A Full-Text Learning to Rank Dataset for Medical Information Retrieval

I let q ∈ {0, 1} and d ∈ {0, 1} be query and document vectors, dimensions indicating word occurrence for dictionaries of size Q and D I score function f(q,d) = qWd = ∑Q i=1 ∑D j=1 qiWijdj , where W ∈ RQ×D is a matrix of word associations between query and document dictionaries I R is a set of tuples (q,d+,d−), document d being more relevant for query q than d− I relevance rank rq,d, rank di erence m(q,d+,d−) = rq,d+ − rq,d− I RankBoost: Lexp = ∑ (q,d+,d−)∈R m(q,d+,d−)ef(q,d +)−f(q,d−)