OCR+AI双引擎驱动：手把手教学构建智能财报分析系统_deepseek 合合信息

from openai import OpenAIclient = OpenAI(api_key=\"\", base_url=\"https://api.deepseek.com\")# Round 1messages = [{\"role\": \"user\", \"content\": \"9.11 and 9.8, which is greater?\"}]response = client.chat.completions.create( model=\"deepseek-reasoner\", messages=messages)reasoning_content = response.choices[0].message.reasoning_contentcontent = response.choices[0].message.content# Round 2messages.append({\'role\': \'assistant\', \'content\': content})messages.append({\'role\': \'user\', \'content\': \"How many Rs are there in the word \'strawberry\'?\"})response = client.chat.completions.create( model=\"deepseek-reasoner\", messages=messages)# ...

# 接收文件并保存file = request.files.get(\'file\')if not file or not file.filename.lower().endswith(\'.pdf\'): return jsonify(error=\"请选择正确的PDF文件\"), 400filename = secure_filename(file.filename)filepath = os.path.join(app.config[\'UPLOAD_FOLDER\'], filename)file.save(filepath)

# 调用合合信息TextIn接口进行PDF解析resp = textin_client.recognize_pdf2md(filepath, { \'page_start\': 0, \'page_count\': 1000, # 设置解析页数为1000页 \'table_flavor\': \'md\', \'parse_mode\': \'scan\', # 设置解析模式为scan模式 \'page_details\': 0, # 不包含页面细节 \'markdown_details\': 1, \'apply_document_tree\': 1, \'dpi\': 144 # 分辨率设置为144 dpi})financial_data = json.loads(resp.text)[\'result\'][\'markdown\']

# 使用DeepSeek进行AI分析analysis = deepseek_client.analyze_financials(financial_data)

# 结构化分析结果html_content = convert_markdown_to_html(analysis)

<div id=\"analysisResult\"> 
🚨 三大核心风险点
 
 
 
净利润增速放缓
 第三季度归母净利润同比下降3.84%（扣非后-7.84%），与营收增长（+19.48%）背离，显示成本压力（研发、销售费用激增）或市场竞争加剧。
 
  
 
 
💡 三大增长亮点
 
  
 
 
📈 估值建议
 
  

四、完整案例

1. 案例完整代码

① app.py

# -*- coding: utf-8 -*-# app.pyfrom flask import Flask, render_template, request, jsonifyimport osimport jsonfrom werkzeug.utils import secure_filenameimport requestsimport pandas as pdfrom openai import OpenAIimport loggingfrom datetime import datetimeimport markdownimport redef get_file_content(filePath): with open(filePath, \'rb\') as fp: return fp.read()class TextinOcr(object): def __init__(self, app_id, app_secret): self._app_id = app_id self._app_secret = app_secret self.host = \'https://api.textin.com\' def recognize_pdf2md(self, image_path, options, is_url=False): \"\"\" pdf to markdown :param options: request params :param image_path: string :param is_url: bool :return: response options = { \'pdf_pwd\': None, \'dpi\': 144, # 设置dpi为144 \'page_start\': 0, \'page_count\': 1000, # 设置解析的页数为1000页 \'apply_document_tree\': 0, \'markdown_details\': 1, \'page_details\': 0, # 不包含页面细节信息 \'table_flavor\': \'md\', \'get_image\': \'none\', \'parse_mode\': \'scan\', # 解析模式设为scan } \"\"\" url = self.host + \'/ai/service/v1/pdf_to_markdown\' headers = { \'x-ti-app-id\': self._app_id, \'x-ti-secret-code\': self._app_secret } if is_url: image = image_path headers[\'Content-Type\'] = \'text/plain\' else: image = get_file_content(image_path) headers[\'Content-Type\'] = \'application/octet-stream\' return requests.post(url, data=image, headers=headers, params=options)class FinancialAnalyst: def __init__(self, api_key): self.client = OpenAI( api_key=api_key, base_url=\"https://api.deepseek.com\", ) def analyze_financials(self, financial_data): \"\"\" 执行深度财务分析（使用官方SDK） \"\"\" prompt = f\"\"\"你是一位顶级证券分析师，请根据以下财务数据进行专业分析： {financial_data} 要求用中文输出： 1. 三个最重要的风险点（用🚨标记） 2. 三个最突出的增长亮点（用💡标记） 3. 估值建议（用📈标记） 格式要求：Markdown列表，每个分类最多3条\"\"\" try: response = self.client.chat.completions.create( model=\"deepseek-reasoner\", messages=[  {\"role\": \"system\", \"content\": \"你是拥有20年经验的证监会持牌财务分析师，擅长发现数据背后的商业逻辑\"},  {\"role\": \"user\", \"content\": prompt} ], temperature=0.2, max_tokens=2000 ) return response.choices[0].message.content except Exception as e: return f\"⚠️ 分析失败：{str(e)}\"app = Flask(__name__)app.config[\'UPLOAD_FOLDER\'] = \'./uploads\'app.config[\'MAX_CONTENT_LENGTH\'] = 100 * 1024 * 1024 # 16MB# 初始化API客户端textin_client = TextinOcr( app_id=\"971357873**********2bef0f5c\", app_secret=\"2d5b5b6267************dbdde26a0\")deepseek_client = FinancialAnalyst( api_key=\"sk-k3k2jeB**********************78XharBWnYAJHneI\")# 配置日志（在Flask应用初始化后添加）logging.basicConfig( level=logging.INFO, format=\'%(asctime)s - %(name)s - %(levelname)s - %(message)s\', handlers=[ logging.FileHandler(\'financial_analysis.log\'), logging.StreamHandler() ])logger = logging.getLogger(__name__)@app.route(\'/\')def index(): return render_template(\'index1.html\')@app.route(\'/analyze\', methods=[\'POST\'])def analyze(): start_time = datetime.now() filepath = None try: # ========== 1. 接收文件 ==========  logger.info(\"收到分析请求 | Headers: %s\", request.headers) file = request.files.get(\'file\') if not file: logger.error(\"未接收到文件\") return jsonify(error=\"请选择要上传的文件\"), 400  if not file.filename.lower().endswith(\'.pdf\'): logger.error(\"文件类型错误: %s\", file.content_type) return jsonify(error=\"仅支持PDF文件\"), 400 # ========== 2. 保存文件 ==========  filename = secure_filename(file.filename) filepath = os.path.join(app.config[\'UPLOAD_FOLDER\'], filename) file.save(filepath) logger.info(\"文件暂存成功 | 路径: %s | 大小: %dKB\",  filepath, os.path.getsize(filepath)//1024) # ========== 3. 解析财报 ==========  logger.info(\"开始解析财报...\") resp = textin_client.recognize_pdf2md(filepath, { \'page_start\': 0, \'page_count\': 1000, # 设置解析页数为1000页 \'table_flavor\': \'md\', \'parse_mode\': \'scan\', # 设置解析模式为scan模式 \'page_details\': 0, # 不包含页面细节 \'markdown_details\': 1, \'apply_document_tree\': 1, \'dpi\': 144 # 分辨率设置为144 dpi }) logger.info(\"request time: %s\", resp.elapsed.total_seconds()) financial_data = json.loads(resp.text)[\'result\'][\'markdown\'] # 记录解析后的财报数据 logger.info(\"原始解析数据: %s\", json.dumps(financial_data, indent=2, ensure_ascii=False)) # ========== 5. 生成分析报告 ==========  logger.info(\"开始AI分析...\") # 获取AI分析结果 analysis = deepseek_client.analyze_financials(financial_data) logger.info(\"分析完成 | 结果长度: %d字符\", len(analysis)) logger.info(\"分析完成 | 结果内容: %s\", analysis) # 结构化分析内容 html_content = convert_markdown_to_html(analysis) result_text = re.sub(r\'.*?\', \'\', html_content, flags=re.DOTALL).replace(\"```markdown\", \"
\") think_text = re.sub(r\'```markdown.*?```\', \'\', html_content, flags=re.DOTALL) logger.info(json.dumps(think_text, indent=2, ensure_ascii=False)) logger.info(json.dumps(result_text, indent=2, ensure_ascii=False)) # ========== 6. 响应结果 ==========  duration = (datetime.now() - start_time).total_seconds() logger.info(\"请求处理完成 | 耗时: %.2fs\", duration) return jsonify(html=result_text) except ValueError as ve: logger.error(\"业务逻辑错误: %s\", str(ve), exc_info=True) return jsonify(error=str(ve)), 400 except requests.exceptions.RequestException as re: logger.error(\"API请求异常: %s\", str(re), exc_info=True) return jsonify(error=\"后台服务暂不可用\"), 503 except Exception as e: logger.critical(\"未处理异常: %s\", str(e), exc_info=True) return jsonify(error=\"系统内部错误\"), 500 finally: # 清理临时文件 if filepath and os.path.exists(filepath): try: os.remove(filepath) logger.info(\"已清理临时文件: %s\", filepath) except Exception as e: logger.warning(\"文件清理失败: %s\", str(e))def convert_markdown_to_html(markdown_text): # 将 Markdown 转换为 HTML html_content = markdown.markdown(markdown_text) return html_contentif __name__ == \'__main__\': os.makedirs(app.config[\'UPLOAD_FOLDER\'], exist_ok=True) app.run(debug=True)

② index.html

<!DOCTYPE html><html><head> <title>智能财报分析系统</title> <link href=\"https://cdn.bootcdn.net/ajax/libs/twitter-bootstrap/5.3.0/css/bootstrap.min.css\" rel=\"stylesheet\"> <script src=\"https://cdn.bootcdn.net/ajax/libs/echarts/5.4.2/echarts.min.js\"></script></head><body><div class=\"container mt-5\"> <h2 class=\"mb-4\">上市公司财报智能分析</h2>  <div class=\"card mb-4\"> <div class=\"card-body\"> <input type=\"file\" id=\"pdfUpload\" accept=\".pdf\" class=\"form-control\"> <div id=\"loading\" class=\"mt-3\" style=\"display:none;\"> <div class=\"spinner-border text-primary\"></div> <span class=\"ms-2\">分析中...</span> </div> </div> </div>  <div id=\"resultPanel\" style=\"display:none;\" class=\"row\">  <div class=\"col-md-6\"> <div class=\"card mb-4\"> <div class=\"card-header\">AI分析结论</div> <div class=\"card-body\">  <div id=\"textAnalysis\" style=\"white-space: pre-line;\"></div> </div> </div> </div>  <div class=\"col-md-6\"> <div class=\"card\"> <div class=\"card-header\">财务报表数据</div> <div class=\"card-body\">  <div id=\"tablesArea\"></div> </div> </div> </div> </div></div><script>document.getElementById(\'pdfUpload\').addEventListener(\'change\', function(e) { const file = e.target.files[0]; const formData = new FormData(); formData.append(\'file\', file); // 显示加载状态 document.getElementById(\'loading\').style.display = \'block\'; document.getElementById(\'resultPanel\').style.display = \'none\'; fetch(\'/analyze\', { method: \'POST\', body: formData }) .then(response => response.json()) .then(data => { if(data.error) { alert(\'错误: \' + data.error); return; } // 渲染分析文本 document.getElementById(\'textAnalysis\').textContent = data.analysis // 渲染表格 renderTables(data.tables) document.getElementById(\'resultPanel\').style.display = \'block\'; }) .finally(() => { document.getElementById(\'loading\').style.display = \'none\'; });});function renderTables(tables) { const container = document.getElementById(\'tablesArea\') container.innerHTML = \'\' // 定义表格样式 const tableStyle = `style=\"width:100%; margin-bottom:2rem; border-collapse:collapse;\"` const thStyle = `style=\"padding:12px; background:#f8f9fa; border:1px solid #dee2e6;\"` const tdStyle = `style=\"padding:12px; border:1px solid #dee2e6; text-align:right;\"` // 遍历所有表格类型 const tableTypes = { \'income_statement\': \'利润表\', \'balance_sheet\': \'资产负债表\', \'cash_flow\': \'现金流量表\' } Object.entries(tableTypes).forEach(([key, title]) => { const data = tables[key] if(data.length === 0) return // 创建表格标题 const titleEl = document.createElement(\'h5\') titleEl.className = \'mt-4 mb-3 text-primary\' titleEl.textContent = title container.appendChild(titleEl) // 创建表格 const table = document.createElement(\'table\') table.className = \'financial-table\' table.setAttribute(\'style\', tableStyle) // 处理表头 const thead = document.createElement(\'thead\') const headerCells = data[0].split(\'|\').filter(c => c.trim()) thead.innerHTML = `  ${headerCells.map(c => `<th ${thStyle}>${c.trim()}`).join(\'\')}  ` table.appendChild(thead) // 处理表格体 const tbody = document.createElement(\'tbody\') data.slice(1).forEach(row => { const cells = row.split(\'|\').filter(c => c.trim()) if(cells.length === 0) return const tr = document.createElement(\'tr\') tr.innerHTML = cells.map(c => ` <td ${tdStyle}>${c.trim().replace(/(\\d)(?=(\\d{3})+(?!\\d))/g, \'$1,\')} `).join(\'\') tbody.appendChild(tr) }) table.appendChild(tbody) container.appendChild(table) container.appendChild(document.createElement(\'div\')).className = \'mb-4\' // 添加空行 })}</script><style>.financial-table { margin-bottom: 2rem;}.financial-table th { background-color: #f8f9fa; min-width: 120px;}.financial-table td { font-family: \'Courier New\', monospace;}</style></body></html>

2. 案例运行效果

我们从雪球上下载某软件公式的2024Q3财报，上传到我们的智能分析项目。

​

五、行业应用案例

1. 金融行业

金融行业对精确的财报分析有着强烈的需求。通过合合信息TextIn与DeepSeek的结合，金融机构能够自动化处理财报文档，并获取AI分析的结果，极大提高了分析效率。无论是用于投资决策还是企业评估，系统都能提供准确、及时的分析报告，帮助决策者抓住市场机会。

2. 制造业

对于制造业企业来说，财报中往往涉及大量的供应链数据、采购订单和合同等文档。通过合合信息TextIn的自动化文档解析与DeepSeek的财务分析，制造业能够迅速获取财报中的核心数据，识别财务风险，并优化资源配置，提高运营效率。

六、总结

合合信息的TextIn与DeepSeek R1的结合，提供了一种全面、自动化的财报分析解决方案。通过自动化文档解析和深度AI分析，企业不仅能够高效提取财务数据，还能精准识别财务中的潜在问题和业务机会。这一智能化的解决方案正在成为金融、制造等行业财报处理的核心工具，助力企业在复杂的市场环境中做出更加明智的决策。