Text Mining of Related Events from Natural Science Literature

摘要

We present an approach to text mining in areas where the entities of interest can not be defined in advance. Our system is aimed at finding related events in natural science literature, in particular, changing/increasing/decreasing variables in Marine science publications. It enables semantic search for events by abstracting from morphological, lexical-semantic and syntactic variations. In addition, generalisation of variables through syntactic pruning helps finding similar variables. Relations between events are induced from cooccurrence frequencies. Extracted information is stored in a property graph database and accessed using the Cypher query language. A user interface presents events as a graph to visualise their type, frequency and relation strength, in combination with their textual sources.