Eduskunnan täysistunnon puheenaiheet 1999–2014
Miten käsitellä LDA-aihemalleja?
Keywords:
LDA, aihemalli, eduskunta, täysistuntokeskustelut, korrelaatioanalyysiAbstract
Latent Dirilecht Allocation (LDA) is one of the most widely used topic models. Here we present and evaluate LDA models of the plenary sessions of the Finnish parliament Eduskunta and assess which amount of topics is the most sensible to an exploratory analysis of the content of the speeches in the Parliament. As a novel methodological application to quantifying the co-occurrence of different topics in the speeches, we analyze the topics generated by the LDA with compositional correlation coefficients which allows using the LDA in a similar fashion to correlated topic model and dynamic topic models, assuming the data is time stamped and the vocabulary remains unchanged. We present a case example by analyzing the correlation between a topic about municipal issues and state-municipalities relationship and topics about budget debate as well as about politics and democracy. There we can notice a large difference before and after 2011 elections: before the discussion was about money, and after, during the debates about a municipal reform bill, which was dropped, the speeches were on democracy.
