Refa: improve GraphRAG similarity sensitivity to numeric differences (#8479)

### What problem does this PR solve?

Improve GraphRAG similarity sensitivity to numeric differences. #8444.

### Type of change

- [x] Refactoring

graphrag/entity_resolution.py

@@ -218,7 +218,20 @@ class EntityResolution(Extractor):
 
         return ans_list
 
+    def _has_digit_in_2gram_diff(self, a, b):
+        def to_2gram_set(s):
+            return {s[i:i+2] for i in range(len(s) - 1)}
+
+        set_a = to_2gram_set(a)
+        set_b = to_2gram_set(b)
+        diff = set_a ^ set_b
+
+        return any(any(c.isdigit() for c in pair) for pair in diff)
+
     def is_similarity(self, a, b):
+        if self._has_digit_in_2gram_diff(a, b):
+            return False
+
         if is_english(a) and is_english(b):
             if editdistance.eval(a, b) <= min(len(a), len(b)) // 2:
                 return True

rag/nlp/__init__.py

@@ -225,15 +225,23 @@ def bullets_category(sections):
 
 
 def is_english(texts):
-    eng = 0
     if not texts:
         return False
-    for t in texts:
+
-        if re.match(r"[ `a-zA-Z.,':;/\"?<>!\(\)-]", t.strip()):
+    pattern = re.compile(r"[`a-zA-Z0-9\s.,':;/\"?<>!\(\)\-]")
-            eng += 1
+
-    if eng / len(texts) > 0.8:
+    if isinstance(texts, str):
-        return True
+        texts = list(texts)
-    return False
+    elif isinstance(texts, list):
+        texts = [t for t in texts if isinstance(t, str) and t.strip()]
+    else:
+        return False
+
+    if not texts:
+        return False
+
+    eng = sum(1 for t in texts if pattern.fullmatch(t.strip()))
+    return (eng / len(texts)) > 0.8
 
 
 def is_chinese(text):