Local Machine REFERENCES [1] e golden record. URL hps://voyager.jpl.nasa.gov/golden-record. [2] S. Bu¨cher and C. L. A. Clarke. Indexing time vs. query time: Trade-os in dynamic information retrieval systems. In Proc. CIKM, pages 317-318, 2005. [3] K.-H. Le, O. Slaery, R. Lu, X. Tang, and V. McCrary. e state of the art and practice in digital preservation. J. Res. Natl. Inst. Stand. Technol., 107(1):93, 2002. [4] H. R. Mohammad, K. Xu, J. Callan, and J. S. Culpepper. Dynamic shard cuto prediction for selective search. In Proc. SIGIR, pages 85-94, 2018. [5] K. Peltonen. Adding full text indexing to the operating system. In Proc. ICDE, pages 386-390, 1997.
Interests
Cache
Crawl User Interests
Cache User !ery 102) will provide an answer set too large for consumer hardware to cache. Implicit feedback of oine search user satisfaction might be used to form smaller, more precise topical shards for search-engines; further improving selective search eciency.
Varying B in a top-B retrieval is likely to alter the user satisfaction for the lengths of documents returned. For smaller B, textbooks may be the most eective way of ensuring coverage of a topic. For larger B, smaller documents might be easier for a user to navigate dierent facets of their information need. TREC collections may be unhelpful in evaluating ΔB due to pooling bias towards shorter documents, and so a collection with judgments for documents of diverse lengths may be needed to observe ΔB and its eect on user satisfaction in the proposed oine search scenario.
Automatically forming oine web caches for a prescribed set of topics on consumer devices may increase the productivity of users without Internet access. is poses new research questions in IR eciency, eectiveness and evaluation.
Acknowledgements. anks to J. S. Culpepper, A. Moat and J. Mackenzie and anonymous reviewers for their feedback.