Applicability of recommender systems to medical surveillance systems
Ricardo Gomes Lage, Frederico Araújo Dur�ão, Peter Dolog, Avaré Stewart · 2011
In traditional Event-Based Surveillance systems, documents are continuously monitored for health threats detection and reporting, following a user-defined set of rules. This monitoring may result in a large set of events detected, overwhelming the user. In addition, such systems would not consider aspects similar to the user's set of defined rules other than exact matches. This prevents the user from discovering similar health threats that could also be of interest. In this paper, we perform a two-fold evaluation of a recommendation algorithm that infers the user preferences from his set of defined rules. In the first part of the evaluation, the participants evaluate the recommendations whether they match their interest in the H1N1 virus. For this evaluation, we achieved a precision rate of 0.81. In the second part, we conduct a group discussion with health surveillance experts where they provide feedback on the recommendations received. We present these results in terms of what should be recommended, and how the recommendations should be evaluated and take place in a surveillance system.