Detection of Stance and Sentiment Modifiers in Political Blogs

Abstract

Abstract. The automatic detection of seven types of modifiers was studied: Certainty, Uncertainty, Hypotheticality, Prediction, Recommendation, Concession/Contrast and Source. A classifier aimed at detecting local cue words that signal the categories was the most successful method for five of the categories. For Prediction and Hypotheticality, however, better results were obtained with a classifier trained on tokens and bigrams present in the entire sentence. Unsupervised cluster features were shown useful for the categories Source and Uncertainty, when a subset of the training data available was used. However, when all of the 2,095 sentences that had been actively selected and manually annotated were used as training data, the cluster features had a very limited effect. Some of the classification errors made by the models would be possible to avoid by extending the training data set, while other features and feature representations, as well as the incorporation of pragmatic knowledge, would be required for other error types.

StanceVis Prime: visual analysis of sentiment and stance in social media texts

“…The existing computational approaches usually focus on AGREEMENT or DISAGREEMENT on a certain topic (Chen and Ku 2016; Mohammad et al 2016Mohammad et al , 2017, and only a few works take a wider view of stance aspects/categories into account, such as NECESSITY and VOLITION (Simaki et al 2017). In this work, we follow the approach taken in our interdisciplinary project, where the researchers in linguistics defined the stance categories of interest (see Table 1) and the experts in computational linguistics implemented a custom stance classifier (Skeppstedt et al 2016b(Skeppstedt et al , 2017. Classification is carried out at the utterance level in a multi-label fashion, i.e., one utterance can be labeled with multiple stance categories simultaneously.…”

“…One of the tasks of our work is to represent the change of public sentiment and stance over time (Aigner et al 2011). ThemeRiver (Havre et al 2000) is a classic example of temporal text data visualization, which uses a stacked area graph metaphor (Byron and Wattenberg 2008) to represent a flow of themes/topics #occurrences of stance in utterances detected with a custom stance classifier (Skeppstedt et al 2016b(Skeppstedt et al , 2017 detected in text data. In a similar fashion, Dörk et al (2010) use a stacked graph to represent topics in the streaming Twitter data, and TextFlow (Cui et al 2011) combines a stacked graph with glyphs and line plots to represent topic evolution.…”

“…Our implementation currently uses the VADER sentiment classifier (Hutto and Gilbert 2014) and a custom logistic regression-based (Hosmer et al 2013) stance classifier (Skeppstedt et al 2016b(Skeppstedt et al , 2017 developed with scikit-learn (Pedregosa et al 2011) for 12 stance categories listed in Table 1. The stance classifier was trained on the text data that was (1) collected from blogs and forums, (2) annotated at the utterance/sentence level as part of the StaViCTA project (Simaki et al 2017), and (3) later augmented with token-level annotations by our collaborators (Skeppstedt et al 2016a) in order to improve the classification performance compared to the initial utterance-level classification.…”

StackGenVis: Alignment of Data, Algorithms, and Models for Stacking Ensemble Learning Using Performance Metrics

“…In this section, we describe how StackGenVis can be used to improve the results of sentiment/stance detection in texts from social media, when compared to previous work from Skeppstedt et al [55]. The authors studied the automatic detection of seven stance categories: certainty, uncertainty, hypotheticality, prediction, recommendation, concession/contrast, and source.…”

“…The data set is very imbalanced, with most cases being on the absence side. Skeppstedt et al [55] used an SVM algorithm to train and build their baseline classifier, and we are going to compare it to our stacking ensemble method in this use case.…”