Systems Good for People Chained to the Rhythm Learning Analogies - PowerPoint PPT Presentation

Making Information Systems Good for People ⓒ

Chained to the Rhythm

Learning Analogies

Analogies Run Amok

What Happened?

What We Do • • • • • •

How Do They Work?

Finding Patterns

Recommender Vocabulary 🌾 🍖 ⚗ 🎂 🚳 👹 🙌 🕵 👰 👾 🌠 💜 👎💳

Recommender Architecture

User-Based Recommendations

Item-Based Recommendations

Matrix Factorization

Other Techniques • • • •

How Do We Know It Worked? Offline evaluation Online evaluation (A/B testing) Lab-style user studies

Experimental Protocol

building researching learning about

LensKit in Use • • • • • • • •

When Recommenders Fail Ekstrand and Riedl, RecSys 2012 😑 🙃 ☹

User-Perceived Differences Ekstrand et al., RecSys 2014

Experiment Features

Results in Differences • • • • • •

Problems with Evaluation Ekstrand and Mahant, FLAIRS 2017 • • ☒ •

Misclassified Decoys

Sturgeon’s Law

Sturgeon’s Decoys

Who Benefits from Recommendations?

Our Question

Fairness in Recommendation and Search Consumers Producers 🕻💄🤷🐷 🧕👶🎆 🧜👹🧚 Individuals Groups

Data • • • • • •

Gender

Age

Differences Exist • • ⇒ • •

Reciprocity [Franklin, 1989]

Giving Users a Voice

LITERATE

Sample of Ethical Issues • • • • • • • •

ACM Code of Ethics

Propagating Bias? (Under Review)

Feedback Loops (Future Work)

Promote Misinformation • • •

Segment Society

Promote Clickbait

Limits of Behavioral Observation

Information Disclosure

In Search of James Comey • • •

Fair Privacy (w/ Hoda Mehrpouyan, FAT* 2018) • • •

Beyond Recommenders • • • • • • •

The Real World of Technology • • • •

Paths Forward • • • • •

Thank You