This work’s title should more aptly start with “Performance improvement for full-text search systems” because it focuses on that area of application. The lengthy paper (25 pages) revisits and expands on previous work aimed at incremental improvements in medium-scale full-text search systems, where medium-scale is defined as an environment where “only a fraction of the [inverted] index is maintained in memory while the remaining fraction is stored in secondary storage.” (“Industry-scale” systems instead cache the entire index.)

Its contribution is the introduction, accompanied by some experimental testing, of a consolidated cacheable index to replace a single-term inverted index cache and a separate term-pairs index cache. The structure of the new index is examined in detail in the body of the work and put to test against somewhat dated, but publicly available, datasets and free-text queries. Metrics are based on the performance of total retrieval, as opposed to cache hit ratios: the argument is that a cache hit ratio is an inadequate indication of overall system performance. The metrics presented, as expected, support the claim of incremental performance. On the other hand, overall system performance is a complex benchmark: its value is simplistically obvious, but the applicability to different experimental circumstances is subject to much interpretation. It would be interesting if the authors chose to assess how suitable this new index is for search systems other than “full-text.” Perhaps the title’s omission of this constraint is not incidental, and we may see experiments in other contexts in the future.