Automatic Classification of Citizens' Appeals

This article presents the results of work on a service that allows classifying citizens' appeals. Citizens' appeals are unstructured text written in natural language. The objective is to determine by text to which executive body the application should be assigned. Based on text set documents, a distance matrix is constructed using word2vec and Word Mover's Distance algorithms. For each new document, the distance vector from that document to each document in the training set is calculated. Applies a K-neighbor classifier with a neighbor count parameter of 10 allows you to define the subject and performer of the case.