A faceted approach to information retrieval

Information retrieval is the core of many applications such as natural language understanding, document clustering, etc. The Boolean model is one of the basic approaches of information retrieval and faceted query belongs to the category of this model. Unfortunately, the characteristic of binary decision has hindered the faceted query from prevailing use. One of the disadvantages is that it often responds to the query with a large quantity of data that should be filtered manually. In this paper we suggest a new faceted method, Facet Analysis Method, to cope with the problem, in which we give Boolean operators an algebraic interpretation to facilitate partial match that is the key feature of Information retrieval. In our approach, a query specifies an ideal document to be searched and retrieval is a ranking among documents in the collection based on their similarities to the ideal document. The Facet Analysis Method enjoys the property of reflexivity, symmetry, and conditioned transitivity, permitting a large variety of applications. To illustrate the applicability and novelty of our approach, we have applied Facet Analysis Method to construct a high-precision, special-purpose search engine, the e-Detective system, which is able to collect crime information from the Internet automatically. The match program of the search engine is described where we first organize search concepts to represent our data request and use Facet Analysis Method to calculate similarities between the search target and Web pages. We further describe the way of error correction and feedback mechanism for tuning term weights to enhance the retrieval efficacy. The system is tested by a difficult and complex search task, finding Web pages auctioning pirated compact discs. The experiment result is evaluated in the notion of the well-recognized recall/precision measure, where we obtain the results with average search precision 0.59, showing the superiority of this new method.