Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
|
@@ -52,31 +52,26 @@ FAQ_DATA = [
|
|
| 52 |
|
| 53 |
def perform_nlp_cosine_matching(user_query):
|
| 54 |
"""
|
| 55 |
-
|
| 56 |
-
|
| 57 |
"""
|
| 58 |
-
# Step 2: Preprocess text using NLP approaches (Lowercase + NLTK Tokenization)
|
| 59 |
try:
|
| 60 |
tokens = nltk.word_tokenize(user_query.lower())
|
| 61 |
cleaned_query = " ".join([w for w in tokens if re.match(r'^\w+$', w)])
|
| 62 |
except Exception:
|
| 63 |
-
# Secure fallback text parsing if NLTK files fail initialization on basic CPU
|
| 64 |
cleaned_query = " ".join(re.findall(r'\w+', user_query.lower()))
|
| 65 |
|
| 66 |
if not cleaned_query.strip():
|
| 67 |
return None, 0.0
|
| 68 |
|
| 69 |
-
# Extract our collected question list text array
|
| 70 |
questions_corpus = [faq["q"] for faq in FAQ_DATA]
|
| 71 |
|
| 72 |
-
# Step 3: Compute mathematical similarity matrices
|
| 73 |
vectorizer = TfidfVectorizer(stop_words='english')
|
| 74 |
tfidf_matrix = vectorizer.fit_transform(questions_corpus)
|
| 75 |
|
| 76 |
query_vector = vectorizer.transform([cleaned_query])
|
| 77 |
similarity_scores = cosine_similarity(query_vector, tfidf_matrix)[0]
|
| 78 |
|
| 79 |
-
# Find highest matching index score
|
| 80 |
best_match_idx = similarity_scores.argmax()
|
| 81 |
highest_score = similarity_scores[best_match_idx]
|
| 82 |
|
|
@@ -89,7 +84,8 @@ def get_thematic_datetime():
|
|
| 89 |
if 11 <= day_int <= 13:
|
| 90 |
suffix = "th"
|
| 91 |
else:
|
| 92 |
-
|
|
|
|
| 93 |
|
| 94 |
formatted_date = f"{now.strftime('%d')}{suffix} {now.strftime('%B %Y')} | Time: {now.strftime('%I:%M:%S %p').lower()}"
|
| 95 |
return f"<div style='text-align: right; font-weight: 600; font-size: 14px; font-family: monospace; letter-spacing: 0.5px;'>{formatted_date}</div>"
|
|
@@ -132,14 +128,9 @@ def bot_side(history, system_message, max_tokens, temperature, top_p):
|
|
| 132 |
yield history
|
| 133 |
return
|
| 134 |
|
| 135 |
-
# Extract the fresh user input message text
|
| 136 |
user_query = extract_text(history[-1]["content"])
|
| 137 |
-
|
| 138 |
-
# Run our NLP Cosine Matcher over our collected database
|
| 139 |
matched_faq, match_score = perform_nlp_cosine_matching(user_query)
|
| 140 |
|
| 141 |
-
# Format a visible dynamic tag noting the similarity result inside the UI panel
|
| 142 |
-
nlp_metadata_prefix = ""
|
| 143 |
faq_context_injection = ""
|
| 144 |
|
| 145 |
if matched_faq and match_score > 0.15:
|
|
@@ -165,11 +156,9 @@ def bot_side(history, system_message, max_tokens, temperature, top_p):
|
|
| 165 |
messages.append({"role": role, "content": content})
|
| 166 |
clean_history.append({"role": role, "content": content})
|
| 167 |
|
| 168 |
-
# Append user question to system message array
|
| 169 |
messages.append({"role": "user", "content": user_query})
|
| 170 |
clean_history.append({"role": "user", "content": user_query})
|
| 171 |
|
| 172 |
-
# Initialize assistant output streaming sequence
|
| 173 |
clean_history.append({"role": "assistant", "content": nlp_metadata_prefix})
|
| 174 |
|
| 175 |
try:
|
|
|
|
| 52 |
|
| 53 |
def perform_nlp_cosine_matching(user_query):
|
| 54 |
"""
|
| 55 |
+
Preprocesses text using tokenization and applies Cosine Similarity
|
| 56 |
+
matching over TF-IDF vectors to find the best matching FAQ.
|
| 57 |
"""
|
|
|
|
| 58 |
try:
|
| 59 |
tokens = nltk.word_tokenize(user_query.lower())
|
| 60 |
cleaned_query = " ".join([w for w in tokens if re.match(r'^\w+$', w)])
|
| 61 |
except Exception:
|
|
|
|
| 62 |
cleaned_query = " ".join(re.findall(r'\w+', user_query.lower()))
|
| 63 |
|
| 64 |
if not cleaned_query.strip():
|
| 65 |
return None, 0.0
|
| 66 |
|
|
|
|
| 67 |
questions_corpus = [faq["q"] for faq in FAQ_DATA]
|
| 68 |
|
|
|
|
| 69 |
vectorizer = TfidfVectorizer(stop_words='english')
|
| 70 |
tfidf_matrix = vectorizer.fit_transform(questions_corpus)
|
| 71 |
|
| 72 |
query_vector = vectorizer.transform([cleaned_query])
|
| 73 |
similarity_scores = cosine_similarity(query_vector, tfidf_matrix)[0]
|
| 74 |
|
|
|
|
| 75 |
best_match_idx = similarity_scores.argmax()
|
| 76 |
highest_score = similarity_scores[best_match_idx]
|
| 77 |
|
|
|
|
| 84 |
if 11 <= day_int <= 13:
|
| 85 |
suffix = "th"
|
| 86 |
else:
|
| 87 |
+
# FIXED: Changed ._get to .get
|
| 88 |
+
suffix = {1: "st", 2: "nd", 3: "rd"}.get(day_int % 10, "th")
|
| 89 |
|
| 90 |
formatted_date = f"{now.strftime('%d')}{suffix} {now.strftime('%B %Y')} | Time: {now.strftime('%I:%M:%S %p').lower()}"
|
| 91 |
return f"<div style='text-align: right; font-weight: 600; font-size: 14px; font-family: monospace; letter-spacing: 0.5px;'>{formatted_date}</div>"
|
|
|
|
| 128 |
yield history
|
| 129 |
return
|
| 130 |
|
|
|
|
| 131 |
user_query = extract_text(history[-1]["content"])
|
|
|
|
|
|
|
| 132 |
matched_faq, match_score = perform_nlp_cosine_matching(user_query)
|
| 133 |
|
|
|
|
|
|
|
| 134 |
faq_context_injection = ""
|
| 135 |
|
| 136 |
if matched_faq and match_score > 0.15:
|
|
|
|
| 156 |
messages.append({"role": role, "content": content})
|
| 157 |
clean_history.append({"role": role, "content": content})
|
| 158 |
|
|
|
|
| 159 |
messages.append({"role": "user", "content": user_query})
|
| 160 |
clean_history.append({"role": "user", "content": user_query})
|
| 161 |
|
|
|
|
| 162 |
clean_history.append({"role": "assistant", "content": nlp_metadata_prefix})
|
| 163 |
|
| 164 |
try:
|