Effects of prior knowledge on the effectiveness of a hybrid user model for information retrieval

摘要

To quickly find relevant information from huge amounts of data is a very challenging issue for intelligence analysts. Most employ their prior domain knowledge to improve their process of finding relevant information. In this paper, we explore the influences of a user's prior domain knowledge on the effectiveness of an information seeking task by using seed user models in an enhanced information retrieval system. In our approach, a user model is created to capture a user's intent in an information seeking task. The captured user intent is then integrated with the attributes describing an information retrieval system in a decision theoretic framework. Our test bed consists of two benchmark collections from the information retrieval community: MEDLINE and CACM. We divide each query set from a collection into two subsets: training set and testing set. We use three different approaches to selecting the queries for the training set: (1) the queries generating large domain knowledge, (2) the queries relating to many other queries, and (3) a mixture of (1) and (2). Each seed user model is created by running our enhanced information retrieval system through such a training set. We assess the effects of having more domain knowledge, or more relevant domain knowledge, or a mixture of both on the effectiveness of a user in an information seeking task.