Master of Science (MS)
Commercial search engines do not return optimal search results when the query is a long or multi-topic one . Long queries are used extensively. While the creator of the long query would most likely use natural language to describe the query, it contains extra information. This information dilutes the results of a web search, and hence decreases the performance as well as quality of the results returned. Kumaran et al.  showed that shorter queries extracted from longer user generated queries are more effective for ad-hoc retrieval. Hence reducing these queries by removing extra terms, the quality of the search results can be improved. There are numerous approaches used to address this shortfall. Our approach evaluates various versions of the query, thus trying to find the optimal one. This variation is achieved by reducing the query length using a combination of n-grams assisted query selection as well as a random keyword combination generator. We look at existing approaches and try to improve upon them. We propose a hybrid model that tries to address the shortfalls of an existing technique by incorporating established methods along with new ideas. We use the existing models and plug in information with the help of n-grams as well as randomization to improve the overall performance while keeping any overhead calculations in check.
Irani, Jehann Kersi, "N-Grams Assisted Long Web Search Query Optimization" (2011). Master's Projects. 211.