Using machine learning to detect sensitive documents on SharePoint
summary
Project applies Random Forest one‑vs‑all classifiers to bag‑of‑words vectors plus regex for secrets to flag sensitive SharePoint files and API keys, deployed as a pipeline that queries the SharePoint API, scores files, and prompts owners via a user notifier app whose responses feed back into training. The system prioritizes low false positives for analyst fatigue, reports per‑document confidence scores, and plans experiments with gradient boosted trees and BERT and scanning S3/Azure/GCP.
AI-written summary · about
chapters (19)
- 0:00Introduction and motivation
- 2:10Talk overview
- 2:51What is machine learning
- 3:52Document classification basics
- 4:13Defining sensitive documents
- 6:00Types of sensitive documents
- 6:43Processing document text
- 8:11Bag of words vectorization
- 8:32Decision trees explained
- 10:14Random forests for classification
- 11:39One versus all classifiers
- 12:42Detecting sensitive info with regex
- 13:44Deploying ML models pipeline
- 14:48Key challenges in project
- 15:50Minimizing false positives
- 16:10Building feedback loops
- 16:33Future projects and expansion
- 17:17Closing and acknowledgments
- 18:01Q&A session
related talks
28mScaling Security Anomaly Detection in Enterprise Knowledge Base with Dask
10mData Security & AI Adoption: A Leader’s Perspective (Customer Spotlight)
46mHunting for Vulnerabilities in Microsoft SharePoint: Case Study
32mWeaponizing Python for Good: Building a Next-Gen CVE Scanner That Detects Zero-Day Vulnerability
48mMethod of detecting vulnerability in WebApps using Machine Learning
31mBuild threat detection models using ML
41mModernizing the Application Penetration Engagement and Reporting Lifecycle
50mNotes from the field Microsoft Information Protection