README.md

metadata

widget:
  - text: >-
      Dih apaan banget dah buang sampah ke sungai begitu. Ada aktivis lingkungan
      yg sampe dipenjara karena menyuarakan peduli lingkungan. Ini pengangguran
      satu malah enak bener buang sampah sembarangan. Pantes lu susah, kelakuan
      lu nyusahin orang lain sih.
    example_title: Example 1
    output:
      - label: Disgust
        score: 0.672
      - label: Anger
        score: 0.282
      - label: Sadness
        score: 0.033
      - label: Joy
        score: 0.004
      - label: Surprise
        score: 0.003
      - label: Trust
        score: 0.003
      - label: Fear
        score: 0.002
      - label: Anticipation
        score: 0.001
  - text: >-
      Februari 2009, wartawan Jawa Pos Radar Bali dibunuh dengan keji karena
      berita korupsi. Januari 2019, Presiden memberikan grasi kepada otak
      pembunuhan Prabangsa, dari seumur hidup menjadi cuma 20 tahun penjara.
      Sebuah langkah mundur yang menyakitkan!
    example_title: Example 2
    output:
      - label: Sadness
        score: 0.604
      - label: Anger
        score: 0.194
      - label: Surprise
        score: 0.127
      - label: Joy
        score: 0.021
      - label: Fear
        score: 0.018
      - label: Disgust
        score: 0.018
      - label: Anticipation
        score: 0.016
      - label: Trust
        score: 0.003
  - text: >-
      Salut banget sama perjalanan hidup mereka ini kalo diproduksi jadi film
      pasti bakal rame dan menginspirasi banget woi
    example_title: Example 3
    output:
      - label: Joy
        score: 0.9637
      - label: Trust
        score: 0.0219
      - label: Anticipation
        score: 0.0079
      - label: Surprise
        score: 0.0029
      - label: Disgust
        score: 0.0013
      - label: Sadness
        score: 0.001
      - label: Anger
        score: 0.0007
      - label: Fear
        score: 0.0006
  - text: >-
      SUMPAH HARUS DIBEBASKAN!!! KENAPA GAK TANGKEPIN KORUPTOR AJA DARIPADA
      NGURUSIN MEME DARI AI GW MARAH BANGET SHIBAL
    example_title: Example 4
    output:
      - label: Anger
        score: 0.9889
      - label: Disgust
        score: 0.0035
      - label: Sadness
        score: 0.0026
      - label: Fear
        score: 0.0015
      - label: Surprise
        score: 0.0012
      - label: Trust
        score: 0.0011
      - label: Anticipation
        score: 0.0009
      - label: Joy
        score: 0.0003
  - text: >-
      ga pernah pacaran, sekarang hidup kesepian bgt. pengen minta kenalin cowo
      ke temen tp mereka jg sama struggle nya. jd nyesel dulu pas sekolah-kuliah
      kenapa ga pernah 'macem2'
    example_title: Example 5
    output:
      - label: Sadness
        score: 0.9526
      - label: Anger
        score: 0.0175
      - label: Fear
        score: 0.0114
      - label: Disgust
        score: 0.0079
      - label: Trust
        score: 0.0038
      - label: Anticipation
        score: 0.0036
      - label: Joy
        score: 0.0019
      - label: Surprise
        score: 0.0013
  - text: >-
      Komisi Penyiaran Indonesia (KPI) meminta agar tayangan televisi
      menampilkan citra positif Polri secara edukatif dan akurat. Hal ini
      disampaikan ketua KPI Pusat Ubaidillah dalam sebuah diskusi panel
    example_title: Example 6
    output:
      - label: Anticipation
        score: 0.4323
      - label: Trust
        score: 0.3996
      - label: Joy
        score: 0.05
      - label: Anger
        score: 0.0388
      - label: Disgust
        score: 0.0362
      - label: Surprise
        score: 0.0186
      - label: Fear
        score: 0.0137
      - label: Sadness
        score: 0.0108
library_name: transformers
license: mit
language:
  - id

Model Details

Model Description

The EmoSense-ID is a model designed to identify and analyze emotions in Indonesian texts based on Plutchik's eight basic emotions: Anticipation, Anger, Disgust, Fear, Joy, Sadness, Surprise, and Trust. This model is developed using the NusaBERT-base and trained using Indonesian tweets categorized into eight emotion categories. The evaluation results of this model can be utilized to analyze emotions in social media, providing insights into users' emotional responses.

Bias

Keep in mind that this model is trained using certain data which may cause bias in the emotion classification process. Therefore, it is important to consider and account for such biases when using this model.

Evaluation Results

The model was trained using the Hyperparameter Tuning technique with Optuna. In this process, Optuna conducted five trials to determine the optimal combination of learning rate (1e-6 to 1e-4) and weight decay (1e-6 to 1e-2). Each trial trained the BERT model with different hyperparameter configurations on the training dataset and then evaluated using the validation dataset. After all the experiments are completed, the best hyperparameter combination is used to train the final model. 

Epoch	Training Loss	Validation Loss	Accuracy	F1	Precision	Recall
1	0.758400	0.583508	0.829932	0.830203	0.833136	0.829932
2	0.370100	0.394630	0.866213	0.865496	0.870364	0.866213
3	0.231500	0.355294	0.884354	0.884585	0.888140	0.884354
4	0.071000	0.322376	0.902494	0.902801	0.904842	0.902494
5	0.129900	0.308596	0.900227	0.900340	0.902132	0.900227

Citation

@misc{Ardiyanto_Mikhael_2024,
    author    = {Mikhael Ardiyanto},
    title     = {EmoSense-ID},
    year      = {2024},
    URL       = {Aardiiiiy/EmoSense-ID-Indonesian-Emotion-Classifier},
    publisher = {Hugging Face}
}