<XML><RECORDS><RECORD><REFERENCE_TYPE>10</REFERENCE_TYPE><REFNUM>7718</REFNUM><AUTHORS><AUTHOR>Cai,D.</AUTHOR><AUTHOR>Van Rijsbergen,C.J.</AUTHOR></AUTHORS><YEAR>2004</YEAR><TITLE>Relevance Weight Revisited</TITLE><PLACE_PUBLISHED>DCS Tech Report</PLACE_PUBLISHED><PUBLISHER>N/A</PUBLISHER><PAGES>1-8</PAGES><ISBN>TR-2004-173</ISBN><LABEL>Cai:2004:7718</LABEL><KEYWORDS><KEYWORD>Information retrieval</KEYWORD></KEYWORDS<ABSTRACT>Relevance weight, developed in the late 70s and early 80s as a device for measuring term value, has been one of the most commonly proposed weighting functions for information retrieval (IR). It has been extensively explored and implemented in retrieval systems, and continues to be often used in today's IR. However, the probability estimation problem has yet to be satisfactorily resolved. This article discusses the problem and presents a new, simple but more effective estimation method. An analysis of the error caused by the 0.5 adjustment factor used in the conventional estimation method is given. </ABSTRACT></RECORD></RECORDS></XML>