Authors: Heimbach, Irina; Gottschlich, Jörg; Hinz, Oliver
Abstract: Most online shops apply recommender systems, i.e. software agents that elicit th ...
Expand
Abstract: Most online shops apply recommender systems, i.e. software agents that elicit the users’ preferences and interests with the purpose to make product recommendations. Many of these systems suffer from the new user cold start problem which occurs when no transaction history is available for the particular new prospective buyer. External data from social networking sites, like Facebook, seem promising to overcome this problem. In this paper, we evaluate the value of Facebook profile data to create meaningful product recommendations. We find based on the outcomes of a user experiment that already simple approaches and plain profile data matching yield significant better recommendations than a pure random draw from the product data base. However, the most successful approaches use semantic categories like music/video, brands and product category information to match profile and product data. A second experiment indicates that recommendation quality seems to be stable for different profile sizes.
Collapse
Semantic filters:
authorizationApache Lucene
Topics:
Facebook recommender system social network database system experience good
Methods:
experiment survey chi squared test hierarchical linear modeling longitudinal research