Based on the work from SemEval-2019 paper: CLARK at SemEval-2019 Task 3: Exploring the Role of Context toIdentify Emotion in a Short Conversation.
# To train
python -m semeval.train --input train.csv --output semeval-clark.mdl
# To predict
python -m semeval.predict --input new_work.csv --saved-model semeval-clark.mdl
# To reproduce results of paper
python -m semeval.test --input dev.csv --saved-model semeval-clark.mdl
id,turn1,turn2,turn3,label
0,Hi!,How are you?,I'm good :),happy
1,Who do you think you are?, I'm just a guy, Well you suck,angry
...
In addition, further work has been done using appraisal variables Patterns of cognitive appraisal in emotion. To use this version of the CLARK, see below:
# To train
python -m train --input train.json --output av-clark.mdl
# To predict
python -m predict --input new_work.json --saved-model av-clark.mdl
# To reproduce results of paper
python -m test --input dev.json --saved-model av-clark.mdl
{
"id":2,
"turn1":"Hi!",
"turn2":"How are you",
"turn3":"I'm good :)",
"label":"happy",
"numUtterances":3,
"hitCountEmotion":5,
"hitCountAppraisal":8
},
...
[
{
"id": 12,
"turn1":
{
"emotion": "joy",
"appraisals": {
"pleasantness": 2.0, "attention": 1.0, "control": 1.0, "certainty": 2.0, "anticipated_effort": 1.0, "responsibility": 1.0
}
},
"turn2":
{
"emotion": "joy",
"appraisals": {
"pleasantness": 2.0, "attention": 1.0, "control": 2.0, "certainty": 2.0, "anticipated_effort": 1.0, "responsibility": 1.0
}
},
"turn3":
{
"emotion": "challenge",
"appraisals": {
"pleasantness": 1.0, "attention": 0.0, "control": 2.0, "certainty": 0.0, "anticipated_effort": 0.0, "responsibility": 2.0
}
},
"fleiss_kappa": 0.44148936170212755
},
...
]