This paper presents a novel ranking approach for complex semantic relationship (semantic association) search based on user preferences. We define a feature vector to describe various statistical and semantic features of a semantic association. Our approach employs a learning-to-rank algorithm to learn a user's preferences to these features, and automatically construct a personalized ranking function for the user. The ranking function is then used to produce ranked lists of semantic association

more »

... earch results for subsequent queries from the user. Results that more closely match the user's preferences gain higher ranks. Our approach is evaluated on a real-world RDF knowledge base we created from Freebase data. The experiment results show that our method significantly improves the ranking quality in terms of capturing user preferences, compared to a typical existing approach.