bug/parse chinese document ( tranditional chinese ) with attachment

govform02.pdf

result_from_official_unstructured.json

result_from_testing_server_installed_unstructured.json

1. first element , official give type : Title, testing server , give UncategorizedText
2. testing server , language , said : kor, it should be chinese , lang detect problem

additional info : about the testing server config

self.extraction_config = {
            'strategy'                      : 'hi_res',                 # 高精度策略
            'model_name'                    : "detectron2_onnx",        # 佈局檢測模型：yolox, detectron2_onnx
            
            'chunking_strategy'             : 'by_title',
            
            
            # 添加 OCR 支援（如果 PDF 是掃描件）
            'ocr_languages'                 : 'chi_tra',
            'languages'                     : ['chi_tra', 'chi_sim', 'eng'],
            
            # 提取圖片配置
            'extract_images_in_pdf'         : True,
            'extract_image_block_types'     : ["Image", "Table"],
            
            # 重要：添加這些參數來改善標題識別
            'pdf_infer_table_structure'     : True,
            'infer_table_structure'         : True,         # 推斷表格結構
            'include_page_breaks'           : False,
            'include_metadata'              : True,
            
            # 調整分塊參數
            'max_characters'                : 4000,
            'new_after_n_chars'             : 3800,
            'combine_text_under_n_chars'    : 500,
            
            #'pdf_image_dpi'                 : 300
            
            # 其它
            #'extract_image_block_to_payload'    : False,  # 不將圖片存入 payload（節省空間）
        }

** file source, from public online , google , just for testing

Start by comparing govform02.pdf with result_from_official_unstructured.json and result_from_testing_server_installed_unstructured.json, then review the supplied extraction_config, especially the hi_res strategy, detectron2_onnx model, OCR languages, and language list. Done means the testing-server output agrees on the first element's Title type and identifies the document as Chinese.