Update app.py

app.py

@@ -1,19 +1,16 @@
 import gradio as gr
 import time
 from datetime import datetime
-import pandas as pd
 from sentence_transformers import SentenceTransformer
 from qdrant_client import QdrantClient
 from qdrant_client.models import Filter, FieldCondition, MatchValue
 import os
-from rapidfuzz import process, fuzz
+from rapidfuzz import fuzz
 from pythainlp.tokenize import word_tokenize
 from pyairtable import Table
 from pyairtable import Api
-import pickle
 import re
 import unicodedata
-from FlagEmbedding import FlagReranker
 
 # Setup Qdrant Client
 qdrant_client = QdrantClient(
@@ -34,12 +31,10 @@ model = SentenceTransformer("BAAI/bge-m3")
 collection_name = "product_bge-m3"
 threshold = 0.45
 
-reranker = FlagReranker('BAAI/bge-reranker-v2-m3', use_fp16=True)
-
 # Utils
 def is_non_thai(text):
     return re.match(r'^[A-Za-z0-9&\-\s]+$', text) is not None
-    
+
 def normalize(text: str) -> str:
     if is_non_thai(text):
         return text.strip()
@@ -72,16 +67,26 @@ def search_product(query):
         return
 
     if len(result) > 0:
-        topk = 10
-        docs = [r.payload.get("name", "") for r in result[:topk]]
-        pairs = [[corrected_query, d] for d in docs]
-        scores = reranker.compute_score(pairs, normalize=True)
-        result[:topk] = sorted(
-            zip(result[:topk], scores),
-            key=lambda x: 0.6 * x[0].score + 0.4 * x[1],
-            reverse=True
-        )
-        result[:topk] = [r[0] for r in result[:topk]]
+      topk = 50  # ดึงมา rerank แค่ 50 อันดับแรกจาก Qdrant
+      result = result[:topk]
+
+      scored = []
+      for r in result:
+          name = r.payload.get("name", "")
+          
+          # ถ้า query สั้นเกินไป ให้ fuzzy_score = 0 เพื่อกันเพี้ยน
+          if len(corrected_query) >= 3 and name:
+              fuzzy_score = fuzz.partial_ratio(corrected_query, name) / 100.0
+          else:
+              fuzzy_score = 0.0
+
+          # รวม hybrid score
+          hybrid_score = 0.6 * r.score + 0.4 * fuzzy_score
+          scored.append((r, hybrid_score))
+
+      # เรียงตาม hybrid score แล้วกรองผลลัพธ์ที่ hybrid score ต่ำเกิน
+      scored = sorted(scored, key=lambda x: x[1], reverse=True)
+      result = [r[0] for r in scored]
 
     elapsed = time.time() - start_time
     html_output = f"<p>⏱ <strong>{elapsed:.2f} วินาที</strong></p>"