Item

Abstract

We describe CompleteSearch, an interactive search engine that offers the user a variety of complex features, which at first glance have little in common, yet are all provided via one and the same highly optimized core mechanism. This mechanism answers queries for what we call context-sensitive prefix search and completion: given a set of documents and a word range, compute all words from that range which are contained in one of the given documents, as well as those of the given documents which contain a word from the given range. Among the supported features are: (i) automatic query completion, for example, find all completions of the prefix “seman” that occur in the context of the word “ontology”, as well as the best hits for any such completion; (ii) semi-structured (XML) retrieval, for example, find all emailmessages with “dbworld” in the subject line; (iii) semantic search, for example, find all politicians which had a private audience with the pope; (iv) DB-style joins and grouping, for example, find the most prolific authors with at least one paper in both “SIGMOD” and “SIGIR”; and (v) arbitrary combinations of these. The prefix search and completion mechanism of Complete- Search is realized via a novel kind of index data structure, which enables subsecond query processing times for collections up to a terabyte of data, on a single PC. We report on a number of lessons learned in the process of building the system and on our experience with a number of publicly used deployments.