Research of the Feminist Movement on the Material of Social Networks

Modern methods of natural language processing allow to use data from social networks for large social research, which was previously impossible. This makes it possible to collect better and more complete statistics compared to the results of social surveys. Therefore, it improves the quality of research, as opinion polls are less representative, and people are less willing and honest to answer them, especially when it comes to sensitive topics. This paper estimates the attitude towards the side effects of combined oral contraceptives. The data collected from social networks better represents the real world, because it represents the opinions of women belonging to different social classes, age and ethnic groups, moreover, women speak out on this topic more confidently than in a real conversation, because the topic is still taboo in society and thus women can be condemned because for using such type of drugs. This paper describes the model which detects the mention of side effects from combined oral contraceptives in the text, as well as the further analysis of the results obtained. Research results – this is a world's first dataset about the side effects of combined oral contraceptives, marked up with the help of crowdsourcing, containing 2387 texts (of which 14% of the elements of the positive class), as well as a model that is trained on this dataset for the task of detecting the presence of side effects from combined oral contraceptives. The result is also statistics on mentions of various side effects, collected using a trained model. These statistics allow you to find out what side effects women most often face and how they feel about it. Key words: crowdsourcing, BERT, sentence BERT, COC