Upload 2 files

app.py

@@ -3,6 +3,7 @@ from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline
 from sentence_transformers import SentenceTransformer, util
 import gradio as gr
 import json
+from fuzzywuzzy import fuzz
 
 # Load pre-trained BERT QA model and tokenizer from Hugging Face model hub
 model_name = "bert-large-uncased-whole-word-masking-finetuned-squad"
@@ -69,7 +70,7 @@ def get_dynamic_context_semantic(question, knowledge_base, knowledge_base_embedd
         for content_item in best_match_entry['content']:
             if 'faq' in content_item:
                 for faq in content_item['faq']:
-                    if faq['question'].lower() in question.lower():
+                    if fuzz.token_sort_ratio(faq['question'].lower(), question.lower()) > 80:
                         return faq['answer']
 
         # If no FAQ is found, check for steps
@@ -83,12 +84,13 @@ def get_dynamic_context_semantic(question, knowledge_base, knowledge_base_embedd
             if 'text' in content_item:
                 return content_item['text']
     
-    return "Lo siento, no encontré una respuesta adecuada para tu pregunta."
+    return "Lo siento, no encontré una respuesta adecuada a tu pregunta."
 
-# Check expanded QA dataset first for a direct answer
-def get_answer_from_expanded_qa(question, expanded_qa_dataset):
+# Use fuzzy matching to find the closest match in the expanded QA dataset
+def get_answer_from_expanded_qa(question, expanded_qa_dataset, threshold=80):
     for item in expanded_qa_dataset:
-        if item['question'].lower() == question.lower():
+        # Use fuzzy matching to find close matches
+        if fuzz.token_sort_ratio(item['question'].lower(), question.lower()) > threshold:
             return item['answer']
     return None
 
@@ -100,7 +102,7 @@ def answer_question(question):
         return direct_answer
 
     # If no direct answer found, use the knowledge base with semantic search
-    context = get_dynamic_context_semantic(question, knowledge_base, knowledge_base_embeddings, threshold=0.55)
+    context = get_dynamic_context_semantic(question, knowledge_base, knowledge_base_embeddings, threshold=0.45)
     return context
 
 # Gradio interface setup
@@ -113,4 +115,4 @@ interface = gr.Interface(
 )
 
 # Launch the Gradio interface
-interface.launch()
+interface.launch(share=True)

requirements.txt

@@ -1,5 +1,6 @@
-gradio
-torch
-transformers
-datasets
-sentence-transformers
+transformers==4.25.1
+torch==1.13.1
+sentence-transformers==2.2.2
+fuzzywuzzy==0.18.0
+scikit-learn==1.1.3
+gradio==3.16.2