Authorship Attribution via Stylometric Analysis and Shapley Values
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to effectively identify anonymous authors of online reviews, which can lead to privacy concerns and legal issues for both review platforms and individuals, while also impacting the utility and credibility of reviews.
Innovation Solution
A two-stage authorship attribution method combining structured data matching with text stylometric analysis, using machine-learning techniques such as random forest and support vector machines to identify anonymous reviewers by analyzing the characteristics of their reviews.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If review platforms display detailed user information (name, location, email) to enable author identification, then the ability to identify reviewers improves, but user privacy protection deteriorates
Solution Approach 1:
The patent introduces an intermediary system that sits between the review platform and potential data intruders. This system uses machine learning models trained on writing style characteristics to predict author identity without requiring direct access to detailed user information. The intermediary translates anonymous review text into probability distributions over possible authors, enabling identification while preserving the anonymity of the review platform's user database.
Solution Approach 2:
The patent replaces the mechanical system of direct information display (showing names, emails, locations) with a computational system based on text analysis. Instead of mechanically revealing identifying information, the system uses natural language processing and machine learning to infer authorship from writing style patterns, substituting direct observation with computational inference.
2Object-affected harmful factors
If review platforms implement strong anonymity protection (hiding all user information), then user privacy protection improves, but the utility of reviews for addressing specific concerns deteriorates
Solution Approach 1:
The patent segments the author identification process into two independent components: (1) the review platform maintains anonymous reviews with no user information exposed, and (2) a separate authorship attribution system uses text analysis to identify authors when needed. This segmentation allows the platform to preserve anonymity while enabling identification through a dedicated system that doesn't compromise the review's utility.
Solution Approach 2:
The authorship attribution system acts as an intermediary that preserves review utility without compromising anonymity. It enables vendors to identify reviewers for legitimate purposes (addressing complaints, improving service) while the review platform itself maintains strong anonymity protections. The intermediary translates between the anonymous review system and the need for author identification.
3Object-affected harmful factors
If platforms allow relatively anonymous reviewing, then user privacy protection improves, but the credibility and accountability of reviews deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where the authorship attribution system provides probability distributions to the review platform. This feedback enables the platform to monitor for potential fake reviews, coordinated attacks, or patterns of malicious behavior while still maintaining reviewer anonymity. The feedback loop allows credibility assessment without revealing identities, balancing anonymity with accountability.
Data Source
AI summary
Privacy, protection, and de-anonymization are issues of societal importance that are implicitly at the core of several key information systems, from electronic health records to online reviews. The system and method herein allows for an identification of an author of anonymous writing based on the text and structured data, subject to practical constraints on the intruder's amount of training data and effort using Shapley values.