Inducing a semantically rich nested event model

Abstract

Research has revealed that getting data with named entities (NEs) labels are laboured intensive and costly. This paper is proposing two approaches to enable NE classes to be added to the semantic role label (SRL) predicateargument structure of Nested Event Model. The first approach associates SRL to Named Entity Recognition (NER), which is named as SRL-NER, to tag the appropriate entity class to the simple argument of the model. The second approach associates SRL to NER by fine-tuning entities in complex argument structures with Automatic Content Extraction (ACE) structure. This approach is called SRL-ACE-NER. Stanford NER tool is used as the benchmark for evaluation. The result shows that the proposed approaches are able to recognize more PERSON entities. However, the approaches are not able to recognize LOCATION/PLACE as efficiently as the benchmark. It is also observed that the benchmark tool is sometimes not able to tag as comprehensively as the proposed approaches. This paper has successfully demonstrated the potential of using a semantically enriched Nested Event Model as an alternative for NER technique. SRL-ACE-NER has achieved an average precision of 92% in recognising PERSON, LOCATION/PLACE, TIME, and ORGANIZATION.