MPI-I-2010-5-003

Modern text analytics applications operate on large volumes of temporal text data such as Web archives, newspaper archives, blogs, wikis, and micro-blogs. In these settings, searching and mining needs to use constraints on the time dimension in addition to keyword constraints. A natural approach to address such queries is using an inverted index whose entries are enriched with valid-time intervals. It has been shown that these indexes have to be partitioned along time in order to achieve efficiency. However, when the temporal predicate corresponds to a long time range requiring the processing of multiple partitions, naive query processing incurs high cost of reading of redundant entries across partitions. We present a framework for efficient approximate processing of keyword queries over a temporally partitioned inverted index which minimizes this overhead, thus speeding up query processing. By using a small synopsis for each partition we identify partitions that maximize the number of final non-redundant results, and schedule them for processing early on. Our approach aims to balance the estimated gains in the final result recall against the cost of index reading required. We present practical algorithms for the resulting optimization problem of index partition selection. Our experiments with three diverse, large-scale text archives reveal that our proposed approach can provide close to 80% result recall even when only about half the index is allowed to be read.