Predicting Party Group from the Lithuanian Parliamentary Speeches

Abstract

A number of recent research works have used supervised machine learning approaches with a bag-ofwords to classify political texts -in particular, speeches and debates-by their ideological position, expressed with a party membership. However, our classification task is more complex due to the several reasons. First, we deal with the Lithuanian language which is highly inflective, has rich morphology, vocabulary, word derivation system, and relatively free-word-order in a sentence. Besides, we have more classes, as the Lithuanian Parliament consists of more party groups if compared to e.g. the European Parliament or the US Senate. Moreover, classes are not stable, because a considerable number of the Lithuanian parliamentarians migrate from one party group to another even within the same parliamentary term. In this research we experimentally investigated the influence of different pre-processing techniques and feature types on two datasets composed of the texts taken from two parliamentary terms. A classifier based on the bag-of-words and token bigrams interpolation gave the best results: i.e. it outperformed random and majority baselines by more than 0.13 points and achieved 0.54 and 0.49 accuracy on the 1 st and the 2 nd dataset, respectively. The error analysis revealed that the same confusion patterns stand for both datasets, besides, majority of these confusions can be explained on the basis of the ideological or pragmatic similarities between those party groups.

Sentiment and position-taking analysis of parliamentary debates: a systematic literature review

“…Within this work, there exist many inconsistencies in the use of terminology, with studies in some cases referring to very similar tasks by different names, while in others the same term may mean quite different things. For example, while Chen, Zhang, Wang, Yang & and Kapočiūtė-Dzikienė & Krupavičius (2014) both attempt to classify speakers according to party affiliation, the former refer to this as "political ideology detection", and the latter as "party group prediction". Conversely, a single term like "sentiment analysis" may be used to refer to, among other things, support/opposition detection (Thomas, Pang & Lee, 2006), a form of opinion-topic modeling (Nguyen, Boyd-Graber & Resnik, 2013), and psychological analysis (Honkela, Korhonen, Lagus & Saarinen, 2014).…”

“…Political and social scientists are responsible for less than half the number of included studies as computer scientists (n = 14), and just 12 studies involve multidisciplinary research. Of these, seven involve both computer scientists and political or social scientists (Kapočiūtė-Dzikienė & Krupavičius, 2014;Lapponi, Søyland, Velldal & Oepen, 2018;Rheault, 2016;Rheault, Beelen, Cochrane & Hirst, 2016;Rudkovsky et al, 2018;Sakamoto & Takikawa, 2017;Van der Zwaan, Marx & Kamps, 2016), three collaboration between linguists and computer scientists (Honkela et al, 2014;Iyyer, Enns, Boyd-Graber & Resnik, 2014;Nguyen et al, 2013), and two that include researchers from three different fields (Diermeier, Godbout, Yu & Kaufmann, 2012;Nguyen, Boyd-Graber, Resnik & Miler, 2015). According to the number of studies published on this subject annually, interest in this area has been increasing over time, particularly in recent years (see Figure 2.…”

“…Balahur et al (2009) combine polarity with party classification, a task that we consider to be a form of ideology detection, but which they name "source classification". Indeed, this is another task that suffers from a lack of clarity over terminology, with some studies considering party affiliation to be a proxy for ideology (Diermeier et al, 2012;Jensen et al, 2012;Kapočiūtė-Dzikienė & Krupavičius, 2014;Taddy, 2013), while others do not make this connection, extracting information about speakers' ideologies from their sentiment towards different topics (Bhatia & P, 2018;Chen et al, 2017;Nguyen et al, 2013), or training a model on examples that have been explicitly labelled by ideology, and not party membership (Iyyer et al, 2014). Yet others perform party classification, making no mention of the relationship between party membership and ideology (Balahur et al, 2009;Burfoot, 2008;Lapponi et al, 2018;Lefait & Kechadi, 2010).…”

“…• Textual: including word (all studies) or character (Kapočiūtė-Dzikienė & Krupavičius, 2014) n-grams, punctuation, embeddings, custom dictionary keyword features (Budhwar et al, 2018), words from particular grammatical categories (Iyyer et al, 2014;Kapočiūtė-Dzikienė & Krupavičius, 2014;Lapponi et al, 2018;Monroe et al, 2017;Naderi & Hirst, 2016;Onyimadu et al, 2013;Sokolova & Lapalme, 2008;Van der Zwaan et al, 2016), presence of questions (Budhwar et al, 2018), word embeddings (Bhatia & P, 2018;Glavaš et al, 2017;Iyyer et al, 2014;Ji & Smith, 2017;Li et al, 2017;Naderi & Hirst, 2016;Rheault, 2016;Rheault et al, 2016) and sentence (Rudkovsky et al, 2018) embeddings, and parse trees (Balahur et al, 2009;Iyyer et al, 2014;Ji & Smith, 2017).…”

Semantic Change in the Language of UK Parliamentary Debates

“…Knowledge of the level of such semantic change existing in a particular domain can assist in the design of systems for downstream natural language processing tasks such as sentiment analysis. For example, training and testing on in-domain data from different periods of time has been shown to negatively affect perfomance in named entity recognition (Fromreide et al, 2014) and sentiment analysis (Kapovciute-Dzikiene and Krupavicius, 2014). Successful analysis of such changes in Hansard could therefore be an important element in the development of civic technology applications for parliamentary analysis.…”