Site Search
Computer Science

Photos

Xiaobu Yuan, Ph.D.Dr. Xiaobu Yuan
Dr. Xiaobu Yuan
Dr. Scott GoodwinDr. Scott Goodwin
Dr. Scott Goodwin
Alioune Ngom, Ph.D.Dr. Alioune Ngom
Dr. Alioune Ngom
Lambton TowerLambton Tower
Lambton Tower
Dr. Luis RuedaDr. Luis Rueda
Dr. Luis Rueda
Christie Ezeife, Ph.D.Dr. Christie Ezeife
Dr. Christie Ezeife
Dr. Robert KentDr. Robert Kent
Dr. Robert Kent
Dr. Ziad Kobti lecturingDr. Ziad Kobti
Dr. Ziad Kobti
Imran Ahmad, Ph.D.Dr. Imran Ahmad
Dr. Imran Ahmad
Arunita Jaekel, Ph.D.Dr. Arunita Jaekel
Dr. Arunita Jaekel
Jessica Chen, Ph.D.Dr. Jessica Chen
Dr. Jessica Chen
Windsor WaterfrontWindsor Waterfront Park
Windsor Waterfront Park
Robin Gras, Ph.D.Dr. Robin Gras
Dr. Robin Gras

Recommending Best Products on E-commerce Clickstream History using Weighted Frequent Pattern Mining

Add this event into your calendar using the iCAL format
  • Fri, 12/08/2017 - 10:30am - 12:30pm




Recommending Best Products on E-commerce Clickstream History using Weighted Frequent Pattern Mining

MSc Thesis Proposal by:

Ying Xiao

Date:  Friday, December 8th, 2017
Time:  10: 30 am – 12:30 pm
Location: 3105, Lambton Tower

Abstract: Clickstream data including clicks, basket placement, and purchase activities have been used by existing E-Commerce recommendation systems such as those in (Kim05Rec, Kim11Rec, Chen13Rec) to address the deficiency of uninformative user-item rating matrix generated from purchase data,that only shows what items a user has purchased previously. Kim05Rec forms a decision tree to find the possibility of adding a product into the shopping cart, then uses the result to enrich the matrix for collaborative filtering (CF) to improve recommendation accuracy, with only the clicked products as candidates. Kim11Rec uses association rule mining which was proved more accurate than Kim03Rec, but it does not care for the rare cases and infrequent users. Chen13Rec compares clickstream sequences by visiting path, browsing frequency and access time for each category to find the similarity between users, but its technique for mining the whole dataset for category-level sequence is not efficient.

In this thesis, we propose a recommendation system for E-commerce on clickstream history (including purchase history) using weighted frequent pattern technique which does not limit the product candidates to the clicked ones, takes care of rare cases and infrequent users, operates on product level recommendations and improves the recommendation accuracy. We enhance the similarity method in Chen13Rec by generalizing it to product level to calculate the similarity between clickstream sequences in each session. For every sequence without purchase, we find the top-N similar sequences with purchases for it. Then use the similarity as a weight for the selected purchases, we propose a weighted (the weights are assigned to transactions instead of items) frequent pattern mining algorithm (WFPM) to calculate the probability of a user purchasing an item (or a combination of items). Two approaches will be introduced; one uses the probability of 1-item patterns from WFPM to enrich the user-item rating matrix before the conventional CF method to generate recommendations; the other directly uses all the results from WFPM as recommendations which may have combinations with multiple products.

Thesis Committee:
Internal Reader: Dr. Jianguo Lu
External Reader: Dr. Zhiguo Hu
Advisor: Dr. Christie Ezeife



csgradinfo@uwindsor.ca
(519)253-3000


See More: