You shall not share!
summary
Agent predicts a user’s private vs public label from that user’s past posts only, representing each item by 20 tags per image (Clarifai or user tags), weighting them with TF‑RF features and classifying with random forest (100 trees). Prototype runs on a Piccolo subset with 59 agents, averaging results over 10 runs, uses cosine similarity threshold 0.7 to replace unseen tags, and shows accuracy gains when recommending tags; no user feedback loop exists yet.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 3:08Approach
- 4:10TFRF
- 6:30Vector representations
- 8:26Sample scenario
- 12:31Dataset
- 13:17Simulation
- 14:29Libraries
- 15:14Performance
- 16:39Frequency
- 18:17Questions
- 20:29Feature vectors
- 21:36Privacy estimation
- 23:34Birth
- 24:01Results
related talks
03mServing Billions of Personalized News Feeds with AI
16mMaking recommendations explainable
18mInstagram in focus
28mMaking Recommendations without Data
42mModeling Aesthetic Identity- Building a Digital Twin
01h 01mWe are the Guardians of our Future
20mAn Innovative Approach to Labeling Ground Truth in Speech
32mFeed generation at ShareChat