PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Demo: using personalized PageRank for keyword based sensor retrieval
Lorand Dali, Alexandra Mortaru and Dunja Mladenić
(2011) Association for Computing Machinery, Hyderabad, India.


PageRank, one of the most important algorithms in information retrieval, was developed to give an estimate of how many users are likely to be visiting a given page on the web at a given moment. This way PageRank gives an approximation of the popularity of a page. Because semantic web resources can be represented in a RDF graph it is easy to adapt the original PageRank algorithm to rank these resources; and indeed much previous work has been done in this direction. However there are important differences between an RDF graph and a graph composed by web pages and hyperlinks. This paper explores to what extent PageRank and other domain and query-independent features can approximate the popularity of a resource in the semantic web. Moreover we study how these features can be combined to derive more robust ranking models. We did our study on DBpedia and Yago, two knowledge bases where the resources have one-to-one mappings to Wikipedia articles. For popularity of a resource we used the Wikipedia access logs as a golden standard.

EPrint Type:Other
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Information Retrieval & Textual Information Access
ID Code:8722
Deposited By:Jan Rupnik
Deposited On:21 February 2012