Relevance Feedback for Web Searches

The growing importance of the World Wide Web in modern daily life and its consequent rapid growth has resulted in the problem of how to effectively manage and query this vast information source. Currently, the primary means of navigating the Web is through the use of search engines such as AltaVista and Yahoo!. These search services tend to return many irrelevant results to user queries. This is because users often submit queries of poor quality due to the difficulty they experience in transforming their information needs into the required query language for the web search engine. To help solve this problem, this project designed and implemented a Java applet called the "Search Assistant." This software builds a layer of functionality on top of the AltaVista search engine and uses a process called "relevance feedback" to facilitate the formation of more effective queries. Relevance feedback works under the assumption that identifying the relevance of documents is more natural than manually formulating a good query. The Search Assistant draws on this concept, and allows users to submit a query, and select the relevant documents from the returned set of results. It then analyses these documents, looking for frequently occurring words. These words are then provided as suggestions to the user for improving the original query. Testing the effectiveness of this technique required a comparison between the quality of results obtained from the basic AltaVista search engine and the Search Assistant. This kind of comparison is difficult to make since the question of what exactly constitutes a "better" set of results arises. Nonetheless, the results of simple, superficial tests revealed that using the Search Assistant, novice users took less time and used fewer queries to find a specified piece of information. These results are promising, and indicate that further study into the use of relevance feedback in web searches is warranted. Future work should focus on performing a more thorough evaluation of the system, and improving the performance of the software.

Thesis originally deposited on 2011-12-29 in version 1.28 of Libra. This thesis was migrated to Libra2 on 2016-11-30 15:23:03.