Of Wines and Reviews: Measuring and Modeling the Vivino Wine Social Network

TL;DR

This paper analyzes the Vivino social network to understand user perceptions of wines, examines biases in ratings, and develops models to predict wine ratings and review preferences based on wine features and review language.

Contribution

It introduces models that predict wine ratings and review preferences, and provides insights into rating biases and user perceptions in a large social wine network.

Findings

Ratings are not biased by wine cost.

Models can predict wine ratings based on characteristics and review language.

User review preferences can be accurately classified.

Abstract

This paper presents an analysis of social experiences around wine consumption through the lens of Vivino, a social network for wine enthusiasts with over 26 million users worldwide. We compare users' perceptions of various wine types and regional styles across both New and Old World wines, examining them across price ranges, vintages, regions, varietals, and blends. Among other things, we find that ratings provided by Vivino users are not biased by cost. We then study how wine characteristics, language in wine reviews, and the distribution of wine ratings can be combined to develop prediction models. More specifically, we model user behavior to develop a regression model for predicting wine ratings, and a classifier for determining user review preferences.