Commit bbd86955 authored by myhloli's avatar myhloli
Browse files

feat(post_proc): enhance title block processing with average line height

- Add average line height calculation for title blocks
- Include page number in title dictionary
- Improve title optimization prompt for better hierarchy- Implement retry mechanism for JSON decoding errors
- Add error logging for title count mismatch
parent 902dcd2c
......@@ -83,24 +83,38 @@ def llm_aided_title(pdf_info_dict, title_aided_config):
if block["type"] == "title":
origin_title_list.append(block)
title_text = merge_para_with_text(block)
title_dict[f"{i}"] = title_text
page_line_height_list = []
for line in block['lines']:
bbox = line['bbox']
page_line_height_list.append(int(bbox[3] - bbox[1]))
if len(page_line_height_list) > 0:
line_avg_height = sum(page_line_height_list) / len(page_line_height_list)
else:
line_avg_height = int(block['bbox'][3] - block['bbox'][1])
title_dict[f"{i}"] = [title_text, line_avg_height, int(page_num[5:])+1]
i += 1
# logger.info(f"Title list: {title_dict}")
title_optimize_prompt = f"""输入的内容是一篇文档中所有标题组成的字典,请根据以下指南优化标题的结果,使结果符合正常文档的层次结构:
1. 保留原始内容:
1. 字典中每个value均为一个list,包含以下元素:
- 标题文本
- 文本行高是标题所在块的平均行高
- 标题所在的页码
2. 保留原始内容:
- 输入的字典中所有元素都是有效的,不能删除字典中的任何元素
- 请务必保证输出的字典中元素的数量和输入的数量一致
2. 保持字典内key-value的对应关系不变
3. 保持字典内key-value的对应关系不变
3. 优化层次结构:
4. 优化层次结构:
- 为每个标题元素添加适当的层次结构
- 标题层级应具有连续性,不能跳过某一层级
- 行高较大的标题一般是更高级别的标题
- 标题从前至后的层级必须是连续的,不能跳过层级
- 标题层级最多为4级,不要添加过多的层级
- 优化后的标题为一个整数,代表该标题的层级
- 优化后的标题只保留代表该标题的层级的整数,不要保留其他信息
IMPORTANT:
请直接返回优化过的由标题层级组成的json,返回的json不需要格式化。
......@@ -110,24 +124,36 @@ Input title list:
Corrected title list:
"""
completion = client.chat.completions.create(
model=title_aided_config["model"],
messages=[
{'role': 'user', 'content': title_optimize_prompt}],
temperature=0.7,
)
json_completion = json.loads(completion.choices[0].message.content)
# logger.info(f"Title completion: {json_completion}")
retry_count = 0
max_retries = 3
json_completion = None
# logger.info(f"len(json_completion): {len(json_completion)}, len(title_dict): {len(title_dict)}")
if len(json_completion) == len(title_dict):
while retry_count < max_retries:
try:
for i, origin_title_block in enumerate(origin_title_list):
origin_title_block["level"] = int(json_completion[str(i)])
completion = client.chat.completions.create(
model=title_aided_config["model"],
messages=[
{'role': 'user', 'content': title_optimize_prompt}],
temperature=0.7,
)
json_completion = json.loads(completion.choices[0].message.content)
# logger.info(f"Title completion: {json_completion}")
# logger.info(f"len(json_completion): {len(json_completion)}, len(title_dict): {len(title_dict)}")
if len(json_completion) == len(title_dict):
for i, origin_title_block in enumerate(origin_title_list):
origin_title_block["level"] = int(json_completion[str(i)])
break
else:
logger.warning("The number of titles in the optimized result is not equal to the number of titles in the input.")
retry_count += 1
except Exception as e:
logger.exception(e)
else:
logger.error("The number of titles in the optimized result is not equal to the number of titles in the input.")
if e is json.JSONDecodeError:
logger.warning(f"JSON decode error on attempt {retry_count + 1}: {e}")
else:
logger.exception(e)
retry_count += 1
if json_completion is None:
logger.error("Failed to decode JSON after maximum retries.")
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment