Abstract:
This thesis aims to design a new recommendation system to be used by web application to recommend interesting items to users. Recently, many web applications are created and published in the internet to allow their users to ac-cess millions of items, and the number of users as well as the number of items is increasing exponentially. These web applications use recommendation tech-niques that are based on users' preferences for items to recommend a few inter-esting items to users.
Collaborative filtering (CF) is currently the most successfully used rec-ommendation system; it is based on the similarities between users and the simi-larities between items. However, users' opinions and items' popularities vary with time. These variations decrease the recommendation accuracy. On the oth-er hand, many researchers investigate ways of using Morkov model in recom-mendation systems; however, the time factor can be better used in new tech-niques.
We introduce a new recommendation system, then we enhance it using the time factor and friends feature. The contributions in this thesis are as follows:
We propose a new technique entitled the basic Markov Chain Recommen-dation System MCRS that is based on Morkov model, and the time factor.
We enhance the basic MCRS using items' popularities in general.
In the second enhancement, we use items' popularities in the last period of time.
In the last enhancement, the basic MCRS is weighted by friends weights of items.
We compare our new technique with the conventional CF recommenda-tion system for the evaluation. We conduct the experiments using dataset from MovieLens and LastFm. The evaluation is done by using precision-recall, accu-racy, area under ROC curve, and mean absolute error.
The result illustrates that the basic MCRS outperforms the conventional CF recommendation system, and the time factor affects positively or negatively in the recommendation. All the enhancements outperform the basic MCRS. The friend weights MCRS outperforms the basic MCRS and its enhancements.
