final v

app.py

@@ -2,11 +2,11 @@ import streamlit as st
 from transformers import pipeline, AutoModelForSequenceClassification, AutoTokenizer
 
 
-@st.cache(allow_output_mutation=True)
+# @st.cache(allow_output_mutation=True)
 def get_model():
     tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
     pulled_model = AutoModelForSequenceClassification.from_pretrained(
-        "NazmusAshrafi/stock_twitter_sentiment_Bert")
+        "NazmusAshrafi/stock_twitter_topic_Bert")
     return tokenizer, pulled_model
 
 
@@ -19,13 +19,36 @@ classifier = pipeline("sentiment-analysis",
 
 # print(classifier("AAP Up Theory = Will go up in future, buy now"))
 
-st.title("Topic Modeling and Sentiment Analysis of Stock Tweets")
+st.title("Find the topic of a stock related tweets")
+st.subheader(
+    'This model can predict 3 topics - :blue[Investment decision], :green[User Volume], :orange[Market crisis] - Entering a topic related tweet will yeild the best results')
+
+
+st.markdown(
+    ':blue[Investment decision example: "AAP Dow Theory = Will go down in future, Do not buy"]')
+st.markdown(
+    ':green[User Volume example: "Loosing a lot of volume, user are leaving this platform"]')
+st.markdown(
+    ':orange[Market crisis example: "Market is in danger because of corona virus"]')
+
+
+st.subheader("", divider='rainbow')
+# 0 = Investment decision
+# 1 = User volume
+# 2 = Market crisis
+
 
 user_input = st.text_area("Enter a tweet about a stock")
 button = st.button("Analyze")
 
+# print(classifier("AAP Down Theory = Will go down in future,do not buy now"))
+
 
 if user_input and button:
     # output
-    classifier("AAP Up Theory = Will go up in future, buy now")
-    st.write("Raw Pred: ", classifier(user_input))
+    st.write("Topic Prediction: ", classifier(user_input)[0]['label'])
+    st.write("Confidence Score: ", classifier(user_input)[0]['score'])
+
+
+st.markdown(
+    'Note: Model may be predicting "Investment decision" way too often, this is because of the data imbalance. The model was trained on 3 topics, but the data was not balanced. The model was trained on a dataset where a large portion of the data was on "Investment decision" topic. This is why the model is biased towards "Investment decision" topic. This is a flaw of this model, and it can be fixed by training the model on a balanced dataset.')