Доповнення Даних Контентом На Іноземних Мовах В Текстовій Класи- Фікації З Використанням Машинного Навчання

Abstract

This paper addresses the problem of insufficient data for text classification tasks and methods of datasets reinforcement. We explore the influence of data augmentation on classification accuracy. Text dataset is complemented by machine translations of similar data from languages of other families and the advantages of this approach are demonstrated. An experimental research is conducted for the sentiment analysis problem for hotel reviews data that shows the difference in accuracy of the model trained on original and translated data. Influence of dataset volume on resulting accuracy is studied and compared for both types of the datasets. It is shown that learning of the model graphs are similar and differs only with some minor static error.