In information extraction systems, it is essential to identify named entities in order to provide the knowledge to be extracted. However, it is not easy to identify these names, because they involve unknown words, and hence the strategy of listing candidates would not work. Also, it is sometimes hard to determine the category of named entities, like distinguishing a person name from a company name.
In English, several rule-based researches show good performance using special dictionaries and limited contextual information. But Korean has no character type information like English. As a result, it is difficult to recognize candidates of named entities and determine the boundary of them.
This thesis proposes the rule-based method for recognizing named entities, especially person names, organization names or locations in Korean texts. The method uses various dictionaries for proper names, prefix, suffix, verb's subcategorization, etc. and it consists of 4 stages according to the type of information used. At the first stage, the information inside an ejeol is used. At the next stage, limited contextual information about surrounding words is considered. At the third stage, subcategorization information of verbs is used to disambiguate the categories of named entities. At the last stage, information between named entities is used for merging named entities recognized in previous stages.
Various experiments have conducted and the contribution of each stage was evaluated. The experimental result shows 90.4% in precision and 83.4% in recall for Korean news articles.
