Institute for Web Science and Technologies · Universität Koblenz - Landau
Institute WeST

Emotional Intensity Recognition Using Unsupervised and Supervised Algorithms for Evaluating Democratic Satisfaction in Microblog Communities

[go to master theses]

Open Master Thesis - Contact a supervisor for more details!

Democratic satisfaction is an important indicator in political science in order to gauge the health of democratic institutions in a country [1]. Measurement usually depends on surveys, where each respondent states satisfaction with democracy. However, especially in younger and challenged democracies, the answer tends to mean whether the respondent is satisfied with the current regime, which differs by ideological orientation.

Theoretical framework
This thesis proposes a measurement of sources of dissatisfaction that are not stated in surveys. By measuring sentiment for keywords related to the objects of political support, we can compare between possible source of dissatisfaction. You may use the following structure [2] to measure whether democratic dissatisfaction stems mostly from:

  • The political community?
  • Regime principles?
  • Regime performance?
  • Regime institutions?
  • Political actors?

Therefore, the hypothesis can be, for example: Dissatisfaction with institutions will be higher than with actors, because citizens in young democracy with recent authoritarian past will trust leaders who are not eager to reform bad institutions. We define democratic (dis-)satisfaction as sentiments expressed for democratic keywords: The student merely needs to define a small number of keywords for each of the five points above, in consideration of the country case(s) that the student may choose (for example, by using [3]).

 Methodology
 Firstly, the student will use semi-automated methods to gather large scale training data from microblogs with different ways of expressing emotions. Secondly, the student survey public emotional lexicons. These lexicons includes sentiment words and phrases, with assigned emotional orientations. The technical aim of this thesis is to improve emotion detection in microblogs by building an adaptive lexicon of words with an assigned emotional intensity in the train phase, and use this lexicon for classification in the test dataset. In this lexicon, a score in the range of [min, max] will be assigned for basic emotions. The unsupervised and supervised algorithms along with sentiment analysis approaches will be combined to generate the lexicon. By assuming that in a large annotated corpus , the emotion of a word can be inferred by investigating the frequency of occurrence of word in corpus. For example, if the frequency of the word rain in the “sadness” class of dataset is more than a threshold, the emotion of word can be assumed as “sad”. Thereby, the student will extract sentimental words from train dataset. Then emotion classification will be formulated as an optimization problem to find optimum emotional-intensity lexicon that minimizes the classification error. The main purpose is to use the lexicon with assigned emotional intensity in order to get a better understanding of the specific language of democratic satisfaction of social media users. Finally, the performance of approach should be evaluated and metrics such as accuracy, precision, recall and F-measure should be reported, and also student should interpret measurements in comparison to common democracy satisfaction indices.

Literature
 [1] Guide, A. "Measuring Public Support for Democracy." IDEA (2017).
[2] Linde, Jonas, and Joakim Ekman. "Satisfaction with democracy: A note on a frequently used indicator in comparative politics." European journal of political research 42.3 (2003): 391-408.
[3] https://datacatalog.worldbank.org/dataset/wps2283-database-political-institutions
Chatterjee, Ankush, et al. "Understanding Emotions in Text Using Deep Learning and Big Data." Computers in Human Behavior 93 (2019): 309-317.
 Hosseini AS. Sentence-level emotion mining based on combination of adaptive Meta-level features and sentence syntactic features. Engineering Applications of Artificial Intelligence. 2017 Oct 1;65:361-74.

Supervisors

  • staab@uni-koblenz.de
  • Professor
  • B 108
  • +49 261 287-2761
  • sadathosseini@uni-koblenz.de
  • Scientific Employee
  • B 007
  • +49 261 287-2966
  • han@uni-koblenz.de
  • Scientific Employee
  • B 006
  • +49 261 287-2864