Filtering forbidden content

The relevance of this study lies in the need to filter content with high accuracy due to the creation of optimal variations of neural network architectures. The solutions available today provide low filtering accuracy, which leads to the blocking of potentially safe content. A complete lack of filtering will lead to the fact that minor users and other vulnerable groups will gain access to it, which is unacceptable. In today’s world, the process of digitization and integration of more and more software products into the life of an ordinary member of society is gaining momentum. Every day there are new apps that help people generate terabytes of content like TikTok, Instagram, Telegram, Linked In, etc. However, with the increase in the amount of content, the number of materials that violate the rules of the platform or the laws of the country of which the user is a citizen also increases. Every day there are photos and videos on social networks that contain smoking, alcohol, weapons, drug propaganda, direct or indirect, etc. At this moment, the phenomenon of content filtering comes to the rescue of users. The object of research is content filtering processes. The subject of research is the methods and technologies of building neural networks for content filtering. The purpose of the work is to increase the accuracy of content filtering by developing a neural network architecture. PyCharm and the Python programming language were chosen as a system development tool. The SciPy and Keras libraries, the Pickle data import and export library, and the Sckit-learn library for working with neural network algorithms were chosen as development tools. The result of the work is the constructed neural network architecture for filtering prohibited content. An analytical review of content filtering methods was conducted. Each of the methods has its own advantages and disadvantages. During the analysis, it was determined that the most effective option at the moment is the use of neural networks that learn from examples of prohibited content for its further filtering An Internet filter is software that limits or controls the content that an Internet user can access, especially when it is used to restrict material delivered over the Internet via the Internet, email, or other means. Content control software determines what content is allowed or blocked. No single solution provides complete coverage, so most companies deploy a mix of technologies to achieve adequate content control according to their policies. Such restrictions can be applied at different levels: a government can try to apply them nationwide, or they can, for example, be applied by an Internet service provider to its customers, an employer to its staff, a school to its students, a library to visitors, a parent to a computer child or individual users to their own computers. The purpose of the article is to analyze the relevance of using neural networks for filtering prohibited content and the phenomenon of content filtering in general. Based on all of the above, we can conclude that it is highly relevant to create various tools for automatic content filtering with an adequate behavior model.