image/svg+xml
CONTEXT-ORIENTATED NEWS FILTERINGFOR WEB 2.0 AND BEYOND
The Problem:Information Overload
David WebsterCentre for InternetComputingThe University of HullScarborough CampusUnited KingdomD.E.Webster@dcs.hull.ac.uk
Weihong HuangFaculty of Computing,Information Systems, and MathematicsKingston UniversityKingston Upon ThamesUnited KingdomW.Huang@Kingston.ac.uk
Darren Mundy,Paul WarrenCentre for InternetComputingThe University of HullScarborough CampusUnited KingdomD.Mundy@hull.ac.uk,p.j.warren@hull.ac.uk
Category Distances
User TopicPreferences
Word SenseDisambiguation
Information overload[1] is currentlya problem on the web today. As anaid against this problem, technologies,such as RSS news syndication, can help byallowing users to subscribe to new feedsin order to have news delivered to themautomatically. Despite this convenience, if a usersubscribes to a dozen news feeds, which produceon average 10 news items a day, then there will be120 news items to read through. We can see thatthere needs to be some way of filtering newsitems, for those which are relevant to the user.
If we can map our news itemcategories (word senses) intoa topic tree, for example DMOZ,it is then possible to measure the distancebetween word senses using tree distancealgorithms[3]. Through distance measureswe can make decisions on what to do witha news item based on the set of its topiccategories.
Through the use of a set of usertopics of interest (or disinterest)and a word sense distancemeasure, we can make a decision on what todo with a news item based upon its topiccategories. Examples of this include, droppingnews items outside of a particular distance ofa set of topics of interest, dropping items basedupon a block list and assigning news items tocategory folders, based on the distancebetween the category and the news item'stopics.
RSS 2.0 and ATOM newsfeeds possess the facilityto contain a set of categorytags for news items.Unfortunately using keywords within these tagspresents us with the problem of word sensedisambiguation[2]. For example the keyword'Java' could refer to a programming language,an island, or coffee. One solution to this is touse a well defined knowledge base and refer toa namespaced word/tag rather than use thekeyword string 'Java'.