ESA-T2N: A Novel Approach to Network-Text Analysis

摘要

Network-Text Analysis (NTA) is a technique for extracting networks of concepts appearing in natural language texts that are linked by a certain measure of proximity. In prior works it has been argued that those networks are a representation of the mental model of the author. Extracting those networks often requires a high amount of domain knowledge of the analyst to specify relevant concepts in advance. Grammatical approaches that discover concepts automatically. However, the resulting networks can contain noisy concept nodes and meaningless edges, and thus, are less interpretable. In this paper, we present a new method that bridges between both approaches for extracting networks from text using Wikipedia as a knowledge base to map phrases occurring in the text to meaningful concepts. The utility of the method is demonstrated along a case study where pivotal moments in the evolution of Brexit debates in the British House of Commons in 2019 are discovered in speech transcripts.