From Html to Usable Data Problems in Meaning and Credibility in the Www Position Paper

Problems in meaning and credibility in the WWW Position paper Joshua Grass and Shlomo Zilberstein Computer Science Department University of Massachusetts Amherst, MA 01003 U.S.A. fjgrass,[email protected] Introduction This paper describes issues in information integration that relate to Value-Driven Information Gathering (VDIG)(Grass & Zilberstein 1996; 1997; 1998). Valuedriven information gathering is the process of querying multiple information sources for information items which are used to make a decision. VDIG works in a resource-bounded environment where it is not possible to gather all the information needed to make a perfect decision. Instead, VDIG keeps statistics on the response expectation of particular sites and the decision model can operate with partial information. The process is referred to as value-driven, because the algorithm determines the value of a query for potential sites and queries the best candidate. The value of a query is determined using the value of information from the decision model, the expectations of a site returning a result at any time in the future, the information the system already knows, and the cost function, which represents the resources the system is allowed to spend in order to make the decision. In this paper we will focus on one aspect of the value-driven process, taking raw information from sites and converting it into a form usable by our decision model1. The decision model we use is an in uence diagram which uses information passed to it from an extraction engine to instantiate nodes. At the present time we rely on hand coding extraction algorithms that convert web sites into a list of feature/value tuples. For our prototype system this approach works well, and we have received good results after testing the system in the domain of making a decision about purchasing a digital camera. Numerous other groups Support for this work was provided in part by the National Science Foundation under grants IRI-9624992, IRI9634938, and INT-9612092. 1It should be noted that although we are dealing with unifying information from distinct information sources in the context of value-driven information gathering, these techniques are valid in domains with few information sources. are developing much more open-ended extraction engines (Doorenbos, Etzioni, & Weld 1997; Ashish & Knoblock 1997a; 1997b; Konopnicki & Shmueli 1995; Genesereth, Keller, & Mueller 1996). Figure 1 shows the in uence diagram use by VDIG to evaluate digital cameras.