Leveraging Linked Data Analysis for Semantic Recommender Systems [chapter]

Andreas Thalhammer
2012 Lecture Notes in Computer Science  
Motivation Traditional (Web) link analysis focuses on statistical analysis of links in order to identify "influencial" or "authorative" Web pages like it is done in PageRank, HITS and their variants [10] . Although these techniques are still considered as the backbone of many search engines, the analysis of usage data has gained high importance during recent years [12] . With the arrival of linked data (LD), in particular Linked Open Data (LOD), 1 new information relating to what actually
more » ... ts different vertices is available. This information can be leveraged in order to develop new techniques that efficiently combine linked data analysis with personalization for identifying not only relevant, but also diverse and even missing information. Accordingly, we can distinguish three problems that motivate the topic of this thesis: Relevance. LOD is well known for providing a vast amount of detailed and structured information. We believe that the information richness of LOD combined with user preferences or usage data can help to understand items and users in a more detailed way. In particular, LOD data can be the basis for an accurate profile which can be useful for recommendation in various domains. As information about items in the user profile is often unstructured and contains only little background knowledge, this information needs to be linked to external sources for structured data such as DBpedia. 2 Also, product and service providers need to link their offers accordingly. Methods for this two-way alignment have to be specified and evaluated. Diversity. According to [5] , recent developments in semantic search focus on contextualization and personalization. However, approaches that semantically enable diverse recommendations for users, also in context to users' profiles, remain barely explored. Of course, this states a complementary way of recommendation that is often only based on ranking by relevance. Consider the example of a news aggregation Web site which ranks articles by popularity. Popular articles are placed at the main page. On the same topic, there are hundreds of additional articles from other news sites and blogs indexed, but not visible on the main page. Of course, these articles get much 1 Linking Open Data -http://ow.ly/8mPMW 2 DBpedia -http://dbpedia.org/ E.
doi:10.1007/978-3-642-30284-8_64 fatcat:rvikpyohzzdavkck5d6ysdlfda