中文摘要：

用统计的方法从单文本中自动抽取关键短语。在实验中验证了频度、首位置作为特征的有效性。用各种方法过滤非法词串，综合短语位置和统计特征对候选短语进行权重计算，并依据关键短语分布规律选择关键短语。另外，通过分析关键短语分布特点为Ⅳ元短语在过滤、按比例选择方面提供了依据。获得了比较好的实验结果：TOP5精确率21．80％，召回率28．27％，F-measure25％；TOP10精确率17．10％，召回率44．50％，F-measure30．80％。

英文摘要：

A statistics- based approach is proposed for automatically extracting keyphrases from Chinese scientific documents. Term frequency and first occurence are valid in the approach. Several filtering methods are utilized to filter invalid terms. Text feature and statistic information are combined to select keyphrases. The final keyphrases are ouputed on the basis of the actual distribution of keyphrases. Keypharases distribution information provides some experiment proof for N-gram fiheration and output by proportion. The experimental results achieve the performance of TOP 5 precision 21.80% ,recall 28.27% ,F-measure 25.00% and TOP 10 precision 17.10% ,recall 44.50% , F-measure 30.80%.