A MODEL OF RECOMMENDER SYSTEM FOR P2P COMPUTER NETWORKS

Abstract

In this article, a research of peer-to-peer network algorithms is conducted. A comparative analysis of data search methods in centralized, decentralized unstructured and decentralized structured peer-to-peer networks is carried out. It has been found that the existing methods of peer-to-peer networks do not use recommender systems to improve data search. A mathematical model of a recommender system for a decentralized P2P network has been developed, taking into account user preferences and the number of transitions to download data. A method of forming recommendations for peer-to-peer computer network users based on the proposed mathematical model has been developed. In P2P networks, there is a problem of indexing and searching files on different network devices. For various reasons, searched files may not be available to a user, even if they were previously added to a system and indexed. For example, computers containing the desired file or routing tables to it or its parts have left a network, or P2P network construction technologies are used with probabilistic search methods that do not always find files located far from the user's computer, etc. Recommender systems are used to build lists of recommendations for users based on their previous actions, including likes, ratings, views, downloads, etc. They make it possible to facilitate the search in a system with a large number of objects, supplementing the classic search, and in some situations even substituting the search. Recommender systems can also be used to rank classical search results. Thus, they can be combined with conventional search algorithms in various ways. In P2P networks, the use of recommender systems can have additional benefits. If a user is searching for a specific file that has been previously added to a network and the file is not found for various reasons, you can provide the user a list of recommendations based on his/her preferences and possibly search query. A model and method of forming recommendation lists in peer-to-peer networks proposed in the paper are designed for the general case and are not tied to a specific search query, can be applied in unstructured and structured decentralized P2P networks to familiarize a user with a content that he/she may like based on the prediction of his/her preferences. This can increase the overall interest of users in a content of a network.