Domain-specific textual commonsense concept acquisition using a corpus
Rushdi Shams, M.S.A. Shahnawaz Chowdhury, S.M. Abu Saleh Shawon · 2011
In this paper, we present a textual commonsense concept acquisition system named SenCept. It works on text of DC electrical circuits and provides commonsense concepts associated with them for better contextualization. SenCept uses a manually developed commonsense knowledge-base that is built upon linguistic information of a domain-specific corpus. We selected representative commonsense knowledge by using several parameters like knowledge weight, average commonsensical distances among knowledge, and normalized mean. To identify commonsense concepts for any sentence, SenCept concentrates on mean of distances between normalized weights of representative sentences and average commonsensical distances among knowledge. We fed 100 sentences to five human subjects and SenCept to evaluate its performance. Results showed that concepts produced by SenCept are originated from textual commonsense in contrast to human analysis that produces concepts from domain knowledge. Moreover, SenCept's Common Concept Rate (CCR) is 43 percent- which is better than that of human analysis.