Information extraction from written documents by relating people and events

In this thesis, the proposed method transforms unstructured texts into structured form using clustering and classification techniques which extract person and location relations depending on time and employ this structured form to generate event based queries on personal trajectory. New York Times corpus which includes 539 thousands of news mostly about politicians is used in training and testing process. Person, location and time phrases are extracted from sentences and normalization of named entities is done in order to track the person trajectory. After that, person-location pairs are extracted on the one-sentence level granularity. Finally, the personal trajectory is obtained based on the action verb phrases between person-location pairs.Time-location pairs are named as events and relation of events with people are extracted from documents. In order to evaluate the method, golden standard dataset is generated and for the different modules of the method, F-Score and Accuracy are calculated.