Sentiment analysis of political communication: combining a dictionary approach with crowdcoding - PubMed (original) (raw)

Sentiment analysis of political communication: combining a dictionary approach with crowdcoding

Martin Haselmayer et al. Qual Quant. 2017.

Abstract

Sentiment is important in studies of news values, public opinion, negative campaigning or political polarization and an explosive expansion of digital textual data and fast progress in automated text analysis provide vast opportunities for innovative social science research. Unfortunately, tools currently available for automated sentiment analysis are mostly restricted to English texts and require considerable contextual adaption to produce valid results. We present a procedure for collecting fine-grained sentiment scores through crowdcoding to build a negative sentiment dictionary in a language and for a domain of choice. The dictionary enables the analysis of large text corpora that resource-intensive hand-coding struggles to cope with. We calculate the tonality of sentences from dictionary words and we validate these estimates with results from manual coding. The results show that the crowdbased dictionary provides efficient and valid measurement of sentiment. Empirical examples illustrate its use by analyzing the tonality of party statements and media reports.

Keywords: Crowdcoding; Media negativity; Negative campaigning; Political communication; Sentiment analysis.

PubMed Disclaimer

Figures

Fig. 1

Fig. 1

Creating a sentiment dictionary_. Notes_ i…number of sentences, j…number of coders, k…number of dictionary words, l…number of tonality ratings, n…number of sentences containing a rated word

Fig. 2

Fig. 2

Histogram of tonality scores of dictionary words (n = 5001)

Fig. 3

Fig. 3

Comparing expert scores and crowdscores. Note: Line indicates linear regression of crowdscores on expert scores

Fig. 4

Fig. 4

Comparing expert scores, crowdscores and automated, dictionary-based scores. Note: Lines indicate linear regression of dictionary-based scores on expert scores (grey line) and crowdscores (black line)

Fig. 5

Fig. 5

OLS regression coefficients (with 95 %-confidence intervals)

Fig. 6

Fig. 6

Mean tonality of campaign coverage on parties and others

Fig. 7

Fig. 7

CrowdFlower Markup Language for the coding of sentence tonality

Fig. 8

Fig. 8

Screenshot of the CrowdFlower coding interface

Similar articles

Cited by

References

    1. Alonso O, Baeza-Yates R. Design and implementation of relevance assessments Using Crowdsourcing. In: Clough P, Foley C, Gurrin C, Jones G, Kraaj W, Lee H, Mudoch V, editors. Advances in Information Retrieval. Berlin: Springer; 2011. pp. 153–164.
    1. Baumeister RA, Bratlavsky E, Finkenauer C. Bad is stronger than good. Rev. Gen. Psychol. 2001;5(4):323–370. doi: 10.1037/1089-2680.5.4.323. - DOI
    1. Baumgartner FR, Jones BD. Agendas and instability in American politics. American politics and political economy series. Chicago: University of Chicago Press; 1993.
    1. Benikova, D., Biemann, C., Reznicek, M.: NoSta-D Named Entity Annotation for German: Guidelines and Dataset. Proceedings of Language Resources and Evaluation Conference (LREC) 2014, 26–31 May, Reykjavik, Iceland (2014)
    1. Benoit K, Conway D, Lauderdale B, Laver M, Mikhaylov S. Crowd-sourced text analysis: Reproducible and agile production of political data. Am. Polit. Sci. Rev. 2016;110(2):278–295. doi: 10.1017/S0003055416000058. - DOI

LinkOut - more resources