On application of data mining techniques to research of manifestos of political parties

The paper presents a new method for determining the proximity of political positions contained in manifestos, i.e. the election programs of political parties, as well as other documents published by the parties to attract voters, which is based on latent semantic analysis. The approach is based on the conjecture that the proximity of political positions reveals itself as syntagmatic proximity of texts of the manifestos. A detailed description of the algorithm is presented, which includes the preprocessing of text, breaking it into fragments, “fragment-word” matrix construction, its normalization, the use of singular value decomposition, and construction of proximity diagrams. Some conclusions obtained from this analysis are briefly outlined.