Intelligent system for user groups socialization with similar interests

Abstract

The article develops a general architectural system of socialization of groups of users with similar interests and functional requirements for it. To process a large part of the information, the system is implemented using the methods of fuzzy text information search and machine learning. thus, N-gram, selection expansion and structured Noisy Channel models are applied. A feature of the implementation is the processing of the text, the analysis of words in the text and the formation of evaluations. A convolutional neural network implementation is designed to determine user authenticity based on facial photo analysis. implementation of fuzzy search algorithms – for processing text data of various volumes to analyze information about each user, form a certain user rating, compare this user with other users to promote further socialization of users whose interests coincide the most. When experimentally checking the accuracy of the developed system by determining the percentage of similarity of current users with the help of N-grams and their connections. Running these algorithms simultaneously is about 15% more accurate than the N-gram algorithm and about 10 % more efficient and accurate than the others algorithm. The operation of the algorithm for linear search of tags in the dictionary and the operation of the Noisy Channel algorithm using the BK-tree are also analyzed. Thanks to which it was possible to achieve significant advantages in the work algorithm, instead of a linear view of the search time, a logarithmic dependence was obtained. A system of synchronous and asynchronous methods also works. At first, the difference is not visible, but the more requests, the faster the system loads and tries to respond to them more by displaying from asynchronous methods.