Exploiting Semantic Product Descriptions for Recommender Systems
Cai-Nicolas Ziegler, Lars Schmidt-Thieme, Georg Lausen
Content-driven and hybrid recommender systems propose products to customers making use of descriptive features and behavioral patterns, likewise. While most approaches exploit classical information retrieval techniques, e.g., nearest-neighbor queries in metric spaces, availability and usage of richer semantic meta-information about products may further improve recommendation quality significantly. Massive taxonomies for product classification are coming of age, e.g, the United Nations Standard Products and Services Classification (UNSPSC), as well as proprietary standards, such as Amazon.com's classification taxonomies for books, DVDs, CDs, and apparel. We exploit suchlike semantic background knowledge in order to leverage powerful inference opportunities for making user profiles, based upon the products these latter customers purchased, more meaningful. Ample empirical analysis, both offline and online, demonstrates our proposal's superiority over common existing approaches when user information becomes sparse and implicit ratings prevail.
[ PDF File ]