Multimedia Data Analysis of Social Networks

Social networks are very popular today and contain a lot of different content, what increases the level of their impact on society. Unfortunately, the content directly contains information materials that violate Russian laws. Manual detection of content with violations of the laws is ineffective. Among technical means for solution of this problem spam filters could be noted, but their realization is not able to fully cope with this problem. Therefore, the problem of multimedia data analysis in social networks on detection of laws' violations is actual and requires new approaches to solve it. This dissertation is devoted to the multimedia data analysis of social networks data on detection of laws' violations of the Russian Federation. The objectives of this work are the development of requirements, analysis of existing methods and tools for creating automatized software tools for solution of this problem, the realization of this system and practical search of the effectiveness of the applied means. For solution the problem of detecting violations in social networks were considered methods such as keyword search, n-gram search and machine learning algorithms. Among the existing social networks for realization of programming tools were chosen VK.com and YouTube. Programs were written in Python v. 3.5 on Ubuntu 16.04 LTS platform. In developing such libraries were used as pymorhy2, sklearn and pymongo. For the analysis of video's audio track was used Yandex SpeechKit Cloud technology, which allows to convert audio to text that makes possible to use the same methods for its analysis, as for text. It was also carried out a practical research of the effectiveness of the methods used for this task. In the results of the work, the objectives were completed in full. It was found that all methods, which were used, and developed software tools are effective for solving the problem.