PhD Thesis Defense: Spencer Lane, Tools for Aggregating, Filtering, Sorting, and Auditing Social Media
Content
Speaker:
Abstract:
The algorithms used to sort and filter social media feeds are generally opaque. The goals of these algorithms are not visible to the user nor is there any way for the user to evaluate to what extent their goals and the goals of the platform are aligned. While there are many proposed solutions, most require actions that are outside the control of end users. Regulatory solutions or modifications to the algorithms themselves require platforms to change their behavior. This dissertation argues for a different approach: empowering users to take control of their social media through the use of a loyal client - a client application that is aligned with the preferences of the user, not the platform.
This work presents three contributions in service of the loyal client concept. The first contribution is the articulation of the loyal client concept, presenting principles for evaluating client loyalty and presenting examples of historical clients that meet these criteria. A loyal client should be protective, interoperable, customizable, and usable.
The second contribution is the design and architecture of a social media client and recommender system built around these principles. This system, Gobo, aggregates content across multiple platforms and combines rules-based recommendation with machine learning models to surface content aligned with user preferences. We conducted a survey evaluating these design principles and present results indicating that users are generally receptive.
The third contribution is a system to allow the auditing of the recommender system, enabling users to evaluate the performance of the algorithms curating their feeds. This work presents the auditing system as well as the design and results of a user study conducted to evaluate the auditing system.
Taken together, these contributions demonstrate that the loyal client is a feasible and promising approach to addressing problems with modern social media, and that algorithmic auditing by end users is achievable.
Advisor:
Ethan Zuckerman