O‘zbek tilidagi sharhlar uchun konturli konvolyutsion neyron tarmoqiga asoslangan aspekt vektorlashtirishning uch bosqichli sentiment tahlili modeli

In this paper, we develop a three-stage aspect-based sentiment analysis (ABSA) model for user
reviews written in Uzbek. In the first stage, aspect terms are extracted from the sentence (ATE); in the second stage, each aspect is assigned an appropriate semantic category (ACC); in the third stage,
the sentiment polarity for every aspect is predicted (ATSA). The three stages are jointly formalized
within a single multi-task neural architecture based on gated convolutional aspect embedding
(GCAE). At the input level, the model combines FST-based morphological analysis, FastText
word embeddings, and Uzbek BERT contextual representations, while the gated convolutional block
performs aspect-dependent feature selection. The proposed approach makes it possible to integrate
lexical, morphological, and contextual information for an agglutinative language like Uzbek, and
provides a solid theoretical basis for practical applications such as online service ranking and
automatic analysis of customer feedback.