Extraction Research Of Chinese Temporal Keywords

Temporal keyword is a natural language phrase, it is used to represent the time points and time interval of documents. At present, in the field of natural language processing, the question to answer or information retrieval, the temporal keywords has been applied widely, and identification of temporal keywords directly affect the use effect of temporal information:automatic question answering system can answer questions about temporal, such as "what time is it" or "when"; in the topic detection and tracking tasks, temporal keywords can be used to determine the order of something happen; In machine translation, depending on the time information order can become more easier for reading or which make the results more smooth.Recently, temporal keyword recognition methods include two categories mainly:The method based on rules and the method based on statistical learning method. This paper will respectively to discuss two research methods above. This article first introduces the research status of Chinese temporal keywords identify, and then respectively use one of them to identify temporal keywords, compare the performance of two methods according to the results of the experiment at last.At present, in the temporal keywords domain, the use of classic method based on rules is the most widely, the paper also makes a deep exploration of rule-based method. First, the paper put forward the thought of describe the tense elements, divide temporal keyword categories and canonical form; On this basis, combine regular expressions and Trie tree structure to build temporal phrase recognition tree which can automatic recognition and classification of Chinese temporal keywords.For the method of statistical learning, this paper firstly analyzes the text sentence structure and then combine analyses results and phrase structure tree to put forward a method of divide phrase in text, which transform text into phrases so that to determine the phrase boundary; Then, we make every of result denoted in vector:On this basis, the introduction of spectral clustering of clustering algorithm, so that recognition problem can be converted to the clustering problem.Finally, the effect of experiment using Chinese Emergency Corpus is good. The method based on statistical learning in accuracy and recall rate and F values are all slightly higher than the recognition method based on rules.