Description
Aspect based sentiment analysis (ABSA) deals with the fine grained analysis of text to extract entities and aspects and analyze sentiments expressed towards them. Previous work in this area has mostly focused on data of short reviews for products, restaurants and services. We explore ABSA for human entities in the context of large documents like news articles. We create the first-of-its-kind corpus containing multiple entities and aspects from US news articles consisting of approximately 1000 annotated sentences in 300 articles. We develop a novel algorithm to mine entity-aspect pairs from large documents and perform sentiment analysis on them. We demonstrate the application of our algorithm to social and political factors by analyzing the campaign for US presidential elections of 2016. We analyze the frequency and intensity of newspaper coverage in a cross-sectional data from various newspapers and find interesting evidence of catering to a partisan audience and consumer preferences by focusing on selective aspects of presidential candidates in different demographics.
Recommended Citation
Ahmad, Raza; Pervaiz, Ahsan; Mannan, Humdah; and Zaffar, Fareed, "Aspect Based Sentiment Analysis for Large Documents with Applications to US Presidential Elections 2016" (2017). AMCIS 2017 Proceedings. 13.
https://aisel.aisnet.org/amcis2017/SocialTechincal/Presentations/13
Aspect Based Sentiment Analysis for Large Documents with Applications to US Presidential Elections 2016
Aspect based sentiment analysis (ABSA) deals with the fine grained analysis of text to extract entities and aspects and analyze sentiments expressed towards them. Previous work in this area has mostly focused on data of short reviews for products, restaurants and services. We explore ABSA for human entities in the context of large documents like news articles. We create the first-of-its-kind corpus containing multiple entities and aspects from US news articles consisting of approximately 1000 annotated sentences in 300 articles. We develop a novel algorithm to mine entity-aspect pairs from large documents and perform sentiment analysis on them. We demonstrate the application of our algorithm to social and political factors by analyzing the campaign for US presidential elections of 2016. We analyze the frequency and intensity of newspaper coverage in a cross-sectional data from various newspapers and find interesting evidence of catering to a partisan audience and consumer preferences by focusing on selective aspects of presidential candidates in different demographics.