Scalable Visualisation of Sentiment and Stance

Scalable Visualisation of Sentiment and Stance Jon Chamberlain1, Udo Kruschwitz1, Orland Hoeber2 1University of Essex, Wivenhoe Park, Colchester CO4 3SQ, UK 2University of Regina, 3737 Wascana Pkwy, Regina, SK S4S 0A2, Canada {jchamb,udo}@essex.ac.uk, orland.hoeber@uregina.ca Abstract Natural language processing systems have the ability to analyse not only the sentiment of human language, but also the stance of the speaker. Representing this information visually from unevenly distributed and potentially sparse datasets is challenging, in particular when trying to facilitate exploration and knowledge discovery. We present work on a novel visualisation approach for scalable visualisation of sentiment and stance and provide a language resource of e-government public engagement of 9,278 user comments with stance explicitly declared by the author. Keywords: sentiment, stance, information visualisation, opinion mining, e-government, language resource 1. Introduction The ever growing amount of user generated content online presents a variety of challenges, one of them being how to summarise large collections of user comments in online discussion forums and social networks, capturing the gist of what has been said and reflecting the different view points on a range of topics. Natural language processing systems have the ability to analyse not only the sentiment of this communication, but also the stance of the speaker. Online participation is increasing at an exponential rate and the desire for consumption of such data has driven inno- vations in communication platforms, in particular on so- cial media platforms such as Facebook1 and Twitter2; how- ever, similar scenarios can exist at much smaller, local scales such as comments on online newspaper reports and e-government consultation schemes. Representing information visually from these unevenly dis- tributed and potentially sparse datasets is problematic, in particular when trying to facilitate exploration and knowl- edge discovery. Traditional linear scaling will result in large values dominating smaller ones; logarithmic scaling addresses this problem, but is not intuitive to casual users. This paper addresses these challenges by discussing related work in the area of sentiment and stance analysis, as well as information visualisation relevant to this field (Section 2.). We then propose a novel approach to scalable visual- isation of sentiment and stance (Section 3.). In Section 4. we outline a case study of how this approach can be used for visualising user participation on an e-government portal, followed by further details of how the language resource was acquired and how it will be shared with the commu- nity (Section 5.). In Section 6. we compare our approach to other visualisation techniques and, in conclusion, we dis- cuss the implications of this approach and its relevance to the NLP researchers (Section 7.). 2. Related Work Summarising and visualising online discussions has be- come ever more important for users to understand the com- 1https://www.facebook.com 2https://twitter.com plexities of multi-faceted issues in modern society. There are many approaches for visualising sentiment or stance as time series data, including timelines (Kucher et al., 2014), dense pixel displays (Hao et al., 2011), summaries of on- line conversations (Riccardi et al., 2016) and summaries of Twitter data (Mohammad et …

Content truncated to 500 words