Update app.py

app.py

@@ -232,72 +232,41 @@ def get_new_passage_random_any(used_passages_set):
 
 
 # ======================================================
-#  Group1: 本文のみ抽出（書き換えなし）
+#  Group1: 本文のみ抽出（★LLMで、Group2と同等の基準）
 # ======================================================
 
-def extract_main_body(text: str) -> str:
-    if not text:
-        return ""
-
-    lines = text.splitlines()
-
-    drop_patterns = [
-        r"^\s*title\s*:\s*.*$",
-        r"^\s*author\s*:\s*.*$",
-        r"^\s*by\s+[A-Z].*$",
-        r"^\s*source\s*:\s*.*$",
-        r"^\s*copyright\s*.*$",
-        r".*©.*",
-        r".*Project\s+Gutenberg.*",
-        r".*GUTENBERG.*",
-        r"^\s*http[s]?://\S+.*$",
-        r"^\s*www\.\S+.*$",
-        r"^\s*\[.*\]\s*$",
-        r"^\s*\(\s*.*\s*\)\s*$",
-        r"^\s*end\s+of\s+.*$",
-        r"^\s*\*{3}.*\*{3}\s*$",
-    ]
-    drop_re = re.compile("|".join(f"(?:{p})" for p in drop_patterns), re.IGNORECASE)
-
-    kept = []
-    for ln in lines:
-        if drop_re.match(ln.strip()):
-            continue
-        kept.append(ln)
-
-    def is_title_like(s: str) -> bool:
-        t = s.strip()
-        if not t:
-            return True
-        if len(t) <= 80 and "." not in t and "!" not in t and "?" not in t:
-            if len(t.split()) <= 12:
-                return True
-        return False
-
-    i = 0
-    skipped = 0
-    while i < len(kept) and skipped < 3 and is_title_like(kept[i]):
-        i += 1
-        skipped += 1
-
-    body_lines = kept[i:]
-
-    while body_lines and body_lines[-1].strip() == "":
-        body_lines.pop()
-
-    out = []
-    blank = False
-    for ln in body_lines:
-        if ln.strip() == "":
-            if not blank:
-                out.append("")
-            blank = True
-        else:
-            out.append(ln.rstrip())
-            blank = False
+def extract_main_body_llm(text: str) -> str:
+    """
+    Group1用：書き換えはしない。
+    ただしGroup2と同等の基準で「本文以外を完全除外」させるため、LLMに本文抽出だけさせる。
+    """
+    prompt = f"""
+Extract ONLY the main body text from the following passage.
+
+Rules:
+- Completely EXCLUDE titles, headings, chapter labels, author names, source information,
+  footnotes, annotations, introductions, and any non-body content.
+- Preserve the original paragraph structure of the main text.
+- Insert exactly ONE blank line between paragraphs.
+- Do NOT create new section breaks, chapter divisions, or headings.
+- Output only the extracted main body text.
+- Do not include explanations, comments, or metadata.
+- Do not include [TEXT START] and [TEXT END] in the output.
+
+[TEXT START]
+{text}
+[TEXT END]
+""".strip()
+
+    with _rewrite_sem:
+        resp = client.chat.completions.create(
+            model="google/gemini-2.5-flash",
+            messages=[{"role": "user", "content": prompt}],
+            temperature=0.0,
+            max_tokens=5000
+        )
+    return resp.choices[0].message.content.strip()
 
-    result = "\n".join(out).strip()
-    return result if result else text.strip()
 
 # ======================================================
 #  Rewrite（同時実行制限付き） Group2で使用
@@ -373,6 +342,7 @@ Formatting Constraints
         )
     return resp.choices[0].message.content.strip()
 
+
 def split_pages(text, max_words=300):
     sentences = re.split(r'(?<=[.!?])\s+', text.strip())
     pages = []
@@ -397,6 +367,7 @@ def split_pages(text, max_words=300):
 
     return pages or [text]
 
+
 # ======================================================
 #  Start（session_stateでユーザー状態管理）
 #  ★Start後：入力はリロードまで固定（1回だけ）
@@ -487,7 +458,7 @@ def start_test(student_id, level_input, group_input, session_state):
         )
 
     if group == 1:
-        processed = extract_main_body(text)
+        processed = extract_main_body_llm(text)  # ★ここが変更点
         measured_fre = orig_lev  # ★要件：Group1はpassage_informationのflesch_scoreを記録
     else:
         processed = rewrite_level(text, level, orig_lev)
@@ -546,6 +517,7 @@ def start_test(student_id, level_input, group_input, session_state):
         lock_student, lock_group, lock_level, lock_start
     )
 
+
 # ======================================================
 #  Next / Prev / Finish（元コードのままの構造 + state参照）
 # ======================================================
@@ -614,6 +586,7 @@ def next_page(pages_json, current_page, total_pages, pid, orig_lev, session_stat
         session_state
     )
 
+
 def prev_page(pages_json, current_page, total_pages, pid, orig_lev, session_state):
     user_id = session_state.get("user_id")
     level = session_state.get("level")
@@ -671,6 +644,7 @@ def prev_page(pages_json, current_page, total_pages, pid, orig_lev, session_stat
         session_state
     )
 
+
 def finish_or_retire(pages_json, current_page, pid, orig_lev, action, session_state):
     user_id = session_state.get("user_id")
     level = session_state.get("level")
@@ -709,7 +683,7 @@ def finish_or_retire(pages_json, current_page, pid, orig_lev, action, session_st
         )
 
     if group == 1:
-        processed = extract_main_body(new_text)
+        processed = extract_main_body_llm(new_text)  # ★ここが変更点
         measured_fre = new_orig_lev
     else:
         processed = rewrite_level(new_text, level, new_orig_lev)
@@ -766,6 +740,7 @@ def finish_or_retire(pages_json, current_page, pid, orig_lev, action, session_st
         session_state
     )
 
+
 # ======================================================
 #  UI（タイトル表示を追加。それ以外は変更しない）
 #  ★追加：パスワード付きログCSVダウンロード
@@ -955,7 +930,7 @@ with gr.Blocks(css=custom_css) as demo:
 
     retire_btn = gr.Button("リタイア")
 
-    # ★変更：Start後に入力をロックするため、入力コンポーネントもoutputsに追加
+    # ★Start後に入力をロックするため、入力コンポーネントもoutputsに追加
     start_btn.click(
         fn=start_test,
         inputs=[student_id_input, level_input, group_input, session_state],
@@ -1045,4 +1020,4 @@ with gr.Blocks(css=custom_css) as demo:
     )
 
 demo.queue(max_size=64)
-demo.launch()
+demo.launch()