Graduation Semester and Year




Document Type


Degree Name

Master of Science in Computer Science


Computer Science and Engineering

First Advisor

Manfred Huber


The current search engines available on the Net are generic in nature. They do not consider user preferences and treat all users information needs in the same way. As a result they frequently return a large number of links, that do not meet the user's information need. This requires more searching to find what the user is looking for. For example if a user is interested in a particular game, e.g. cricket, and enters the query world cup, a generic search engine would return links of all the sports that hold a world cup. The user has to browse through a considerable number of non-relevant pages before he is able to get to links he is looking for. This is also because search engines don't have the ability to ask a few questions and they also can not rely on judgment and past experience to rank web pages, in the way humans can. This raises the issue of customizing a generic search engine to consider user preferences. There have been a number of attempts in the past to personalize the search for information on the Net. These systems are based on relevance feedback methods, similarity measures, or storing a user profile explicitly or implicitly. Some of them have shown impressive results in query expansion and providing pages similar to the user's interest. Here we propose a novel system for personalizing web search. Our method is based on creating a user profile as he performs his routine searches in a given user category. In this customization, the user is allowed to create personal user categories within which he could search for information on the Net without getting too many irrelevant links in his search results. The application enhances the query by adding words that are generated from the user profile stored for a particular user category. It uses information about the probabilistic co-occurrence of words in the user profile with other words in the query as a measure for adding words. The snippets returned from the generic search engine are then classified on the basis of the user profile and are re-ordered according to a measure representing the interest of the user. With our method the user not only gets personalized query expansion but also receives re-ordered search results from the search engine. In this system the query expansion is made to optimize the estimated returns of the search engine, taking into account the classification accuracy and re-ranking of results. The system add words that would give the user the best possible returns according to his user profile.


Computer Sciences | Physical Sciences and Mathematics


Degree granted by The University of Texas at Arlington