Fast Approximate Title Matching

Given a set of titles and a query, we want to find the title with the largest similarity to the query. The queries may contain spelling mistakes and words which are not part of the correct entry. Furthermore, no word separators in the query are expected and words of the correct title may be missing. Titles are usually short and consist of multiple words and numbers. We analyse several existing similarity measures like Edit distance and Jaccard similarity. Besides the quality, efficiency is the other requirement of our algorithm and therefore we analyse and compare the best existing algorithms for efficient fuzzy match [CGGM03] and efficient duplicate detection [XWLY08]. We propose a new similarity measure as well as algorithm for its efficient computation in order to improve the existing algorithms in terms of quality and efficiency. We show on two different collections that our algorithm is in most of the cases more efficient than the related work, yet achieves better quality.