86 C-N. Ziegler 3.5 Supplementary Fine-Tunir Taxonomy-driven profile generation renders another mechanism, dubbed "topic diver sification"[24], feasible. Hereby, our novel approach, optionally applicable on top of recommendation generation, allows rearrangement of the active user ais recommen- dation list in order to better reflect ais full range of interests, considering the impact of specific topics ai implicitly declares interest in. For instance, suppose that novels classifying under Modern German Poetry have twice the share of Social Psychology in a'is reading list. Then post-processing of a;'s recommendation list by means of topic diversification procedures allows to fully account for that fact To our best knowledge, no similar approaches exist or have been documented in literature affiliated with recommender systems. Moreover, topic diversification becomes even more valuable when making recommendations across diverse domains of interest e.g, books, DVDs, apparel, etc. Other enhancements include considering explicit product ratings for recommenda- tion generation whenever available. However, note that most scenarios only allow for collecting implicit ratings, e.g., purchase data, product mentions, etc, rather than explicit 4 Real-World Deployment Section 3. 1 exposed our envisioned information infrastructure. We will show that such an architecture may actually come into life and become an integral part of the Semantic Web Social Networks. FOAF defines machine-readable homepages based upon RDF and allows weaving acquaintance networks. Golbeck [5] proposed some modifica tions making FOAF support"real"trust relationships instead of mere acquaintance ship. Product Rating Information. Moreover, FOAF networks seamlessly integrate with so-called"weblogs", which are steadily gaining momentum. These personalized online diaries"are especially valuable with respect to product rating information For instance, some crawlers extract certain hyperlinks from weblogs and analyze their makeup and content. Hereby, those referring to product pages from large cat alogslikeAmazon.com(http://www.amazon.com)countasimplicitvotesforthese goods. Mappings between hyperlinks and some sort of unique identifier are required for diverse catalogs, though. Unique identifiers exist for some product groups like books, which are given"International Standard Book Numbers, i.e., ISBNS. Ef forts to enhance weblogs with explicit, machine-readable rating information have also been proposed and are becoming increasingly popular. For instance, BLAM! (http://www.pmbrowserinfo/hublog/)allowscreatingbookratingsandhelpsem- bedding these into machine-readable weblogs Product Classification Taxonomies. besides user-centric information, i.e. agent ais trust relationships ti and product ratings ri, taxonomies for product classifica- tion play an important role within our approach. Luckily, these taxonomies exist for86 C.-N. Ziegler 3.5 Supplementary Fine-Tuning Taxonomy-driven profile generation renders another mechanism, dubbed “topic diver￾sification” [24], feasible. Hereby, our novel approach, optionally applicable on top of recommendation generation, allows rearrangement of the active user ai’s recommen￾dation list in order to better reflect ai’s full range of interests, considering the impact of specific topics ai implicitly declares interest in. For instance, suppose that novels classifying under Modern German Poetry have twice the share of Social Psychology in a i’s reading list. Then post-processing of ai’s recommendation list by means of topic diversification procedures allows to fully account for that fact. To our best knowledge, no similar approaches exist or have been documented in literature affiliated with recommender systems. Moreover, topic diversification becomes even more valuable when making recommendations across diverse domains of interest, e.g., books, DVDs, apparel, etc. Other enhancements include considering explicit product ratings for recommenda￾tion generation whenever available. However, note that most scenarios only allow for collecting implicit ratings, e.g., purchase data, product mentions, etc., rather than explicit ones. 4 Real-World Deployment Section 3.1 exposed our envisioned information infrastructure. We will show that such an architecture may actually come into life and become an integral part of the Semantic Web: – Social Networks. FOAF defines machine-readable homepages based upon RDF and allows weaving acquaintance networks. Golbeck [5] proposed some modifica￾tions making FOAF support “real” trust relationships instead of mere acquaintance￾ship. – Product Rating Information. Moreover, FOAF networks seamlessly integrate with so-called “weblogs”, which are steadily gaining momentum. These personalized “online diaries” are especially valuable with respect to product rating information. For instance, some crawlers extract certain hyperlinks from weblogs and analyze their makeup and content. Hereby, those referring to product pages from large cat￾alogs like Amazon.com (http://www.amazon.com) count as implicit votes for these goods. Mappings between hyperlinks and some sort of unique identifier are required for diverse catalogs, though. Unique identifiers exist for some product groups like books, which are given “International Standard Book Numbers”, i.e., ISBNs. Ef￾forts to enhance weblogs with explicit, machine-readable rating information have also been proposed and are becoming increasingly popular. For instance, BLAM! (http://www.pmbrowser.info/hublog/) allows creating book ratings and helps em￾bedding these into machine-readable weblogs. – Product Classification Taxonomies. Besides user-centric information, i.e., agent ai’s trust relationships ti and product ratings ri, taxonomies for product classifica￾tion play an important role within our approach. Luckily, these taxonomies exist for