[Paper Review] Exploring Linkablility of Community Reviewing

Mishari Al Mishari, Gene Tsudik|arXiv (Cornell University)|Nov 1, 2011

Spam and Phishing Detection20 references3 citations

TL;DR

This paper investigates the linkability of anonymous reviews on community review platforms using simple statistical models on a large Yelp dataset. It finds that up to 99% of reviews from active contributors can be linked with high confidence based on writing style alone, revealing significant privacy risks due to consistent authorial traits in reviews.

ABSTRACT

Large numbers of people all over the world read and contribute to various review sites. Many contributors are understandably concerned about privacy; specifically, about linkability of reviews (and accounts) across review sites. In this paper, we study linkability of community reviewing and try to answer the question: to what extent are "anonymous" reviews linkable, i.e., likely authored by the same contributor? Based on a very large set of reviews from a popular site (Yelp), we show that a high percentage of ostensibly anonymous reviews can be linked with very high confidence. This is despite the fact that we use very simple models and equally simple features set. Our study suggests that contributors reliably expose their identities in reviews. This has important implications for cross-referencing accounts between different review sites. Also, techniques used in our study could be adopted by review sites to give contributors feedback about privacy of their reviews.

Motivation & Objective

• To assess the extent to which seemingly anonymous reviews on community review sites can be linked to the same author.
• To investigate whether consistent writing styles across reviews enable high-confidence re-identification of contributors.
• To explore the feasibility of detecting spam or self-reviewing behavior through review linkability.
• To provide a privacy-aware feedback mechanism for review sites to inform users about the linkability of their reviews.

Proposed method

• Applied a Naïve Bayes classifier using only alphabetical letter frequency distributions as features to classify anonymous reviews into author groups.
• Used Kullback-Leibler divergence to measure similarity between review feature distributions for linkability assessment.
• Employed Laplace smoothing to handle zero-frequency features in probability estimation.
• Evaluated linkability using a large dataset of over 1 million reviews from approximately 2,000 active Yelp contributors.
• Treated each contributor’s set of reviews as a single author profile and tested the ability to group anonymous reviews to the correct profile.
• Assessed performance using precision, recall, and F1-score, with results reported across varying numbers of anonymous reviews per author.

Experimental results

Research questions

• RQ1 To what extent can anonymous reviews be linked to the same author using only basic text features?
• RQ2 How does the linkability of reviews vary with the number of reviews per contributor?
• RQ3 Can simple models like Naïve Bayes and KL divergence achieve high accuracy in identifying authorship from review text alone?
• RQ4 What are the implications of high linkability for user privacy across multiple review platforms?
• RQ5 Can this approach be used to detect self-reviewing or spam behavior in review ecosystems?

Key findings

• Up to 83% of anonymous reviews can be linked to their original authors using only alphabetical letter frequency distributions.
• Linkability reaches as high as 99% for contributors who have written 60 or more reviews.
• The Naïve Bayes model with simple features achieves high precision and recall, demonstrating strong performance even without complex linguistic features.
• The study confirms that users consistently expose identifiable traits in their writing, making them highly linkable across reviews.
• The results suggest that cross-referencing accounts across different review sites is highly feasible due to stable writing patterns.
• The method can be used by review platforms to provide users with feedback on the privacy risk of their reviews.