[Note] 人類行為大數據分析: 資料科學如何應用在教育及醫療領域
講者:李祈均
**What is BEHAVIORAL SIGNAL PROCESSING (BSP) ?
**Compute Human Behavior Traits and States for Domain Experts Decision Making
- Help experts to do things they know in a more efficient manner at scale
- Develop novel behavioral analytics framework for possible scientific discovery
人類行為分析方式
- 訊號產生 — -> 訊號擷取 — -> 訊號分析 — -> 決策(行為判定)
- 單向:
行為產生機制 → 外顯行為 →內在狀態 行為認知機制 → 主觀量測 → 決策
行為產生機制 → 外顯行為 — -> 內在狀態 行為認知機制 → Self Report → 決策
- 雙向: 行為產生機制 → 外顯行為 — -> 行為認知機制 → 行為產生機制 → 外顯行為 — -> 行為認知機制 → 主觀量測 → 決策
COMPUTING BEHAVIORAL TRAITS & STATES FOR DECISION MAKING & ACTION
- QUANTITATIVE
- EFFICIENCY
- SUPPLMENTARY
- POSSIBILITY
ELEMENTS OF BSP
- Fetching
- Pre-processing
- Modeling
- Assessment
資料代表性
- 辨識目標的評分
1. 運用 established instrument
2. Scientific-rigor
3. Ensure domain-applicable analytics 產出 - 如何找出有用的資料
1. 討論、討論、討論
2. 收的過程影響後面預處理
3. 想像未來應用
4. 領域專家意見太重要 - ADOS (Autism Diagnostic Observation Schedule)
1. Subject interacts with a psychologist for ~45 minutes
2. Current gold standard, research-level observational coding
3. Psychologists are trained using stringent training protocol
4. Semi-structured assessment in eliciting socio-communicative behavior of the ASD children for diagnostics
5. Multiple subparts events (14) on rating of a wide range number of socio-communicative behavior (28)
充分預處理
訊號預處理
Label 預處理
- **Voice Activity Detector
**Speech signal
1. Energy every frame
frame = 25ms
standard deviation (normalize D.C. offset)
2. 閥值Threshold
speech percentage in the wav
3. Speech Segments
Energy > Threshold Energy - **建議
**1. 資料收集過程重要、重要、重要
2. 熟悉資料是必要、必要、必要
3. 有些真的太錯、知道錯、不要用
4. 要被學的label需要被分析徹底
5. domain experts 很厲害
6. 跨領域閱讀