Table 6 Feature importance in the node placement classifier and the size prediction regression model at k = 2 and k = 5, both for the Twitter CVE dataset
From: On the challenges of predicting microscopic dynamics of online conversations
Feature | Node | Size (k=2) | Size (k=5) |
|---|---|---|---|
Global | |||
Current size | 0.029 | – | – |
Depth | – | 0.0 | 0.005 |
Root edges | – | 0.0 | 0.11 |
Recent edges | – | 0.0 | 0.0 |
Parent delay (mean, median) | – | 0.0, 0.0 | 0.083, 0.082 |
Node | |||
Node index | 0.54 | – | – |
Node depth | 0.28 | – | – |
Node parent delay | 0.013 | – | – |
Subtree size | 0.021 | – | – |
#Children | 0.043 | – | – |
#Grandchildren | 0.008 | – | – |
#Siblings | 0.012 | – | – |
#Cousins | 0.002 | – | – |
#Uncles | 0.002 | – | – |
User (mean and median over conversations initiated by the user) | |||
Max breadth | 0, 0 | – | – |
Depth | 0.003, 0.002 | – | – |
\(1{\text {st}}\) Breadth | \(<10^{-3}\), 0 | – | – |
Lifetime | 0, 0 | – | – |
Size | \(<10^{-3}\), \(<10^{-3}\) | – | – |
Root edges | \(<10^{-3}\), \(<10^{-3}\) | – | – |
Recent edges | 0.011, 0.011 | – | – |
#Posts | 0 | – | – |
Content (25 features, range given)) | |||
Root text | 0–0.002 | 0–0.21 | 0–0.12 |
Temporal | |||
Response time (mean, median) | – | 0.02, 0.006 | 0.028, 0.009 |
\(1{\text {st}}\) Response time | – | 0.43 | 0.028 |
Root time of day | \(<10^{-3}\) | 0.012 | 0.009 |
Root day of week | 0 | 0.0 | 0.0 |