KondoCloud: Improving Information Management in Cloud Storage via Recommendations Based on File ...
summary
KondoCloud is a web file-browser that surfaces ML recommendations to retrieve, move, or delete cloud files by comparing each file to others using logistic regression classifiers over 10 metadata and content features (similarities pre-computed). Trained from 69 participants and crowd-labeled pairs, classifiers reach F1 >= 0.72. In a 62-participant evaluation the system shows a 25.5% accepted rate (36.2% of recommendations completed including manual actions) and increases cross-folder deletions, though scalability and grouped recommendations remain open issues.
AI-written summary · about
chapters (13)
- 0:00Intro
- 1:09Methodology
- 2:06Observation Study Summary
- 2:34Distribution of Actions
- 3:10Co-occurrence of actions
- 3:39Actions over time
- 4:34KondoCloud Recommendations
- 4:55Recommendation Specifics
- 5:45Evaluation Study Participants
- 5:57Recommendation Acceptance (36 participants)
- 6:46Without Recommendations
- 7:10Recommendations at a Distance
- 7:49Improvements
related talks
34mMoving notebooks into the cloud: challenges and lessons learned
34mScaling Cloud Development Environments on K8s: Enterprise-Scale Adoption Lessons
43mHow KION builds realtime personalized recommendations and storefronts
35mEnhancing User Engagement Through Personalized Recommendation at Dow Jones
13mAI-Driven Data Integration at Scale
28mScaling Security Anomaly Detection in Enterprise Knowledge Base with Dask
43mRecommendation Systems in Market Research
42mWhich Car Fits My Life? - mobile.de’s Approach to Recommendations