Geotagging Location Information Extracted from Unstructured Data (Short Paper)

Location information is an essential element of location-based services and is used in various ways. Unstructured data contain different types of location information, but coordinate values are required to determine the exact location. In Twitter, a typical social network service (SNS) platform of unstructured data, the number of geotagged tweets is low. If we can estimate the location of text by geotagging a large number of unstructured data, we can estimate the location of the event in real-time. This study is a base study on extracting the location information by using the named entity recognizer provided by the Exobrain API and applying geotagging to unstructured data in Hangul (Korean). We used Chosun news articles, which are grammatically correct and well organized, instead of tweets to extract three location-related categories, namely "location," "organization," and "artifact". We used the named entity recognizer and geotagged each sentence in combination of the fields in each category. The results of the study showed that 61% of the 800 test sentences did not have the location-related information, thus hindering geotagging. In 11.75% of the test sentences, geotagging was possible with only the given location information extracted using the named entity recognizer. The remaining 27.25% of the sentences contained information on more than two locations from the same subcategories and hence required location estimation from candidate locations. In future research, we plan to apply the results of this study to develop location estimation algorithm that makes use of the extracted location-related entities from purely unstructured data such as that on SNSs.