청와대 청원 글의 어조 분석과 주제어 시각화 - 데이터과학적 분석 과정을 활용하여 -

초록

This research aims to investigate the text of Blue House petitions in order to identify different tones in relation to domain fields and make a visual representation of topic words embedded in these texts. The data covers the beginning of the Blue House petition web service up to the recent petitions, as many web-scraped as possible yet limited to those not unopened by the administrator, from August of 2017 to mid-January of 2021. We saved the petition text data structurally with their respective subject domain, date, title, body text, and agreement number totals so as to investigate the contained information conditionally. Firstly, we investigated sentence endings from petition texts according to subject domains and agreement number and found that linguistic tone, which is mainly expressed in the verb and its ending, is critical in agreement number totals. Secondly, we separated two data sets, one with all of the petitions and another with influential petitions—those which were agreed to by over 200,00 people so that they must be answered publicly—and obtained the distributional semantics of these texts using word embedding. Finally, we found that these texts contain different characteristics of tones with respect to agreement numbers and visualised subject topics using word embedding as distributional semantics from all of the petitions and influential petitions which have higher number of agreements respectively.

키워드