Robustness of well-designed retrieval performance measures under optimal user behavior
✍ Scribed by Conlon, John R. ;Conlon, Sumali J.
- Publisher
- John Wiley and Sons
- Year
- 1998
- Tongue
- English
- Weight
- 101 KB
- Volume
- 49
- Category
- Article
- ISSN
- 0002-8231
No coin nor oath required. For personal study only.
✦ Synopsis
This article considers a simple model of an information user behavior, and based on these, derives three alternaretrieval (IR) technology, and uses this model to examine tive performance measures.
performance measures for IR systems. In particular, we Thus, different assumptions about behavior lead to difshow that, if a user is employing an IR system optimally, ferent performance measures. Furthermore, it is useful then reasonable performance measures will tend to be to base these behavioral assumptions on a model of an relatively robust, or insensitive, to small variations in assumptions about the user's behavior. A simple empirical optimizing user, for at least two reasons. First, the optiillustration of this approach is reported, though more mized value of the user's objective function offers a natuwork will be needed to determine exactly how robust ral point of comparison with proposed performance meaperformance measures will be in practical applications.
sures (see Cooper, 1973a). Second, user optimization has interesting implications for the properties of performance measures. Thus, in this article we show that, if a user * To whom all correspondence should be addressed.