Strategies for Filtering Unwanted Comments in Social Media
Andrii Podorozhniak, Vasyl Oliinyk, Nataliia Liubchenko · 2024
In today's digital age, social media platforms have become an indispensable part of everyday life, revolutionizing communication and information exchange. However, with the freedom of expression afforded by these platforms comes the challenge of managing unwanted messages, ranging from intrusive advertisements to spam and propaganda. This issue has gained increasing relevance as undesirable comments can significantly distort public opinion and contribute to the spread of misinformation. To combat this problem, a comprehensive algorithm for filtering unwanted messages in social media comments has been developed. Grounded in text analysis, this algorithm employs three popular methods: convolutional neural networks, support vector machines, and k-nearest neighbors. Notably, the algorithm automatically determines whether a message is unwanted, based on the collective output of these methods. Before final decisionmaking, comment text undergoes preprocessing using efficient text processing algorithms to reduce noise and enhance analysis quality. Text preprocessing may involve steps such as stop-word removal, lemmatization, and vectorization. The resulting text is then inputted into each algorithm, which examines it for characteristics of unwanted messages. The strength of this comprehensive approach lies in its consideration of diverse text characteristics, improving the accuracy of unwanted message detection. To validate the algorithm's effectiveness, tests were conducted in the Telegram messenger, demonstrating successful filtration of unwanted messages and contributing to enhanced communication quality and prevention of undesirable information dissemination across social media platforms. Overall, the development and deployment of comprehensive algorithm for filtering unwanted messages play a crucial role in combating propaganda and misinformation, fostering a safer and more informed online environment.