自动文摘基集语句的提取与润色的数学模型*

针对统计和理解相结合的自动文摘方法,提出了一种新的内容词、有效词和特征词的动态加权函数以及句子重要性的动态加权函数.鉴于基于统计的自动文摘结果常常出现语句间缺乏连贯性及信息冗余的问题,设计了句间语义距离测试函数,并通过大量实验确定语句间语义距离的上限和下限.上限用于控制语句间的逻辑联系,下限用于解决文摘结果信息冗余的问题.实验结果证明,该模型能有效地提取文章中的重点语句,且很好地解决了统计文摘语句不连冠的瓶颈问题.

Mathematical Model for Abstracting and Embellishing Sentences in Automatic Abstraction

Two kinds of dynamic weighting functions were presented.One was content words,effective words and characteristic words weighting function and the other was sentence weighting function.Considering the problems of lack of logic between sentences and information redundancy were usually exist in the statistical-based abstraction result,a distinguish function based on semantic distance was purposed.The upper limit of such a semantic distance was used in controlling the logic continuity between sentences obtained from a mass of experiments as well as the lower one was used in information redundancy resolving.Combining the dynamic weighting functions,the semantic-distance-based distinguish function and the automatic text structure analysis method presented above,a basic abstraction was generated.Then,the basic abstraction was condensed.This integrated approach demonstrated the impressive effectiveness by its outstanding performance in experiment.