Table 5.
Accuracy and features for 7 machine learning models.
| Machine learning model and top 10 features | Accuracy (Pearson r)a | Feature weight of SMOregb | |
| Positive emotional self-disclosure | .44 |
|
|
|
|
Positive emotion |
|
0.32 |
|
|
Word count per sentence |
|
0.28 |
|
|
Religion |
|
0.25 |
|
|
<Please + VERB> |
|
–0.21 |
|
|
Sentence count |
|
0.16 |
|
|
<SUBJECT_I + positive_ADJECTIVE> |
|
0.13 |
|
|
Negation |
|
–0.10 |
|
|
We |
|
0.07 |
|
|
Financial concerns |
|
–0.07 |
|
|
Strong subjectivity |
|
0.07 |
| Negative emotional self-disclosure | .59 |
|
|
|
|
Anxiety |
|
1.18 |
|
|
Anger |
|
0.51 |
|
|
<SUBJECT_I> |
|
0.40 |
|
|
Sadness |
|
0.28 |
|
|
<SUBJECT_I + negative_ADJECTIVE> |
|
0.27 |
|
|
Death |
|
0.23 |
|
|
Negation |
|
0.18 |
|
|
Strong subjectivity |
|
0.17 |
|
|
Word count per sentence |
|
0.14 |
|
|
Sentence count |
|
0.14 |
| Positive informational self-disclosure | .45 |
|
|
|
|
Positive emotion |
|
0.31 |
|
|
Religion |
|
0.27 |
|
|
Sadness |
|
–0.25 |
|
|
Sentence count |
|
0.25 |
|
|
Word count per sentence |
|
0.23 |
|
|
<Please + VERB> |
|
–0.20 |
|
|
<SUBJECT_I + positive_ADJECTIVE> |
|
0.16 |
|
|
Routine and schedule |
|
0.13 |
|
|
Biological processes |
|
–0.13 |
|
|
Auxiliary verb |
|
–0.12 |
| Negative informational self-disclosure | .64 |
|
|
|
|
Anxiety |
|
0.42 |
|
|
Sentence count |
|
0.41 |
|
|
Any |
|
0.32 |
|
|
Biological processes |
|
0.28 |
|
|
Tumor treatment |
|
0.26 |
|
|
<SUBJECT_I> |
|
0.26 |
|
|
<SUBJECT_I + positive_ADJECTIVE> |
|
–0.25 |
|
|
Anger |
|
0.24 |
|
|
I |
|
0.23 |
|
|
Lymphedema |
|
0.21 |
| Question asking | .78 |
|
|
|
|
Sentence count |
|
–0.82 |
|
|
Religion |
|
–0.72 |
|
|
Word count per sentence |
|
–0.64 |
|
|
Positive emotion |
|
–0.59 |
|
|
Question marks |
|
0.52 |
|
|
Any |
|
0.50 |
|
|
Proper nouns |
|
–0.40 |
|
|
<Please + VERB> |
|
0.36 |
|
|
Spiritual |
|
–0.30 |
|
|
Negation |
|
0.27 |
| Emotional support provision | .81 |
|
|
|
|
Sentence count |
|
0.55 |
|
|
Emotional support |
|
0.46 |
|
|
We |
|
0.45 |
|
|
She/He |
|
–0.44 |
|
|
You |
|
0.37 |
|
|
Question marks |
|
–0.33 |
|
|
Strong subjectivity |
|
0.24 |
|
|
Adjusting to diagnosis |
|
0.23 |
|
|
Be verbs |
|
0.23 |
|
|
Positive life events |
|
–0.23 |
| Informational support provision | .85 |
|
|
|
|
Sentence count |
|
1.13 |
|
|
Word count per sentence |
|
0.38 |
|
|
Question marks |
|
–0.33 |
|
|
Spiritual |
|
–0.26 |
|
|
Postsurgery problems |
|
0.22 |
|
|
I |
|
–0.20 |
|
|
<If + you> |
|
0.20 |
|
|
Strong subjectivity |
|
–0.19 |
|
|
Forum communication |
|
–0.17 |
|
|
Tumor treatment |
|
0.16 |
a The accuracy correlation is the Pearson product moment correlation between the average of 10 human judgments and the output of the machine learning model.
b The output feature weight of the support vector machine regression model shows the strength of the association between the presence of a feature in a message and human judgments of that message.