My research focuses on extracting events and their attributes. In particular, my Ph.D. research concentrates on a single event---possession. Possession is an asymmetric semantic relation, where an entity (possessor) possesses another entity (possessee). Throughout my Ph.D., I work on automatically extracting possession and its attributes. In this section, I briefly explain my research published so far. For more details on my research read my publications.
In this work, we present a new corpus for tracking concrete objects as they change hands over time. We then generate a timeline using the corpus which includes possessors, the certainty of possession, and the duration of possession. In addition to possessors that are named entities, we extract possessors that are not identified by conventional NERs. We also present a strategy to place possessors who do not have an explicit time mention on the timeline.
In this research, we work on Wikipedia articles on historical artifacts, which mostly includes paintings like the Guernica and the Mona Lisa. The goal is to identify all possessors of a chosen artifact and the years in which they were possessed. We present a mechanism to pair the possessors and years, along with a LSTM ensemble to predict possession relations.
In this work, we extract the presence of possession in sentence level. We then identify the type of possession and temporally anchor the possession relation with respect to the verb in the sentence. We automatically extract these relations using SVM and deep learning models.