DSpace Repository

Sliding window association rule mining

Show simple item record

dc.contributor.author Onanong Nopkhun en_US
dc.date.accessioned 2015-01-12T10:40:00Z
dc.date.available 2015-01-12T10:40:00Z
dc.identifier.other AIT Thesis no.CS-03-27 en_US
dc.identifier.uri http://www.cs.ait.ac.th/xmlui/handle/123456789/286
dc.description Pathum Thani, Thailand : Asian Institute of Technology, 2003 en_US
dc.description 47, 4, 15 p. : ill. en_US
dc.description.abstract In rapidly thriving barter exchange business, human brokers as business mediators take care of all their members in various ways such as what they will purcha se, what products or services will be provided, and purchase r ecommendation. For recommendation, brokers need to observe all their members’ purchase beha vior. How effective the recommendation is depends on an individual broker’s experience. In order to help new brokers generate useful recommendation in a short time, a recommendation engine, a semi-automated broker, is proposed in this thesis. The objec tive is to observe and predict purchases based on sparse real business transactions. Among several data mining techniques, a novel approach, Sliding Window Association Rule Mining (SWARM) algorithm, is proposed to disc over the association ru les from temporal transactional data. The correlations of produc ts purchased within a given time window are considered as the rules which are used for re commendations at appropria te thresholds. In the experiment, three parameters: minconf, minsup and winsize, were used to find out the affect of these parameters on the association rules. The experimental results of testing with real data show reasonable and accurate association rules. The results were evaluated via an eval uation program and a human expert. First, the evaluation program shows that small time window is found to be slightly more helpful than large time window. Second, the human expert who cl assified how useful or interesting these rules are shows that there are still some rules relatively useless to generate recommendations. As one of the confidence measures, lift is appropriated for ranking top-N rules and also improving the generated rules to become more efficient for recommendations.
dc.relation.ispartof Thesis no. CS-03-27 en_US
dc.relation.ispartof Asian Institute of Technology. Thesis no. CS-03-27 en_US
dc.subject Barter en_US
dc.subject Exchange en_US
dc.subject Telemarketing en_US
dc.subject Data mining en_US
dc.title Sliding window association rule mining en_US
dc.type Thesis en_US

Files in this item

This item appears in the following Collection(s)

Show simple item record

Search DSpace

Advanced Search


My Account