WebTech and Security Lab

Current Projects
Majority Web (Coselections)

When a user makes a selection from a set of search results, they are making an implicit relevance feedback between the search term and the chosen webpage. This is especially so for image searches where we have found that the relevance feedback is much more accurate.

We have been taking this a step further - when a user makes more than one selection from the search results, which we call a coselection, then we have found that there is also an implicit mutual relevance between the two selected websites.

There are a number of things that we postulate that coselections can be used for. We aggregate (or cluster) webpages using coselections as the similarity function, and we have shown that this creates clusters of webpages which are non-ambiguous - there is no other algorithm that can do this. With these non-ambiguous clusters, we are working on all of the following:

  • separating out multiple senses of the same word or phrase
  • detecting ambiguous terms
  • finding synonyms
  • finding translations
all of these completely automatically and without using any external resources.

We have recently published a paper that summarises all the experimental work so far on this project (Ashman, Antunovic, Chaprasit, Smith and Truran, see the publications page).