欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
首先用向量空间模型工具Lucene从全部网页正文信息中检索，再用语言模型工具Lemur对结果集进行重排序，然后将两次的结果进行融合，远回融合结果的前1000篇文档作为最终结果集。构造查询输入时，从主题的〈title〉字段和〈dese〉字段选择关键词，并依据tf＊idf的思想对关键词赋予权值。时正式评测的50个主题集检索，获得的三项评价指标为：程序自动构造查询时，MAP=0．3107，P@10=0．624，R-Preeision=0．3672；人工构造查询时，MAP=0．3538，P@10=0．684，R-Preelsion=0．4078。
英文摘要：
A rough set of relevant results is returned by Lucene, which based on vector space model, after searching all web pages, and is then reranked by Lemur, a language model based tool, to form a second set of relevant results. These two sets are combined by a linear interpolation into one set afterward and the top 1000 pages in it are returned as final results. When formulating queries from topics, key words of queries.are selected from 〈 title 〉 fields and 〈 desc 〉 fileds of topics, and weights of them are calculated using a modified ff ＊ idf method. In the official evaluation on 50 topics, MAP 0. 3107, P@ 10 0. 624, R-Precision 0. 3672 and MAP 0. 3538, P@ 100. 684, R-Precision 0. 4078 are achieved with queries constructed automatically and artificially respectively.
同期刊论文项目
同项目期刊论文
期刊信息
位置：