Probabilistic retrieval: thresholding for automatic filtering
Stephen E. Robertson · 1999
The paper discusses work in progress on a particular problem that arises in filtering systems. The work is discussed more fully and with more technical detail by S.E. Robertson et al. (1999) and S.E. Robertson and S. Walker (2000). It is concluded that: adaptive filtering is more difficult than ranked retrieval; the thresholding task is different from the scoring task, and should be kept separate; in the early life of a profile, probabilistic modelling is needed and seems to be feasible; we have only half a model for the value of feedback tradeoff; and we need a large number of experiments to check out some of the variables. (4 pages)