important research topic that focuses on analyzing the relationships between products to recommend items to users based on their preferences or previous activity. These systems are used extensively in different applications varying across domains to recommend items ranging from books to music. Many companies, such as Amazon, Netflix, and Spotify, leverage recommender systems to drive further engagement and revenue by delivering value through a scalable way of personalizing content for their users.
Current recommender systems recommend items based on two factors: users and items. For example, if a user purchases a product, then the recommender system will recommend similar products based on the users' previous purchases or similar social circles. In certain domains, such as clothing and electronics, the focus of compatibility relationships between products should be analyzed and used to recommend products to offer a complementary product, not a similar one.
In our thesis, we propose a new definition of compatibility to provide a new and improved recommender system strictly for item compatibility. Compared to traditional recommender systems, compatiblity recommender systems provide more accurate item recommendations for users. Our thesis currently focuses on analyzing the compatibility relationships within top-level categories in Amazon data but can be applied to any domain where compatibility is important. In order to do so, we define a general definition of compatibility, analyze a large product dataset and map product relationships, create a model to identify compatible items, and compare our results with other models. We will be analyzing the Cell Phone & Accessories category with our compatibility definition. Compared to other recommender systems, our compatibility recommender system is able to recommend compatible items at a higher accuracy and can therefore be used to provide users with a more personalized experience. | en |
