基于浅层语义分析的主题事件的时间识别

时间识别是自然语言处理中极其重要的课题。事件中与主题相关的时间信息体现了事件在时间维度的主题特征。当前面向事件的时间识别大多是基于句子或短语的,并采用静态时间值机制。本文提出了一个面向主题事件的时间识别模型。该模型采用参考时间动态选择机制对时间表达式规范化。结合事件抽取和浅层语义分析,将浅层语义分析结果和时间表达式进行映射,将基于句子或短语的时间识别转化为基于篇章的时间识别,从而识别主题事件片段的时间。实验表明所提出的方法使主题事件片段的时间识别的性能提高了9.6%。

Temporal recognition for topic event based on shallow semantic parsing

Temporal recognition is a key subject in natural language processing community. The topic-related temporal information reflects the topic feature of topic events on temporal dimensionality. Most temporal recognition for events was sentence-oriented or phrase-oriented and employed static time-value machine. A temporal recogtion model for topic events was proposed in this paper. Temporal expressions were normalized with reference time dynamic-choosing mechanism in this model. Combining event extraction and shallow semantic parsing, semantic roles were mapped to temporal expressions. Document-oriented temporal recognition was implemented using sentence-oriented or phrase-oriented temporal recognition, consequently, temporal recognition for topic event segments was realized. Results show that performances of temporal recognition for topic event segments are improved by 9.6%.