A Multi-level Hybrid Method for Dynamic Extraction and Normalization of Chinese Temporal Expressions

Since the semantic meaning of Chinese is intricate and complicated, traditional methods have deficiency in extracting the temporal information from Chinese texts. By adopting a way that combines the rule-based with the statistical-based methods, this paper puts forward a novel method for Chinese Temporal Expression Normalization based on context. The construction of this approach mainly employs sieve algorithm based on Bi-LSTM and benchmark time dynamic selection algorithm based on context. In this way, the dependency between contextual temporal information can be effectively used, and the "absolute time" and the "offset time" proposed in this paper also can be adopted for reasoning and calculation. Through the experimental testing in the TempEval2 task of SemEval-2010, we get precision, recall and F-score values that are 96.74%, 89.29% and 92.87%, respectively. Compared with the traditional rule-based method, the approach in the paper achieves better results overall, and solves the discrimination difficulty in part of the temporal phrases by combining the contextual semantics.