A critical investigation of recall and precision as measures of retrieval system performance
ACM Transactions on Information Systems (TOIS)
Expected reading effort in focused retrieval evaluation
Information Retrieval
| Hi-index | 0.00 |
Standard Information Retrieval (IR) metrics are not well suited for new paradigms like XML IR in which retrievable information units are document elements. These units are neither predefined nor independent, and the elements returned by IR systems may overlap and contain near misses. Part of the problem stems from the classical hypotheses on the user behaviour that do not take into account the structural or logical context of document elements or the possibility of navigation between retrievable units. The Expected Precision Recall with User Model (EPRUM) metric is based on a more realistic user model which encompasses a large variety of user behaviours. In this paper, we present the EPRUM metric used for evaluating the official submissions of INEX 2005 and detail the settings we used. We do not present the full derivation of the EPRUM metric but we give a thorough example of its computation along with the complete set of formulas needed to compute precision at different recall values. We also discuss the implication of such a metric on several key problems of XML Information Retrieval as the notion of the ideal list and the problem of the overlap.