Initialize project and create a first sentiment classifier prototype with streamlit and a pretrained bert model

.gitignore

@@ -0,0 +1,4 @@
+__pycache__
+*.csv
+.DS_Store
+*.h5

README.md

@@ -0,0 +1 @@
+# bert-sentiment-analysis

models/models.py

@@ -0,0 +1,21 @@
+"""
+Module to load the project models
+"""
+import os
+import tensorflow_text
+import tensorflow as tf
+import tensorflow_hub as hub
+
+
+CURRENT_DIR = os.path.dirname(os.path.abspath(__file__))
+
+
+def load_sentiments_model():
+    """
+    Load pretrained model
+    """
+    model_path = os.path.join(CURRENT_DIR, "sentiments_bert_model.h5")
+    model = tf.keras.models.load_model(
+        model_path, custom_objects={"KerasLayer": hub.KerasLayer}, compile=False
+    )
+    return model

requirements.txt

@@ -0,0 +1,5 @@
+pandas
+jupyter
+numpy
+tensorflow
+tensorflow-text

sentiment_analysis.py

@@ -0,0 +1,46 @@
+"""
+Sentiment analysis streamlit webpage
+"""
+import streamlit as st
+from sentiment_classificator import classify_sentiment
+
+
+def get_representative_emoji(sentiment: str) -> str:
+    """
+    From a sentiment return the representative emoji
+    """
+    if sentiment == 'positive':
+        return "😃"
+    elif sentiment == 'negative':
+        return "😞"
+    else:
+        return "😐"
+
+
+def main() -> None:
+    """
+    Build streamlit page for sentiment analysis
+    """
+    st.title("Sentiment Classification")
+
+    # Initialize session state variables
+    if 'enter_pressed' not in st.session_state:
+        st.session_state.enter_pressed = False
+
+    # Input text box and button
+    input_text = st.text_input("Enter your text here:")
+    button_clicked = st.button("Classify Sentiment")
+
+    if button_clicked or st.session_state.enter_pressed:
+        # Process the input text with the sentiment classifier
+        sentiment = classify_sentiment(input_text)
+
+        # Get the representative emoji
+        emoji = get_representative_emoji(sentiment)
+
+        # Show the response and emoji
+        st.write(f"Sentiment: {sentiment.capitalize()} {emoji}")
+
+
+if __name__ == "__main__":
+    main()

sentiment_classificator.py

@@ -0,0 +1,34 @@
+"""
+Module to classify text into positive or negative sentiments
+"""
+import sys
+import tensorflow as tf
+from models.models import load_sentiments_model
+
+sentiments_model = load_sentiments_model()
+MAX_NEG = 0.4
+MIN_POS = 0.6
+
+
+def classify_sentiment(input_text: str) -> str:
+    """
+    Receives a string and classifies it in positive, negative or none
+    """
+    result = tf.sigmoid(sentiments_model(tf.constant([input_text])))
+    if result < MAX_NEG:
+        return "negative"
+    elif result > MIN_POS:
+        return "positive"
+    else:
+        return "-"
+
+
+if __name__ == "__main__":
+    if len(sys.argv) < 2:
+        print(
+            f"Usage: python {sys.argv[0]} <text to classify>")
+        sys.exit(1)
+    # Get the input string from command line argument
+    input_text = sys.argv[1]
+    sentiment = classify_sentiment(input_text)
+    print("Sentiment of the sentence: ", sentiment)