Improving Search Engines via Classification

In this dissertation, we study the problem of how search engines can be improved by making use of classification. Given a user query, traditional search engines output a list of results that are ranked according to their relevance to the query. However, the ranking is independent of the topic of the document. So the results of different topics are not grouped together within the result output from a search engine. This can be problematic as the user must scroll though many irrelevant results until his/her desired information need is found. This might arise when the user is a novice or has superficial knowledge about the domain of interest, but more typically it is due to the query being short and ambiguous. One solution is to organise search results via categorization, in particular, the classification. We designed a target testing experiment on a controlled data set, which showed that classification-based search could improve the user’s search experience in terms of the numbers of results the user would have to inspect before satisfying his/her query. In our investigation of classification to organise search results, we not only consider the classification of search results, but also query classification. In particular, we investigate the case where the enrichment of the training and test queries is asymmetrical. We also make use of a large search engine log to provide a comprehensive topic specific analysis of search engine queries. Finally we study the problem of ranking the classes using some new features derived from the class. The contribution of this thesis is the investigation of classification-based search in terms of ranking the search results. This allows us to analyze the