Identification and modeling of discoverers in online social systems

TL;DR

This paper introduces a framework to identify 'discoverers' in online social systems who find popular items early, revealing their unique role and improving prediction and ranking algorithms by considering temporal discovery patterns.

Contribution

The paper presents a novel analytical framework and network model to identify discoverers, demonstrating their significance across various social systems and challenging existing ranking methods.

Findings

Discoverers are present in many online systems.

Discoverers can predict future popular items.

Classical ranking algorithms fail to account for discovery timing.

Abstract

The dynamics of individuals is of essential importance for understanding the evolution of social systems. Most existing models assume that individuals in diverse systems, ranging from social networks to e-commerce, all tend to what is already popular. We develop an analytical time-aware framework which shows that when individuals make choices -- which item to buy, for example -- in online social systems, a small fraction of them is consistently successful in discovering popular items long before they actually become popular. We argue that these users, whom we refer to as discoverers, are fundamentally different from the previously known opinion leaders, influentials, and innovators. We use the proposed framework to demonstrate that discoverers are present in a wide range of systems. Once identified, they can be used to predict the future success of items. We propose a network model which reproduces the discovery patterns observed in the real data. Furthermore, data produced by the model pose a fundamental challenge to classical ranking algorithms which neglect the time of link creation and thus fail to discriminate between discoverers and ordinary users in the data. Our results open the door to qualitative and quantitative study of fine temporal patterns in social systems and have far-reaching implications for network modeling and algorithm design.