abhijit-buet / continuous-emotion-prediction

Continuous Emotion Prediction was a challenge given in AVEC 2017. Three emotion dimension -arousal,valence & liking were predicted.LSTM-RNN model was used.Features were bag of words.Here various LSTM-RNN models were used to fit the data.1.Bidirectional LSTM,2.Multi tasking ,3.Many to many mapping.We got improvement from the baseline score as .562 for arousal,.543 for valence & .3512 for liking.

Continuous-Emotion-Prediction

Continuous Emotion Prediction was a challenge launched by AVEC 2017. Three emotion dimensions -arousal,valence and liking were required to predict.

Proposed Method

We trained a deep LSTM-RNN structure using Bag-of-Words features. Here Different LSTM-RNN variants were used to train, including, 1.Bidirectional LSTM, 2.Multi tasking , 3.Many to Many Mapping.

Result

For evaluation, Pearson Correlation Coefficient(PCC) was used. We got PCC about 0.562 for arousal, 0.543 for valence and 0.3512 for liking. Our work got 10% improvement from the baseline.