Rule and Maximum Entropy Model Based Method of Temporal Expression Recognition

摘要

Temporal relation exits between concepts of time and event. It provides a natural clue for information organization. This paper studies the method of temporal expression recognition on the basis of information extraction and temporal denotation. We analyze the features of temporal expressions and conclude that the temporal expressions are constructed by temporal expression baseline and temporal expression director, which respectively indicate time reference and time point or time range and are the theory reference of recognition rules. Rule based method is used to extract explicit time information because of the certainty feature of it. Machine learning based method is used to recognize the implicit time information because there are no definite rules in it. The experiment shows the blended method is of relatively high accuracy and has higher recall rate than an individual method.