Dialect-Sensitive Sentiment Analysis for Uzbek News Content Using Traditional Methods

This paper discusses a combinational algorithm for sentiment analysis of Uzbek texts based exclusively on traditional (rule-oriented) methods. The authors introduce a script conversion procedure: Cyrillic characters are replaced with Latin, which eliminates the ambiguity of further text analysis. The next stage is dialect normalization, where specific regional forms of words are reduced to a formal equivalent to reduce the number of missed emotionally charged lexemes. The process is completed by lexical analysis of sentiment based on a dictionary of words, each of which is assigned numerical scores from -2 to +2, taking into account “amplifying” adverbs (for example, “very”,”incredibly”). The proposed approach was tested on two different samples, where the first includes more formal news, and the second - user comments, distinguished by stylistic and dialectal variability. The results indicate that the algorithms show fairly high values of accuracy and completeness, indicating the practical value of the solution. Particular emphasis is placed on the fact that when working with the Uzbek language it will be useful to take into account dialectism and mixed writing, otherwise a significant portion of emotionally charged words risk being unrecognized.