Retrieving Web Pages Using Content, Links, URLs and Anchors

For this year’s web track, we concentrated on the entry page finding task. For the content-only runs, in both the ad-hoc task and the entry page finding task, we used an information retrieval system based on a simple unigram language model. In the Ad hoc task we experimented with alternatieve approaches to smoothing. For the entry page task, we incorporated additional information into the model. The sources of information we used in addition to the document’s content are links, URLs and anchors. We found that almost every approach can improve the results of a content only run. In the end, a very basic approach, using the depth of the path of the URL as a prior, yielded by far the largest improvement over the content only results.