Akademik Veri Yönetim Sistemi
Yükseköğretim Kurumları Destekli Proje, 2015 - 2016
With the improvements of internet technologies, recommender
systems have been used in many areas. Recommender systems recommends the appropriate
items with the respect of users’ past choices and features without any efforts
of users. As the result of many researches and studies at last years, recommender
systems have been used in many areas. The most popular recommender systems are
movie (MovieFinder.com), music (last.fm), news (Google News), book (Amazon.com),
restaurant and holiday (booking.com) recommender systems. Recommender systems
approaches are Collaborative Filtering, Content-Based Filtering and Hybrid
Filtering. These systems make recommends to users for an item. System do this
with the help of user’s own past preferences and similar user’s past
preferences. The accuracy of the systems’ recommends is very important for users.
There are many research for the purpose of improving recommends’ quality. Generally,
users make decision with using these recommends.
For some purchase of items and services, collecting
user’s opinion for different criteria is more efficient. For example, an user
may want to share opinion for different criteria that are movie’s directing,
visual effects, acting, in spite of sharing opinion for just one criteria. For
this purpose, researches developed multi-criteria recommender systems that
provide more accurate and effective evaluating for items to users (YahooMovies,
Booking.com). All research for multi-criteria recommender systems have used
numerical data until now. For example, a system’s items may have evaluation
criteria that is specified preferences at 5-star. Users share opinions about
items like numerical data in this scale and the system can make recommends to
users for items in this scale. But,
using of numerical data in multi-criteria systems that have several criteria,
effects negatively to system’s performance.
Expecting users evaluate their opinions for each
criteria like numerical data is a negative situation for users and systems. In
spite of making recommend like numerical data, using like-unliked (1-0)
improves to system’s performance. For this purpose, there are algorithms that
makes recommends for single criteria recommenders systems. The Simple Bayesian
Classifier is used in these studies and proposed the systems that makes
recommends based on binary data of users. But, for multi-criteria recommender
systems, in literature, there is no research for this purpose. It is planned to
be primarily research applicability of multi criteria recommender systems based
on binary data. Besides, for improving quality of recommends, it is aimed that
making more accurate recommends with using mathematical techniques in
literature and developed new mathematical techniques.
The studies planned during the project will fill
the gap in the literature and to be novel. Because of this, there is high potential of
publishing the results in international indexed journals. Duration of the
project was envisaged as 6 months. Graduate student, who will work in the project,
will complete the graduate thesis during the project period.