نوع مقاله : مقاله پژوهشی
عنوان مقاله English
نویسندگان English
In recent years, Telegram groups and channels become considerable parameter in user’s opinion. Previous researches show the importance of sentiments in forecasting models and researches find significant relation between sentimental analysis and stock price prediction. They consider buy, sell and hold contents as independent variable. But in reality, there is a difference between people who publish their opinion in social media. Means people consider the speaker’s background and also for the viral messages there are same situation. In this research we consider importance of the publisher and message in social media. Then we separate content into different variables. We analyse 15.736.204 Telegram messages and according to our sentimental dictionary, we tag them then put merge them with 19.312 records related to symbols in different days. Selected symbols are the stocks in 30 company index portfolio and the duration of survey is from march 2019 to august 2021. Six variable selection approach was considered in this survey, for each symbol (among 36 selected companies) we train different artificial neural network which the number of neurons in hidden layer varies from 6 to 10 and the lag of data for input variable differs from 1 to 5. In order to remove initial values effect in neural network, each network runs 10 times. Results showed binary prediction in a model which separates messages by importance of people and amount of view has better performance than other models and RMSE error in that model for return forecasting significantly lower than other models.
کلیدواژهها English
