Difference between revisions of "Recommender Systems: Thesis Description"
Latest revision as of 14:41, 30 January 2009
Natasha Eilbert Senior Conference Prof. Deepak Kumar January 30, 2009
Recommender Systems Thesis
My thesis will explore Recommender Systems. These systems are computer programs (often located on the World Wide Web) that recommend to a given user a group of items that the user might like. Recommender systems can also be described as estimators which approximate how much a particular user will rate a given item based on information about either the item, the user, or similar users. Examples of such systems include Pandora, which recommends music, and Amazon.com, which recommends books and other items. There are a variety of different kinds of recommendation systems, as well as a number of issues (security, privacy) that surround such systems. There are a wide range of applications for these systems and a number of applications for these systems within other fields of computer science (e.g., interacting multi-agent systems, market economy and trust modeling for computers). In my thesis I will delve into some of these issues and aspects of recommendation systems. My plan is to develop my senior thesis in two parts.
The first part will be to look at past and current recommender systems and to analyze how the field of recommender systems stands. I will be tracing the systems from a historical viewpoint in order to best understand the developments made in the field. To that end, my thesis will look at early recommender systems and how continuing developments have improved and changed such systems. The thesis will cover the various types of recommender systems currently available, including personalized, social (which uses collaborative filtering techniques), item-based, and hybrid recommender systems. As my particular area of interest concerns artificial intelligence, I intend for my thesis work to highlight machine learning methods used in recommender systems. I will explore issues including privacy and trust modeling which relate to recommender systems, though the extent to which my thesis discusses these topics will depend on my research findings and my interest level during research work.
For the second part of my thesis, I plan to design my own recommender system or offer suggestions towards more efficient or creative systems. My system will either explore a topic not yet focused on in the world of recommender systems (e.g. Bryn Mawr course recommendations) or will implement a new type of recommending technique (which would likely be a novel hybrid of existing techniques). The exact topic or technique will be determined after further research into recommender systems.
While my current goal is to design and potentially implement such a recommender system, I leave the possibility open for altering that goal at some point in my research. I may deem rendering a recommender system to be out of the scope of this project. Or, I may decide that continuing research and synthesis of the (vast) amount of material currently available on recommendation systems to be more important or of greater interest than designing a new system. In that case, I plan to provide a wide-ranging analysis of the current systems as well as recommendations for further improvement within the field.
At the start of a given project, it makes sense to begin with questions about which the researcher is curious; hence I will end the discussion of my thesis topic just where my thesis research begins. Questions I seek to answer in this project include the following. The most basic question is to understand what a recommendation system is. Past that, how do these systems work? What is the history of recommender systems and in what ways have recommender systems developed? What are the best currently standing techniques in the field, according to various criteria (e.g. efficiency, usability, applicability)? What insights do the best current techniques consider to create a good system (e.g. does the computer coding for a good system implement an underlying concept from psychology)? Do different types of recommender systems fit better for different recommendations? What techniques and topics are most heavily concentrated on in the field of recommendation systems, and what requires continuing work? There are many more questions I could and want to ask, but these will at least provide a beginning framework for my project.