A hybrid peer-to-peer recommendation system architecture based on locality-sensitive hashing

Abstract

Recommendation systems have become ubiquitous recently as they help to mitigate information overflow of the nowadays life. The vast majority of current recommendation system approaches are centralized. Although centralized recommendations have several significant advantages, they also have two main drawbacks: single point of failure and the necessity for users to share their preferences. In this paper, a system architecture of a peer-topeer recommendation system with limited preferences disclosure is proposed. The proposed architecture is based on a locality-sensitive hashing of user preferences and an anonymized distributed hash table approach to peer-to-peer design.

Ditto: Towards Decentralised Similarity Search for Web3 Services

“…Some works have proposed to use a DHT to store the mapping of similarity signatures to content [40], [41]. Haghani et al [42] leverage a cyclic DHT space based on Chord [23] to provide nearest neighbour search and queries within a range.…”

Stochastically Robust Personalized Ranking for LSH Recommendation Retrieval

“…In face of these challenges, recent approaches turn to indexing schemes to overcome the prohibitive cost of performing exhaustive top-k recommendation search for each user. In particular, one of the most popular such schemes is Locality-Sensitive Hashing (LSH) (Shrivastava and Li 2015;Bachrach et al 2014;Fraccaro, Paquet, and Winther 2016;Le and Lauw 2017;Hsieh et al 2017;Liu and Wu 2016;Koenigstein and Koren 2013;Qi et al 2017;Smirnov and Ponomarev 2014). In the prevalent binary variant, LSH approximates relative distances between data points, by computing the Hamming distance between the corresponding codes, and hashing similar data points to similar codes with high probability.…”