Update app.py

app.py

@@ -2,18 +2,22 @@ import streamlit as st
 import pandas as pd
 
 from transformers import pipeline
+from sentence_transformers import SentenceTransformer, util
+semantic_model = SentenceTransformer('all-MiniLM-L6-v2')
 
 @st.cache(allow_output_mutation=True)
 def get_model(model):
 	return pipeline("fill-mask", model=model, top_k=100)#seto maximum of tokens to be retrieved after each inference to model
 
+
 HISTORY_WEIGHT = 100 # set history weight (if found any keyword from history, it will priorities based on its weight)
 
 st.caption("This is a simple auto-completion where the next token is predicted per probability and a weigh if appears in user's history")
 
-history_keyword_text = st.text_input("Enter users's history keywords (optional, i.e., 'Gates')", value="Gates")
+history_keyword_text = st.text_input("Enter users's history keywords (optional, i.e., 'Gates')", value="")
 
 text = st.text_input("Enter a text for auto completion...", value='Where is Bill')
+semantic_text = st.text_input("Enter users's history semantic (optional, i.e., 'Microsoft')", value="Microsoft")
 
 model = st.selectbox("choose a model", ["roberta-base", "bert-base-uncased"])
 
@@ -24,12 +28,26 @@ if text:
     data_load_state = st.text('Inference to model...')
     result = nlp(text+' '+nlp.tokenizer.mask_token)
     data_load_state.text('')
+    
+    predicted_embeddings = model.encode(result['sequence'], convert_to_tensor=True)
+    semantic_history_embeddings = model.encode(semantic_text.spllit(','), convert_to_tensor=True)
+    
+    cosine_scores = util.cos_sim(embeddings1, embeddings2)
+    
     for index, r in enumerate(result):
+        result[index]['score']=cosine_scores[index][index]
         if r['token_str'].lower().strip() in history_keyword_text.lower().strip() and len(r['token_str'].lower().strip())>1:
             #found from history, then increase the score of tokens
             result[index]['score']*=HISTORY_WEIGHT
             
+    
+    
     #sort the results        
     df=pd.DataFrame(result).sort_values(by='score', ascending=False)
+    
+    
+
+
+    
     #show the results as a table
     st.table(df)