app.py

import streamlit as st
from pipeline.model import MultiTaskModel
from pipeline.preprocessing import Preprocessor, EMOTION_LABELS, TOXICITY_LABELS

# Load the model
st.title("ChattyTicket Model: Emotion and Toxicity Classification of Valorant chat messages")

st.write("""Discover our innovative model designed to tackle toxicity in Valorant, a popular multiplayer game. Leveraging advanced multi-task learning, our model combines Bi-LSTM and BERT architectures for superior performance. This approach enables accurate classification of toxic conversations and emotional content within the game.

Key Features:
 - Multi-Task Learning: Simultaneous classification of toxicity and emotion.
 - Bi-LSTM & BERT Integration: Enhanced accuracy with a BERT pre-trained backbone.
 - High Accuracy: 91.81% in toxicity detection and 86.74% in emotion prediction.
Insights & Impact:
 - Emotional Landscape: Identifies prevalent emotions like anger and instances of cyberbullying.
 - Healthier Communities: Provides insights for fostering positive gaming environments.
ChattyTicket API:
    - Evaluate chat text for emotion and toxicity with our user-friendly API. Positive user feedback highlights its effectiveness and interactivity, with ongoing improvements based on user input to better handle nuances like sarcasm.

Experience the future of online gaming moderation with our model and contribute to a healthier, more enjoyable gaming community.
         """
)

st.table(
    {
        "Emotion": EMOTION_LABELS,
        "Toxicity": TOXICITY_LABELS,
    }
)

@st.cache_resource
def loading_model(bert_base):
    if bert_base == "bert-base-uncased":
        print("Loading base model")
        return MultiTaskModel(preprocessor=Preprocessor())
    elif bert_base == "bert-base-multilingual-cased":
        print("Loading multilingual model")
        return MultiTaskModel(is_multilingual=True, preprocessor=Preprocessor(is_multilingual=True))
    else:
        return None
    
def clear():
    loading_model.clear()

bert_base = st.selectbox("Select a model", ("bert-base-uncased", "bert-base-multilingual-cased"), placeholder="Select a model", on_change=clear)


model = loading_model(bert_base)

# Get user input
user_input = st.text_input("Enter a Valorant chat message:")
st.write("You entered:", user_input)
# Predict
prediction = model.predict(user_input)
emotions, toxicitys = prediction
decoded = model.decode(prediction)

col1, col2 = st.columns(2)

with col1:
    st.write(f"The predicted emotion is: {decoded[0][0][0]}" )
    for i, emotion in enumerate(emotions[0]):
        st.write(f"{EMOTION_LABELS[i]}: {(emotion*100):.2f}%")
        st.progress(float(emotion))
with col2:
    st.write(f"The predicted toxicity is: {decoded[1][0][0]} ")
    for i, toxicity in enumerate(toxicitys[0]):
        st.write(f"{TOXICITY_LABELS[i]}: {(toxicity*100):.2f}%")
        st.progress(float(toxicity))