Efficient Time-Travel on Versioned Text Collections

The availability of versioned text collections such as the Internet Archive opens up opportunities for time-aware exploration of their contents. In this paper, we propose time-travel retrieval and ranking that extends traditional keyword queries with a temporal context in which the query should be evaluated. More precisely, the query is evaluated over all states of the collection that existed during the temporal context. In order to support these queries, we make key contributions in (i) defining extensions to well-known relevance models that take into account the temporal context of the query and the version history of documents, (ii) designing an immortal index over the full versioned text collection that avoids a blowup in index size, and (iii) making the popular NRA algorithm for top-k query processing aware of the temporal context. We present preliminary experimental analysis over the English Wikipedia revision history showing that the proposed techniques are both effective and efficient.