Metagraph Knowledge Base and Natural Language Processing Pipeline for Event Extraction and Time Concept Analysis

Abstract

Analysis of sentence meaning is necessary to improve the precision and recall of information retrieval. Event extraction is one of methods used for this purpose. An important step in solving this problem is the time extraction and processing. It is necessary to classify time intervals and determine the relation of succession for them. Intervals can be expressed with specific dates and times, pronouns and tense nouns or they are implied with successive events description. The authors propose to use a metagraph knowledge base and a natural language processing pipeline. The metagraph has the emergence property for a detailed description of a specific concept using a graph fragment. This representation of knowledge allows setting different types of relations between time concepts and states of objects. The study was carried out to extract time intervals from OpenCorpora dataset. The number of time concepts in knowledge base was calculated according to their types.