关键词: 地址语义, 要素特征字, 转移概率, 无词典

Abstract:

There are a large number of Chinese address text in the Internet that contains rich spatial location information. In order to obtain the address location information in the text more effectively, we propose a Chinese address location information recognition method based on address semantics. According to the statistics of word frequency of the training corpus, we obtain a set of address feature words and word transition probability. Then, we construct a feature word transition probability matrix. Finally, combining with the string maximum joint probability algorithm, we put forward an address recognition method which does not depend on address dictionary and tagging of the part of speech. Experimental results show that the exact match rate of the method is 76.85% for ambiguous Chinese addresses with prominent feature words, and the recognition accuracy is 93.11%. Compared with the mechanical matching algorithm and the methods for constructing the transition probability matrix based on experience, experimental results verify the feasibility and effectiveness of the proposed method.

Key words: address semantics, feature character word, transfer probability, without dictionary