Efficient Web Searching Using Temporal Factors

Abstract

Web traversal robots are used to gather information periodically from large numbers of documents distributed throughout the Web. In this paper we study the issues involved in the design of algorithms for performing information gathering of this kind more efficiently, by taking advantage of anticipated variations in access times in different regions at different times of the day or week. We report and comment on a number of experiments showing a complex pattern in the access times as a function of the time of the day. We look at the problem theoretically, as a generalisation of single processor sequencing with release times and deadlines, in which performance times (lengths) of the tasks can change in time. The new problem is called Variable Length Sequencing Problem (VLSP). We show that although the decision version of VLSP seems to be intractable in the general case, it can be solved optimally for lengths 1 and 2. This result opens the possibility of practicable algorithms to schedule searches efficiently when expected access times can be categorised as either slow or fast. Some algorithms for more general cases are examined and complexity results derived.

Research partially supported by DFGSonderforschungsbereich 376 “Massive Parallelität: Algorithmen, Entwurfsmethoden, Anwendungen. The research of this author was partly done while visiting the University of Liverpool.

Supported in part by NUF-NAL (The Nuffield Foundation Awards to Newly Appointed Lecturers) award.

Supported by EPSRC grant GR/L/77089