BUILDING TOOLS FOR EVALUATING USER OPINION ON YOUTUBE CHANNEL

Abstract

In this thesis, I will propose an analytical tool that evaluates users' opinions through interaction with Youtube. The system includes two main functions: automatic interaction and automatic evaluation of user opinions. The auto-interaction section uses the Selenium framework to get the specific positions of the function buttons on the Youtube website, through which the system can conduct interactions with the channel automatically. Similarly, the user opinion rating section also uses the Selenium framework to get the location of comments and execute some javascript commands. From the comments retrieved, the system will use the specific machine learning models Underthesea and VaderSentiment to analyze and categorize these comments. The system uses a small number of pre-made Google accounts to perform interactive tasks with YouTube channel with randomly selected channel data. Particularly for the rating of users using Youtube channels with random videos and all comments are also collected in order of each channel on average over 200 comments for a video. The experimental results were relatively stable with auto-interaction, functions such as auto-register, auto-like, and auto-watch videos that worked well, and no errors with average interaction time for each. videos in about 4 to 5 minutes (including watching advertisements). The machine learning model used for opinion evaluation works well, giving high accuracy results with 70% for the Underthesea library model for Vietnamese language analysis and 80% for the model using VaderSentiment English language analysis.