Automated sentiment analysis of short texts

Digital technologies are transforming traditional patterns of user behavior, increasingly shifting communication toward mobile devices that serve as personal assistants and multifunctional tools. This transition highlights the growing need to assess the emotional attitude of transmitted messages. Mobile communication imposes constraints on message length and style, emphasizing brevity and reducing contextual depth. For sentiment analysis of short sets of text and extraction of emotional characteristics, this study proposes the use of binary classification as a preprocessing stage for data arrays, combined with a floating temporal context window to refine the processed information. Recurrent neural networks are employed alongside the binary classifier to enhance analytical accuracy while maintaining computational efficiency. It is demonstrated that the results of this work can be improved by supplementing traditionally used datasets with information collected directly from users' mobile devices during their daily activities. The aim of this work is to improve the quality of sentiment analysis of short sets of user texts by developing and testing a method for automated generation of a trusted dataset. Existing datasets contain a significant amount of incorrectly labeled information, which impacts the final quality of the analysis. The proposed methods achieved correct answer rates of 96% on the training dataset and 92% on the validation dataset.