IMPROVING E-COMMERCE COLLABORATIVE RECOMMENDATIONS BY SEMANTIC INFERENCE OF NEIGHBORS’ PRACTICAL EXPERTISE 6 th International Workshop on Semantic Media Adaptation and Personalization SMAP 2011 Vigo, December 2011 Manuela I. Martín Vicente University of Vigo (Spain)
Introduction Recommender systems Save users from dealing with overwhelming amounts of information, providing them with personalized suggestions. E-commerce Collaborative filtering recommenders Based on offering to the active user items that were appealing to other individuals with similar preferences (neighbors). Neighborhood formation stage Two users alike – similar ratings to the same items Prediction stage Interest in an item a user does not know – neighbors’ ratings 2
Antecedents and motivation Expertise – “skill or knowledge that a person has in a particular field” Considering neighbors’ expertise in the recommendation process can increase accuracy Problem – lack of information about the skills of the users in a recommender system Our proposal Measure of users’ practical expertise Transparently to users – only requires their consumption history A semantic approach 3
Domain ontology and user profiles User modeling by reusing the knowledge formalized in the ontology IDs to product instances 4
Inference of practical expertise 5 User’s expertise in a leaf class of products (C L ) Two premises: Variety of product instances E.g. Wine – different brands of wine Variety of attributes of the products E.g. Wine – from varied countries of origin, produced with diverse grape varieties, of different qualities, etc. Accordingly, two components: Quantitative variability Qualitative variability
Quantitative variability 6 Depends on the number of different instances of C L the user profile stores Not linked to a specific ontology Two factors: Variability – user Population factor – ontology
Qualitative variability (I) The more diverse attributes of the products (instances of C L ) in a user profile, the higher the qualitative variability Semantic dissimilarity (SemDis) 1. SemDis attributes Depth Least Common Ancestor (LCA) 2. SemDis instances 7
Qualitative variability (II) 8
Conclusions 9 Practical measure On the basis of the products the users have tried Our strategy exploits users’ consumption histories Available in any e-commerce recommender system Reasoning techniques to inspect the semantics of the products stored in the users’ profiles Considering neighbors’ expertise can greatly improve recommendation results
Thank you for your attention! 10
IMPROVING E-COMMERCE COLLABORATIVE RECOMMENDATIONS BY SEMANTIC INFERENCE OF NEIGHBORS’ PRACTICAL EXPERTISE 6 th International Workshop on Semantic Media Adaptation and Personalization SMAP 2011 Vigo, December 2011 Manuela I. Martín Vicente University of Vigo (Spain)