Authorship Verification Using Mixture of Experts Voting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current authorship attribution and verification technologies face challenges in accurately determining authorship in open-set scenarios, where the answer can be 'none of the above,' and are flawed in reliability when verifying authorship for a single candidate.
Innovation Solution
A novel approach using a 'mixture of experts' method with multiple automated authorship attribution systems and a voter box to analyze a pool of authors, allowing for the determination of author identity, including the possibility of 'none of the above,' by applying mathematical analysis to proportions of votes from a large number of experts, and utilizing only the candidate's prior writings for verification without distractor sets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional authorship attribution methods using distractor sets are used, then the system can compare candidate author against other authors, but the system cannot reliably determine 'none of the above' in open-set scenarios and produces unwarranted assumptions
Solution Approach 1:
The patent extracts and eliminates the distractor set from the authorship verification process. Instead of comparing a candidate author against a set of other authors' writings, the system uses only the candidate author's own prior writings as samples. This extraction of the problematic distractor set allows the system to reliably determine authorship verification without being forced to choose from predetermined authors, enabling accurate 'none of the above' conclusions in open-set scenarios.
Solution Approach 2:
The patent inverts the traditional authorship attribution approach by switching from a closed-set comparison model (candidate vs. distractor authors) to an open-set verification model (candidate vs. candidate's own prior work). This inversion allows the system to verify whether a specific author wrote a text without being constrained by predetermined author lists, thereby enabling reliable detection when none of the named authors is the actual author.
2Measurement precision
If multiple automated authorship attribution systems are used in a mixture of experts approach, then the measurement precision improves, but the device complexity increases significantly
Solution Approach 1:
The patent merges multiple automated authorship attribution systems into a unified verification framework. Instead of treating each system as a separate entity, the invention combines their outputs through a voter box that aggregates votes from multiple experts (automated systems) to reach a consensus decision. This merging approach maintains the measurement precision benefits of multiple systems while managing complexity through a unified architectural structure.
Solution Approach 2:
The patent creates a universal verification framework that can handle multiple authorship attribution systems through a single voter box interface. The voter box serves multiple functions: it collects votes from different automated systems, aggregates their results, and produces a unified verification outcome. This multi-functionality allows the system to leverage the precision of multiple experts without requiring separate processing pipelines for each system.
Data Source
AI summary
Novel distractorless authorship verification technology optionally combines with novel algorithms to solve authorship attribution as to an open set of candidates—such as without limitation by analyzing the voting of “mixture of experts” and outputting the result to a user using the following: if z (z=pi−pj√ pi+pj−(pi−pj)2/n) is larger than a first predetermined threshold then author j cannot be the correct author; or if z (z=pi−pj√ pi+pj−(pi−pj)2/n) is smaller than a second predetermined threshold then author i cannot be the correct author; or if no author garners significantly more votes than all other contenders then none of the named authors is the author of a document in question—in a number of novel applications. Personality profiling and authorship attribution may also be used to verify user identity to a computer.