چكيده به لاتين
The attraction of easy international communication with low cost in social networks has led to an increasing number of active members in them, and they expand the level of international communication and interaction by sharing information, such as personal information, opinions on a specific topic, and liking and discussing a topic. This has made the analysis of large data in social networks an irreplaceable opportunity for companies, organizations and governments to achieve their goals such as discovering the target market, satisfaction through virtual space and social psychology in relation to a specific issue. While in this space, due to the wide usage, the high number of software related to this space, and the ever-increasing raise of this data, processing this data, which has different formats, is challenging enough on one hand the challenge of changes in the structure of this space should be managed separately. According to these explanations, the aim of this study is to create a system to predict the capital market, which can predict the future stock price based on the previous stock price along with sentiment analysis in virtual space using machine learning models.In this thesis, the algorithms used in this way have been introduced in detail, we have shown its advantages and disadvantages, and we have designed the best core for a recommender system and using the appropriate algorithm in natural language processing. The proposed system in this research uses the prices of different symbols in the previous days and the opinions of users on social media related to this index to predict the future stock price or its increase or decrease. The result obtained with a simple method without using user comments is compared and the efficiency of processing comments is shown in increasing the prediction accuracy. In order to implement and test the proposed solution, we created a dataset consisting of 247,822 records of daily comments with three hash tags: Shasta, Khodro, and Burs, and for its validation, the dataset received from the Iranian Stock Exchange Organization is also used, which indicates An increase or decrease in the value of a symbol per day. Finally, the result of the work is between five and six percent of the average error (94% of accuracy). Also, in order to calculate the impact of sentiment analysis on capital market value forecasting correctly, we reviewed the entire process once without considering the sentiment score.
