The distribution of the amount of preference information across customers is not the same in every domain of recommendation problems. It is necessary to treat each user differently based on their available preference information. In this dissertation, we have proposed three novel recommender system approaches that depend on each user’s preference information and can produce recommendations in a user-specific, parametric way. This parametric approach allows different weights to be assigned to different kinds of page-page similarity features used in the recommendation process, depending on the user group to which a particular user belongs. This novel approach of incorporation of different kinds of item-item (or page-page) similarities is shown to result in a significant increase in recommendation accuracy. In our first approach, we incorporated content-based and co-occurrence-based page-page similarities parametrically, by determining relative weights of the two component page-page similarities in user-specific way. We implemented a Map-Reduce based, parametric, hybrid recommendation system in order to solve the scalability issues. Experimental results showed better accuracy for this unique, scalable, and user-specific parametric approach, compared to that of another related work. In our second approach, we used clustering-based, hybrid recommendation system in user-specific way to get better accuracy and to further alleviate scalability issues by exploiting pre-computed clusters. This clustering-based incorporation approach showed better result than our first approach for users having extremely small amount of preference information. Finally, in our third approach, we proposed a graph-based, hybrid recommendation system. Two graphs using, respectively, content similarity and co-occurrence similarity were created. An approach involving features derived from these two graphs to make web page recommendations was introduced. For each user-page pair, one combined feature component was first obtained by making a weighted summation of the eight feature sets from each graph. Use of supervised learning for deriving feature weights to obtain combined feature components showed much more promising results, compared to the first two methods. Finally, the two feature components, from the two graphs were combined in user-specific way to train a model and make recommendations. To the best of our knowledge, ours is the first such effort in the recommender systems context.
|Advisor:||Raghavan, Vijay V.|
|Commitee:||Benton, Ryan, Chu, Henry, Loganantharaj, Rasiah|
|School:||University of Louisiana at Lafayette|
|School Location:||United States -- Louisiana|
|Source:||DAI-B 77/06(E), Dissertation Abstracts International|
|Keywords:||Clustering, Collaborative-filtering, Graph, Hybrid-recommender, Large-scale, Page-content|
Copyright in each Dissertation and Thesis is retained by the author. All Rights Reserved
The supplemental file or files you are about to download were provided to ProQuest by the author as part of a
dissertation or thesis. The supplemental files are provided "AS IS" without warranty. ProQuest is not responsible for the
content, format or impact on the supplemental file(s) on our system. in some cases, the file type may be unknown or
may be a .exe file. We recommend caution as you open such files.
Copyright of the original materials contained in the supplemental file is retained by the author and your access to the
supplemental files is subject to the ProQuest Terms and Conditions of use.
Depending on the size of the file(s) you are downloading, the system may take some time to download them. Please be