Why Bigger Windows Are Better Than Smaller Ones

We investigate the use of multi-term query concepts to improve the performance of text-retrieval systems that accept \natural-language" queries. A relevance feedback process is explained that massively expands an initial query with single and multi-term concepts. The multi-term concepts are modelled as a set of words appearing within windows of varying sizes. Experimental results suggest that windows of larger size yield improvements in average precision. The reason for this improvement is explored. A window size relaxation process that yields a signiicant reduction in expanded query size with no performance loss is also described.