A study of parameter tuning for term frequency normalization

摘要: Most current term frequency normalization approaches for information retrieval involve the use of parameters. The tuning these parameters has an important impact on overall performance system. Indeed, a small variation in involved parameter(s) could lead to precision/recall values. are dependent document collections. As consequence, effective parameter value cannot be obtained given new collection without extensive training data. In this paper, we propose novel and robust method parameter(s), by measuring effect within query terms. illustration, apply our Amati \& Van Rijsbergen's so-called 2. experiments ad-hoc TREC-6,7,8 tasks TREC-8,9,10 Web tracks show that is independent collections able provide reliable good performance.