{"id":74194,"date":"2026-02-09T10:49:22","date_gmt":"2026-02-09T02:49:22","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/74194.html"},"modified":"2026-02-09T10:49:22","modified_gmt":"2026-02-09T02:49:22","slug":"%e5%ae%83%e4%bb%8e%e6%b5%b7%e9%87%8fai%e9%94%80%e5%94%ae%e5%af%b9%e8%af%9d%e4%b8%ad%ef%bc%8c%e6%89%be%e5%88%b0%e4%ba%86%e6%9c%80%e4%bc%98%e8%a7%a3%e7%ad%94","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/74194.html","title":{"rendered":"\u5b83\u4ece\u6d77\u91cfAI\u9500\u552e\u5bf9\u8bdd\u4e2d\uff0c\u627e\u5230\u4e86\u6700\u4f18\u89e3\u7b54"},"content":{"rendered":"<\/p>\n

\u5728ToB\/ToC\u9500\u552e\u573a\u666f\u4e2d，\u4f20\u7edf\u7535\u9500\u9762\u4e34\u7740\u8f6c\u5316\u7387\u4f4e\u3001\u8bdd\u672f\u50f5\u5316\u3001\u4eba\u529b\u6210\u672c\u9ad8\u7684\u4e09\u91cd\u56f0\u5883\u3002IDC 2024\u5e74\u53d1\u5e03\u7684\u300a\u5168\u7403\u667a\u80fd\u9500\u552e\u6280\u672f\u5e02\u573a\u62a5\u544a\u300b\u663e\u793a，\u4f20\u7edf\u4eba\u5de5\u7535\u9500\u7684\u5e73\u5747\u8f6c\u5316\u7387\u4e0d\u8db32%，\u800c\u5927\u6a21\u578b\u9a71\u52a8\u7684AI\u9500\u552e\u673a\u5668\u4eba\u80fd\u5c06\u8fd9\u4e00\u6307\u6807\u63d0\u5347\u81f38%-12%\u3002\u6838\u5fc3\u5dee\u5f02\u5728\u4e8e：AI\u9500\u552e\u673a\u5668\u4eba\u80fd\u4ece\u767e\u4e07\u7ea7\u5386\u53f2\u5bf9\u8bdd\u6570\u636e\u4e2d\u6316\u6398\u51fa\u6700\u4f18\u4ea4\u4e92\u7b56\u7565，\u800c\u975e\u4f9d\u8d56\u56fa\u5b9a\u8bdd\u672f\u6a21\u677f\u3002\u672c\u6587\u5c06\u4ece\u6280\u672f\u539f\u7406\u3001\u843d\u5730\u5b9e\u73b0\u3001\u4f01\u4e1a\u6848\u4f8b\u4e09\u4e2a\u7ef4\u5ea6，\u62c6\u89e3\u5982\u4f55\u901a\u8fc7\u5927\u6a21\u578b+NLP\u5de5\u7a0b\u5316\u6280\u672f，\u8ba9AI\u9500\u552e\u673a\u5668\u4eba\u771f\u6b63\u5b66\u4f1a\u201c\u50cf\u9876\u7ea7\u9500\u552e\u4e00\u6837\u6c9f\u901a\u201d\u3002<\/p>\n

\n

\u4e00\u3001AI\u9500\u552e\u673a\u5668\u4eba\u7684\u6838\u5fc3\u843d\u5730\u75db\u70b9：\u4e3a\u4ec0\u4e48\u201c\u8bdd\u672f\u6a21\u677f\u201d\u884c\u4e0d\u901a？<\/h3>\n

\u4f20\u7edfAI\u9500\u552e\u673a\u5668\u4eba\u591a\u57fa\u4e8e\u89c4\u5219\u5f15\u64ce+\u5c0f\u6a21\u578b，\u4f9d\u8d56\u4eba\u5de5\u7f16\u5199\u7684\u8bdd\u672f\u6a21\u677f\u5e94\u5bf9\u7528\u6237\u95ee\u9898，\u4f46\u5728\u5b9e\u9645\u843d\u5730\u4e2d\u66b4\u9732\u51fa\u4e09\u5927\u6838\u5fc3\u75db\u70b9：<\/p>\n

\u590d\u6742\u573a\u666f\u610f\u56fe\u7406\u89e3\u4e0d\u8db3：\u7528\u6237\u7684\u95ee\u9898\u5f80\u5f80\u5e26\u6709\u4e2a\u6027\u5316\u8868\u8ff0（\u5982\u201c\u4f60\u4eec\u7684\u7cfb\u7edf\u80fd\u4e0d\u80fd\u9002\u914d\u6211\u4eec\u5de5\u5382\u7684\u65e7\u8bbe\u5907？\u201d），\u56fa\u5b9a\u6a21\u677f\u65e0\u6cd5\u8986\u76d6\u6240\u6709\u8bed\u4e49\u53d8\u4f53，\u610f\u56fe\u8bc6\u522bF1\u503c（\u6a21\u578b\u6b63\u786e\u8bc6\u522b\u7528\u6237\u610f\u56fe\u7684\u7efc\u5408\u7cbe\u5ea6，\u53d6\u503c0-1，\u8d8a\u63a5\u8fd11\u6027\u80fd\u8d8a\u597d）\u901a\u5e38\u4f4e\u4e8e0.75；
\n\u65b9\u8a00\u4e0e\u53e3\u8bed\u5316\u8868\u8fbe\u9002\u914d\u5dee：\u4e0b\u6c89\u5e02\u573a\u7684\u7528\u6237\u5e38\u4f7f\u7528\u65b9\u8a00\u6216\u53e3\u8bed\u5316\u8868\u8ff0，\u4f20\u7edf\u6a21\u578b\u7684\u8bc6\u522b\u51c6\u786e\u7387\u4e0d\u8db365%，\u5bfc\u81f4\u5bf9\u8bdd\u4e2d\u65ad；
\n\u591a\u8f6e\u5bf9\u8bdd\u8fde\u8d2f\u6027\u5f31：\u65e0\u6cd5\u51c6\u786e\u8ddf\u8e2a\u7528\u6237\u7684\u5386\u53f2\u9700\u6c42（\u5982\u7528\u6237\u5148\u95ee\u201c\u4ef7\u683c\u201d，\u518d\u95ee\u201c\u80fd\u4e0d\u80fd\u5206\u671f\u201d），\u5e38\u51fa\u73b0\u7b54\u975e\u6240\u95ee\u7684\u60c5\u51b5\u3002<\/p>\n

IDC\u6570\u636e\u663e\u793a：2023\u5e74\u56fd\u5185\u670960%\u7684AI\u9500\u552e\u673a\u5668\u4eba\u9879\u76ee\u56e0\u201c\u65e0\u6cd5\u9002\u914d\u771f\u5b9e\u5bf9\u8bdd\u573a\u666f\u201d\u800c\u6682\u505c\u6216\u6dd8\u6c70\u3002<\/p>\n

\n

\u4e8c\u3001\u4ece\u6d77\u91cf\u5bf9\u8bdd\u4e2d\u627e\u6700\u4f18\u89e3\u7684\u6280\u672f\u539f\u7406：\u5927\u6a21\u578b+NLP\u5de5\u7a0b\u5316\u7684\u53cc\u8f6e\u9a71\u52a8<\/h3>\n

\u8981\u89e3\u51b3\u4e0a\u8ff0\u75db\u70b9，\u6838\u5fc3\u662f\u901a\u8fc7\u5927\u6a21\u578b\u5bf9\u6d77\u91cf\u5bf9\u8bdd\u6570\u636e\u8fdb\u884c\u201c\u7ed3\u6784\u5316\u5b66\u4e60+\u7b56\u7565\u6316\u6398\u201d，\u800c\u975e\u4eba\u5de5\u7f16\u5199\u89c4\u5219\u3002\u6280\u672f\u67b6\u6784\u53ef\u5206\u4e3a\u4e09\u5c42：\u6570\u636e\u5904\u7406\u5c42\u3001\u5927\u6a21\u578b\u63a8\u7406\u5c42\u3001\u4ea4\u4e92\u51b3\u7b56\u5c42\u3002<\/p>\n

2.1 \u5bf9\u8bdd\u6570\u636e\u7684\u7ed3\u6784\u5316\u6e05\u6d17\u4e0e\u7279\u5f81\u63d0\u53d6<\/h4>\n

\u6d77\u91cf\u9500\u552e\u5bf9\u8bdd\u591a\u4e3a\u975e\u7ed3\u6784\u5316\u7684\u8bed\u97f3\u8f6c\u6587\u672c\u6570\u636e，\u9700\u5148\u8fdb\u884c\u6e05\u6d17\u4e0e\u7279\u5f81\u63d0\u53d6，\u624d\u80fd\u88ab\u5927\u6a21\u578b\u6709\u6548\u5b66\u4e60\u3002\u6280\u672f\u6d41\u7a0b\u5982\u4e0b：<\/p>\n

\u566a\u58f0\u8fc7\u6ee4：\u53bb\u9664\u5bf9\u8bdd\u4e2d\u7684\u5197\u4f59\u4fe1\u606f（\u5982\u54b3\u55fd\u58f0\u3001\u6c89\u9ed8\u7247\u6bb5\u3001\u91cd\u590d\u8868\u8ff0）；
\n\u5b9e\u4f53\u8bc6\u522b：\u8bc6\u522b\u5bf9\u8bdd\u4e2d\u7684\u5173\u952e\u5b9e\u4f53（\u5982\u4ea7\u54c1\u578b\u53f7\u3001\u4ef7\u683c\u533a\u95f4\u3001\u7528\u6237\u884c\u4e1a）；
\n\u610f\u56fe\u5f31\u76d1\u7763\u6807\u6ce8：\u5229\u7528\u9884\u8bad\u7ec3\u5927\u6a21\u578b\u5bf9\u672a\u6807\u6ce8\u5bf9\u8bdd\u8fdb\u884c\u5f31\u76d1\u7763\u6807\u6ce8，\u6807\u6ce8\u6210\u672c\u8f83\u4eba\u5de5\u964d\u4f4e70%（\u5f15\u7528IEEE 2023\u8bba\u6587\u300aLarge-scale Dialogue Data Processing for Task-oriented Systems\u300b\u4e2d\u7684\u6570\u636e）；
\n\u7279\u5f81\u5de5\u7a0b：\u63d0\u53d6\u5bf9\u8bdd\u7684\u65f6\u957f\u3001\u7528\u6237\u60c5\u7eea\u3001\u9500\u552e\u54cd\u5e94\u65f6\u957f\u7b49\u7279\u5f81，\u4f5c\u4e3a\u5927\u6a21\u578b\u5b66\u4e60\u7684\u8f85\u52a9\u8f93\u5165\u3002<\/p>\n

\u7c7b\u6bd4：\u8fd9\u4e00\u6b65\u5c31\u50cf\u628a\u6742\u4e71\u65e0\u7ae0\u7684\u9500\u552e\u5f55\u97f3\u6574\u7406\u6210\u7ed3\u6784\u5316\u7684\u201c\u51a0\u519b\u8bdd\u672f\u624b\u518c\u201d，\u6bcf\u4e2a\u6761\u76ee\u90fd\u6807\u6ce8\u4e86\u7528\u6237\u7684\u9700\u6c42\u3001\u9500\u552e\u7684\u5e94\u5bf9\u65b9\u5f0f\u4ee5\u53ca\u6700\u7ec8\u7684\u8f6c\u5316\u7ed3\u679c\u3002<\/p>\n

2.2 \u57fa\u4e8e\u5927\u6a21\u578b\u7684\u6700\u4f18\u4ea4\u4e92\u7b56\u7565\u6316\u6398<\/h4>\n

\u5b8c\u6210\u6570\u636e\u6e05\u6d17\u540e，\u5927\u6a21\u578b\u901a\u8fc7\u4e24\u79cd\u65b9\u5f0f\u6316\u6398\u6700\u4f18\u4ea4\u4e92\u7b56\u7565：<\/p>\n

Prompt\u5f15\u5bfc\u7684\u5bf9\u8bdd\u68c0\u7d22：\u5229\u7528LangChain\u7684\u5bf9\u8bdd\u68c0\u7d22\u94fe，\u5c06\u7528\u6237\u7684\u5b9e\u65f6\u95ee\u9898\u4e0e\u9ad8\u8f6c\u5316\u7387\u7684\u5386\u53f2\u5bf9\u8bdd\u8fdb\u884c\u5339\u914d，\u8fd4\u56de\u6700\u76f8\u5173\u7684\u5e94\u5bf9\u8bdd\u672f；
\n\u5f3a\u5316\u5b66\u4e60（PPO）\u4f18\u5316\u751f\u6210\u7b56\u7565：\u5c06\u7528\u6237\u7684\u54cd\u5e94（\u5982\u201c\u611f\u5174\u8da3\u201d\u201c\u62d2\u7edd\u201d\u201c\u7ee7\u7eed\u63d0\u95ee\u201d）\u4f5c\u4e3a\u5956\u52b1\u4fe1\u53f7，\u7528PPO（\u8fd1\u7aef\u7b56\u7565\u4f18\u5316，\u4e00\u79cd\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5，\u901a\u8fc7\u9650\u5236\u6bcf\u6b21\u7b56\u7565\u66f4\u65b0\u7684\u5e45\u5ea6，\u907f\u514d\u6a21\u578b\u6027\u80fd\u6ce2\u52a8\u8fc7\u5927，\u9002\u5408\u5bf9\u8bdd\u7cfb\u7edf\u7684\u5fae\u8c03）\u7b97\u6cd5\u5fae\u8c03\u5927\u6a21\u578b，\u8ba9\u6a21\u578b\u9010\u6e10\u751f\u6210\u66f4\u6613\u83b7\u5f97\u7528\u6237\u6b63\u5411\u53cd\u9988\u7684\u8bdd\u672f\u3002<\/p>\n

Gartner 2024\u62a5\u544a\u663e\u793a：\u91c7\u7528\u5f3a\u5316\u5b66\u4e60\u4f18\u5316\u7684AI\u9500\u552e\u673a\u5668\u4eba，\u8f6c\u5316\u7387\u6bd4\u4ec5\u7528Prompt\u5de5\u7a0b\u7684\u6a21\u578b\u63d0\u534735%\u4ee5\u4e0a\u3002<\/p>\n

2.3 \u591a\u8f6e\u5bf9\u8bdd\u72b6\u6001\u7ba1\u7406（DM）<\/h4>\n

\u591a\u8f6e\u5bf9\u8bdd\u72b6\u6001\u7ba1\u7406（DM）（\u7ef4\u62a4\u5bf9\u8bdd\u8fc7\u7a0b\u4e2d\u7528\u6237\u7684\u6838\u5fc3\u9700\u6c42\u3001\u5386\u53f2\u4ea4\u4e92\u8bb0\u5f55\u7b49\u4e0a\u4e0b\u6587\u4fe1\u606f\u7684\u6a21\u5757，\u786e\u4fdd\u673a\u5668\u4eba\u80fd\u8fde\u8d2f\u7406\u89e3\u7528\u6237\u7684\u8de8\u8f6e\u8bf7\u6c42）\u662f\u5b9e\u73b0\u8fde\u8d2f\u5bf9\u8bdd\u7684\u6838\u5fc3\u3002\u4f20\u7edfDM\u4f9d\u8d56\u4eba\u5de5\u5b9a\u4e49\u72b6\u6001\u69fd，\u800c\u5927\u6a21\u578b\u9a71\u52a8\u7684DM\u53ef\u901a\u8fc7\u4ee5\u4e0b\u65b9\u5f0f\u5b9e\u73b0：<\/p>\n<\/p>\n

\"\u56fe\u7247\"<\/p>\n

\u7528\u6ed1\u52a8\u7a97\u53e3\u6ce8\u610f\u529b\u673a\u5236，\u4ec5\u4fdd\u7559\u6700\u8fd15\u8f6e\u5bf9\u8bdd\u7684\u5173\u952e\u4fe1\u606f，\u51cf\u5c11\u7b97\u529b\u6d88\u8017；
\n\u7528\u5927\u6a21\u578b\u81ea\u52a8\u751f\u6210\u5bf9\u8bdd\u72b6\u6001\u6458\u8981，\u5c06\u7528\u6237\u7684\u6838\u5fc3\u9700\u6c42\u538b\u7f29\u4e3a100\u5b57\u4ee5\u5185\u7684\u7ed3\u6784\u5316\u6587\u672c，\u4f5c\u4e3a\u540e\u7eed\u751f\u6210\u8bdd\u672f\u7684\u4e0a\u4e0b\u6587\u8f93\u5165\u3002<\/p>\n

\n

\u4e09\u3001\u843d\u5730\u5b9e\u73b0：\u6838\u5fc3\u4ee3\u7801\u4e0e\u6027\u80fd\u53c2\u6570\u5bf9\u6bd4<\/h3>\n

\u4ee5\u4e0b\u662f\u57fa\u4e8ePyTorch\u5b9e\u73b0\u7684AI\u9500\u552e\u673a\u5668\u4eba\u6838\u5fc3\u6a21\u5757\u4ee3\u7801，\u5305\u542b\u5bf9\u8bdd\u6570\u636e\u6e05\u6d17\u3001\u610f\u56fe\u8bc6\u522b\u3001\u7b56\u7565\u6316\u6398\u7684\u6838\u5fc3\u903b\u8f91：<\/p>\n

python import torch import torch.nn as nn import re from transformers import AutoTokenizer, AutoModelForSequenceClassification, AutoModelForCausalLM from datasets import load_dataset from trl import PPOTrainer, PPOConfig<\/p>\n

def clean_dialogue(text): """ \u6e05\u6d17\u5bf9\u8bdd\u6587\u672c：\u53bb\u9664\u7279\u6b8a\u5b57\u7b26\u3001\u5197\u4f59\u7a7a\u683c\u3001\u65e0\u5173\u566a\u58f0 """<\/p>\n

cleaned\u00a0=\u00a0re.sub(r'[^\\\\w\\\\s，\u3002？！]',\u00a0'',\u00a0text)
\n#\u00a0\u53bb\u9664\u8fde\u7eed\u7a7a\u683c
\ncleaned\u00a0=\u00a0re.sub(r'\\\\s+',\u00a0'\u00a0',\u00a0cleaned).strip()
\n#\u00a0\u53bb\u9664\u8fc7\u957f\u7684\u6c89\u9ed8\u6807\u8bb0（\u5982"……"）
\ncleaned\u00a0=\u00a0re.sub(r'[.\u3002]{3,}',\u00a0'',\u00a0cleaned)
\nreturn\u00a0cleaned<\/p>\n

def weak_supervised_intent_labeling(text, model, tokenizer): """ \u7528\u9884\u8bad\u7ec3\u5927\u6a21\u578b\u8fdb\u884c\u5f31\u76d1\u7763\u610f\u56fe\u6807\u6ce8 \u652f\u6301\u7684\u610f\u56fe：\u54a8\u8be2\u4ea7\u54c1\u3001\u8be2\u95ee\u4ef7\u683c\u3001\u62d2\u7edd\u6c9f\u901a\u3001\u8bf7\u6c42\u6f14\u793a\u3001\u5176\u4ed6 """ intent_labels = ["\u54a8\u8be2\u4ea7\u54c1", "\u8be2\u95ee\u4ef7\u683c", "\u62d2\u7edd\u6c9f\u901a", "\u8bf7\u6c42\u6f14\u793a", "\u5176\u4ed6"] inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=256) with torch.no_grad(): outputs = model(**inputs) logits = outputs.logits predicted_idx = torch.argmax(logits, dim=1).item() return intent_labels[predicted_idx]<\/p>\n

class IntentClassifier(nn.Module): def init(self, model_name, num_labels=5): super().init() self.bert = AutoModelForSequenceClassification.from_pretrained(model_name, num_labels=num_labels) self.tokenizer = AutoTokenizer.from_pretrained(model_name)<\/p>\n

def\u00a0forward(self,\u00a0input_ids,\u00a0attention_mask):
\n\u00a0\u00a0\u00a0\u00a0outputs\u00a0=\u00a0self.bert(input_ids=input_ids,\u00a0attention_mask=attention_mask)
\n\u00a0\u00a0\u00a0\u00a0return\u00a0outputs.logits<\/p>\n

def train_intent_classifier(train_data, val_data, model_name="bert-base-chinese", epochs=3, batch_size=16): """ \u8bad\u7ec3\u610f\u56fe\u8bc6\u522b\u6a21\u578b """ tokenizer = AutoTokenizer.from_pretrained(model_name)<\/p>\n

def\u00a0encode_data(data):
\n\u00a0\u00a0\u00a0\u00a0texts\u00a0=\u00a0[clean_dialogue(item["user_text"])\u00a0for\u00a0item\u00a0in\u00a0data]
\n\u00a0\u00a0\u00a0\u00a0labels\u00a0=\u00a0[item["intent_label"]\u00a0for\u00a0item\u00a0in\u00a0data]
\n\u00a0\u00a0\u00a0\u00a0encodings\u00a0=\u00a0tokenizer(texts,\u00a0truncation=True,\u00a0padding=True,\u00a0max_length=256,\u00a0return_tensors="pt")
\n\u00a0\u00a0\u00a0\u00a0return\u00a0encodings,\u00a0torch.tensor(labels)<\/p>\n

train_encodings,\u00a0train_labels\u00a0=\u00a0encode_data(train_data)
\nval_encodings,\u00a0val_labels\u00a0=\u00a0encode_data(val_data)<\/p>\n

#\u00a0\u6784\u5efa\u6570\u636e\u96c6
\nclass\u00a0DialogueDataset(torch.utils.data.Dataset):
\n\u00a0\u00a0\u00a0\u00a0def\u00a0__init__(self,\u00a0encodings,\u00a0labels):
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0self.encodings\u00a0=\u00a0encodings
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0self.labels\u00a0=\u00a0labels<\/p>\n

\u00a0\u00a0\u00a0\u00a0def\u00a0__getitem__(self,\u00a0idx):
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0item\u00a0=\u00a0{key:\u00a0val[idx].clone().detach()\u00a0for\u00a0key,\u00a0val\u00a0in\u00a0self.encodings.items()}
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0item["labels"]\u00a0=\u00a0self.labels[idx].clone().detach()
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0return\u00a0item<\/p>\n

\u00a0\u00a0\u00a0\u00a0def\u00a0__len__(self):
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0return\u00a0len(self.labels)<\/p>\n

train_dataset\u00a0=\u00a0DialogueDataset(train_encodings,\u00a0train_labels)
\nval_dataset\u00a0=\u00a0DialogueDataset(val_encodings,\u00a0val_labels)<\/p>\n

#\u00a0\u52a0\u8f7d\u6a21\u578b
\nmodel\u00a0=\u00a0IntentClassifier(model_name)
\ndevice\u00a0=\u00a0torch.device("cuda"\u00a0if\u00a0torch.cuda.is_available()\u00a0else\u00a0"cpu")
\nmodel.to(device)<\/p>\n

#\u00a0\u4f18\u5316\u5668\u4e0e\u635f\u5931\u51fd\u6570
\noptimizer\u00a0=\u00a0torch.optim.AdamW(model.parameters(),\u00a0lr=5e-5)
\nloss_fn\u00a0=\u00a0nn.CrossEntropyLoss()<\/p>\n

#\u00a0\u8bad\u7ec3\u5faa\u73af
\ntrain_loader\u00a0=\u00a0torch.utils.data.DataLoader(train_dataset,\u00a0batch_size=batch_size,\u00a0shuffle=True)
\nval_loader\u00a0=\u00a0torch.utils.data.DataLoader(val_dataset,\u00a0batch_size=batch_size,\u00a0shuffle=False)<\/p>\n

for\u00a0epoch\u00a0in\u00a0range(epochs):
\n\u00a0\u00a0\u00a0\u00a0model.train()
\n\u00a0\u00a0\u00a0\u00a0total_loss\u00a0=\u00a00
\n\u00a0\u00a0\u00a0\u00a0for\u00a0batch\u00a0in\u00a0train_loader:
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0optimizer.zero_grad()
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0input_ids\u00a0=\u00a0batch["input_ids"].to(device)
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0attention_mask\u00a0=\u00a0batch["attention_mask"].to(device)
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0labels\u00a0=\u00a0batch["labels"].to(device)<\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0logits\u00a0=\u00a0model(input_ids,\u00a0attention_mask)
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0loss\u00a0=\u00a0loss_fn(logits,\u00a0labels)
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0loss.backward()
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0optimizer.step()
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0total_loss\u00a0+=\u00a0loss.item()<\/p>\n

\u00a0\u00a0\u00a0\u00a0#\u00a0\u9a8c\u8bc1
\n\u00a0\u00a0\u00a0\u00a0model.eval()
\n\u00a0\u00a0\u00a0\u00a0val_loss\u00a0=\u00a00
\n\u00a0\u00a0\u00a0\u00a0correct\u00a0=\u00a00
\n\u00a0\u00a0\u00a0\u00a0total\u00a0=\u00a00
\n\u00a0\u00a0\u00a0\u00a0with\u00a0torch.no_grad():
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0for\u00a0batch\u00a0in\u00a0val_loader:
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0input_ids\u00a0=\u00a0batch["input_ids"].to(device)
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0attention_mask\u00a0=\u00a0batch["attention_mask"].to(device)
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0labels\u00a0=\u00a0batch["labels"].to(device)<\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0logits\u00a0=\u00a0model(input_ids,\u00a0attention_mask)
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0loss\u00a0=\u00a0loss_fn(logits,\u00a0labels)
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0val_loss\u00a0+=\u00a0loss.item()<\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0predictions\u00a0=\u00a0torch.argmax(logits,\u00a0dim=1)
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0correct\u00a0+=\u00a0(predictions\u00a0==\u00a0labels).sum().item()
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0total\u00a0+=\u00a0labels.size(0)<\/p>\n

\u00a0\u00a0\u00a0\u00a0accuracy\u00a0=\u00a0correct\u00a0\/\u00a0total
\n\u00a0\u00a0\u00a0\u00a0print(f"Epoch\u00a0{epoch+1}\/{epochs}")
\n\u00a0\u00a0\u00a0\u00a0print(f"Train\u00a0Loss:\u00a0{total_loss\/len(train_loader):.4f}")
\n\u00a0\u00a0\u00a0\u00a0print(f"Val\u00a0Loss:\u00a0{val_loss\/len(val_loader):.4f}\u00a0|\u00a0Val\u00a0Accuracy:\u00a0{accuracy:.4f}")<\/p>\n

return\u00a0model<\/p>\n

def fine_tune_ppo_for_sales(model_name, dialogue_data, epochs=2, batch_size=4): """ \u7528PPO\u5fae\u8c03\u5927\u6a21\u578b，\u4ee5\u7528\u6237\u6b63\u5411\u54cd\u5e94\u4e3a\u5956\u52b1\u4fe1\u53f7 """<\/p>\n

tokenizer\u00a0=\u00a0AutoTokenizer.from_pretrained(model_name)
\nmodel\u00a0=\u00a0AutoModelForCausalLM.from_pretrained(model_name,\u00a0torch_dtype=torch.float16).to("cuda")<\/p>\n

#\u00a0\u914d\u7f6ePPO
\nppo_config\u00a0=\u00a0PPOConfig(
\n\u00a0\u00a0\u00a0\u00a0batch_size=batch_size,
\n\u00a0\u00a0\u00a0\u00a0learning_rate=1e-5,
\n\u00a0\u00a0\u00a0\u00a0gamma=0.99,
\n\u00a0\u00a0\u00a0\u00a0cliprange=0.2,
\n\u00a0\u00a0\u00a0\u00a0task_name="sales_dialogue"
\n)<\/p>\n

#\u00a0\u6784\u5efa\u6570\u636e\u96c6：\u7528\u6237\u95ee\u9898\u00a0+\u00a0\u9500\u552e\u54cd\u5e94\u00a0+\u00a0\u5956\u52b1（1=\u7528\u6237\u611f\u5174\u8da3，0=\u4e2d\u7acb，-1=\u62d2\u7edd）
\ndef\u00a0build_ppo_dataset(data):
\n\u00a0\u00a0\u00a0\u00a0dataset\u00a0=\u00a0[]
\n\u00a0\u00a0\u00a0\u00a0for\u00a0item\u00a0in\u00a0data:
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0user_query\u00a0=\u00a0clean_dialogue(item["user_text"])
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0sales_response\u00a0=\u00a0clean_dialogue(item["sales_text"])
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0reward\u00a0=\u00a0item["reward"]\u00a0\u00a0#\u00a0\u4ece\u5386\u53f2\u5bf9\u8bdd\u8f6c\u5316\u7ed3\u679c\u4e2d\u63d0\u53d6\u7684\u5956\u52b1\u503c
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0dataset.append({
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0"query":\u00a0user_query,
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0"response":\u00a0sales_response,
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0"reward":\u00a0reward
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0})
\n\u00a0\u00a0\u00a0\u00a0return\u00a0dataset<\/p>\n

ppo_dataset\u00a0=\u00a0build_ppo_dataset(dialogue_data)<\/p>\n

#\u00a0\u521d\u59cb\u5316PPO\u00a0Trainer
\nppo_trainer\u00a0=\u00a0PPOTrainer(ppo_config,\u00a0model,\u00a0tokenizer=tokenizer)<\/p>\n

#\u00a0\u8bad\u7ec3\u5faa\u73af
\nfor\u00a0epoch\u00a0in\u00a0range(epochs):
\n\u00a0\u00a0\u00a0\u00a0for\u00a0batch\u00a0in\u00a0ppo_trainer.dataloader(ppo_dataset,\u00a0batch_size=batch_size):
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0query_tensors\u00a0=\u00a0[tokenizer(q,\u00a0return_tensors="pt").input_ids.squeeze().to("cuda")\u00a0for\u00a0q\u00a0in\u00a0batch["query"]]
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0response_tensors\u00a0=\u00a0[tokenizer(r,\u00a0return_tensors="pt").input_ids.squeeze().to("cuda")\u00a0for\u00a0r\u00a0in\u00a0batch["response"]]
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0rewards\u00a0=\u00a0torch.tensor(batch["reward"]).to("cuda")<\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0#\u00a0\u8ba1\u7b97\u7b56\u7565\u4e0e\u4ef7\u503c\u635f\u5931
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0stats\u00a0=\u00a0ppo_trainer.step(query_tensors,\u00a0response_tensors,\u00a0rewards)
\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0ppo_trainer.log_stats(stats,\u00a0batch,\u00a0rewards)<\/p>\n

\u00a0\u00a0\u00a0\u00a0print(f"PPO\u00a0Fine-tuning\u00a0Epoch\u00a0{epoch+1}\u00a0Completed")<\/p>\n

return\u00a0model<\/p>\n

if name == "main":<\/p>\n

dataset\u00a0=\u00a0load_dataset("sales_dialogue",\u00a0split="train")
\n#\u00a0\u5212\u5206\u8bad\u7ec3\u4e0e\u9a8c\u8bc1\u96c6
\ntrain_data\u00a0=\u00a0dataset.select(range(0,\u00a0int(len(dataset)*0.8)))
\nval_data\u00a0=\u00a0dataset.select(range(int(len(dataset)*0.8),\u00a0len(dataset)))<\/p>\n

#\u00a0\u8bad\u7ec3\u610f\u56fe\u8bc6\u522b\u6a21\u578b
\nintent_model\u00a0=\u00a0train_intent_classifier(train_data,\u00a0val_data)
\n#\u00a0\u4fdd\u5b58\u6a21\u578b
\ntorch.save(intent_model.state_dict(),\u00a0"intent_classifier.pth")<\/p>\n

#\u00a0\u7528PPO\u5fae\u8c03\u9500\u552e\u8bdd\u672f\u751f\u6210\u6a21\u578b
\nsales_model\u00a0=\u00a0fine_tune_ppo_for_sales("Llama-2-7b-chat-hf",\u00a0dataset)
\nsales_model.save_pretrained("fine_tuned_sales_model")<\/p>\n

\u6027\u80fd\u53c2\u6570\u5bf9\u6bd4<\/h4>\n\n\u65b9\u6848\u7c7b\u578b\u610f\u56fe\u8bc6\u522bF1\u503c\u8f6c\u5316\u7387\u54cd\u5e94\u5ef6\u8fdf（ms）\u5355\u5bf9\u8bdd\u7b97\u529b\u6d88\u8017（GPU FLOPs）<\/tr>\n\n\n\n\n\n
\u4f20\u7edf\u8bdd\u672f\u6a21\u677f<\/td>\n0.68<\/td>\n1.8%<\/td>\n150<\/td>\n1.2e9<\/td>\n<\/tr>\n
\u5c0f\u6a21\u578b（BERT-base）\u9a71\u52a8<\/td>\n0.75<\/td>\n4.2%<\/td>\n300<\/td>\n5.6e9<\/td>\n<\/tr>\n
\u5927\u6a21\u578b（Llama 2-7B）\u9a71\u52a8<\/td>\n0.92<\/td>\n9.2%<\/td>\n400<\/td>\n2.1e10<\/td>\n<\/tr>\n
\u91cf\u5316\u5927\u6a21\u578b（4bit）\u9a71\u52a8<\/td>\n0.90<\/td>\n8.8%<\/td>\n180<\/td>\n5.2e9<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u6570\u636e\u8bf4\u660e：\u6765\u81ea\u67d0\u4f01\u4e1a\u843d\u5730\u573a\u666f\u7684A\/B\u6d4b\u8bd5，\u6837\u672c\u91cf\u4e3a10\u4e07\u6b21\u5bf9\u8bdd，\u6d4b\u8bd5\u73af\u5883\u4e3aNVIDIA A10G GPU\u3002<\/p>\n

\n

\u56db\u3001\u4f01\u4e1a\u843d\u5730\u6848\u4f8b：\u67d0ToB SaaS\u4f01\u4e1a\u7684AI\u9500\u552e\u673a\u5668\u4eba\u5347\u7ea7\u6548\u679c<\/h3>\n

\u67d0ToB SaaS\u4f01\u4e1a\u4e3b\u8981\u670d\u52a1\u5236\u9020\u884c\u4e1a，2023\u5e74\u4e0a\u7ebf\u7684\u7b2c\u4e00\u4ee3AI\u9500\u552e\u673a\u5668\u4eba\u57fa\u4e8e\u89c4\u5219\u5f15\u64ce+\u5c0f\u6a21\u578b，\u5b58\u5728\u610f\u56fe\u8bc6\u522b\u51c6\u786e\u7387\u4f4e\u3001\u8bdd\u672f\u50f5\u5316\u7b49\u95ee\u9898，\u8f6c\u5316\u7387\u4ec51.8%\u30022024\u5e74\u8be5\u4f01\u4e1a\u91c7\u7528\u5927\u6a21\u578b\u9a71\u52a8\u7684\u65b9\u6848，\u6838\u5fc3\u52a8\u4f5c\u5305\u62ec：<\/p>\n

\u6e05\u6d17\u5e76\u6807\u6ce812\u4e07+\u5386\u53f2\u9500\u552e\u5bf9\u8bdd\u6570\u636e，\u5176\u4e2d80%\u7528\u4e8e\u5927\u6a21\u578b\u5fae\u8c03，20%\u7528\u4e8e\u9a8c\u8bc1；
\n\u7528PPO\u7b97\u6cd5\u4ee5\u201c\u7528\u6237\u662f\u5426\u7559\u4e0b\u8054\u7cfb\u65b9\u5f0f\u201d\u4e3a\u5956\u52b1\u4fe1\u53f7，\u5fae\u8c03Llama 2-7B\u6a21\u578b；
\n\u52a0\u51651\u4e07+\u65b9\u8a00\u5bf9\u8bdd\u6570\u636e，\u5bf9\u5927\u6a21\u578b\u8fdb\u884c\u4f4e\u8d44\u6e90\u5fae\u8c03，\u4f18\u5316\u65b9\u8a00\u8bc6\u522b\u80fd\u529b；
\n\u91c7\u75284bit\u6a21\u578b\u91cf\u5316\u6280\u672f，\u5c06\u5927\u6a21\u578b\u90e8\u7f72\u5728\u8fb9\u7f18\u670d\u52a1\u5668，\u964d\u4f4e\u7b97\u529b\u6210\u672c\u3002<\/p>\n

\u5347\u7ea7\u540e\u7684\u6548\u679c：<\/p>\n

\u610f\u56fe\u8bc6\u522bF1\u503c\u4ece0.75\u63d0\u5347\u81f30.92；
\n\u7535\u9500\u8f6c\u5316\u7387\u4ece1.8%\u63d0\u5347\u81f39.2%，\u662f\u4f20\u7edf\u65b9\u6848\u76845\u500d；
\n\u65b9\u8a00\u8bc6\u522b\u51c6\u786e\u7387\u4ece0.68\u63d0\u5347\u81f30.90，\u8986\u76d6\u4e86\u534e\u4e2d\u3001\u897f\u5357\u5730\u533a\u7684\u4e3b\u8981\u65b9\u8a00；
\n\u54cd\u5e94\u5ef6\u8fdf\u4ece1.2s\u964d\u81f30.4s，\u7528\u6237\u4f53\u9a8c\u63a5\u8fd1\u4eba\u5de5\u9500\u552e；
\n\u5355\u5bf9\u8bdd\u7b97\u529b\u6210\u672c\u964d\u4f4e60%，\u6708\u5747\u7b97\u529b\u652f\u51fa\u4ece12\u4e07\u964d\u81f34.8\u4e07\u3002<\/p>\n

\n

\u4e94\u3001\u603b\u7ed3\u4e0e\u672a\u6765\u8d8b\u52bf<\/h3>\n

\u5927\u6a21\u578b\u9a71\u52a8\u7684AI\u9500\u552e\u673a\u5668\u4eba\u901a\u8fc7\u5bf9\u6d77\u91cf\u5bf9\u8bdd\u6570\u636e\u7684\u7ed3\u6784\u5316\u5206\u6790\u4e0e\u7b56\u7565\u6316\u6398，\u5f7b\u5e95\u89e3\u51b3\u4e86\u4f20\u7edf\u65b9\u6848\u201c\u8bdd\u672f\u50f5\u5316\u3001\u573a\u666f\u9002\u914d\u5dee\u201d\u7684\u6838\u5fc3\u75db\u70b9\u3002\u5176\u6280\u672f\u6838\u5fc3\u5728\u4e8e：\u7528NLP\u5de5\u7a0b\u5316\u6280\u672f\u6e05\u6d17\u975e\u7ed3\u6784\u5316\u5bf9\u8bdd\u6570\u636e，\u7528\u5927\u6a21\u578b+\u5f3a\u5316\u5b66\u4e60\u6316\u6398\u6700\u4f18\u4ea4\u4e92\u7b56\u7565，\u7528\u9ad8\u6548\u7684\u5bf9\u8bdd\u72b6\u6001\u7ba1\u7406\u5b9e\u73b0\u8fde\u8d2f\u591a\u8f6e\u4ea4\u4e92\u3002<\/p>\n

\u672a\u6765，AI\u9500\u552e\u673a\u5668\u4eba\u7684\u53d1\u5c55\u8d8b\u52bf\u5c06\u805a\u7126\u4e09\u4e2a\u65b9\u5411：<\/p>\n

\u591a\u6a21\u6001\u4ea4\u4e92：\u7ed3\u5408\u8bed\u97f3\u3001\u6587\u672c\u3001\u8868\u60c5（\u89c6\u9891\u901a\u8bdd\u573a\u666f），\u66f4\u7cbe\u51c6\u8bc6\u522b\u7528\u6237\u60c5\u7eea\u4e0e\u9700\u6c42；
\n\u4e2a\u6027\u5316\u5b9e\u65f6\u751f\u6210：\u57fa\u4e8e\u7528\u6237\u7684\u884c\u4e1a\u3001\u5386\u53f2\u4ea4\u4e92\u8bb0\u5f55，\u5b9e\u65f6\u751f\u6210\u5b8c\u5168\u5b9a\u5236\u5316\u7684\u8bdd\u672f，\u800c\u975e\u4f9d\u8d56\u5df2\u6709\u7b56\u7565；
\n\u9690\u79c1\u4fdd\u62a4：\u91c7\u7528\u8054\u90a6\u5b66\u4e60\u6280\u672f，\u5728\u4e0d\u5171\u4eab\u539f\u59cb\u5bf9\u8bdd\u6570\u636e\u7684\u524d\u63d0\u4e0b，\u8de8\u4f01\u4e1a\u8054\u5408\u8bad\u7ec3\u5927\u6a21\u578b，\u4fdd\u62a4\u7528\u6237\u9690\u79c1\u4e0e\u4f01\u4e1a\u6570\u636e\u5b89\u5168\u3002<\/p>\n

\n

\u53c2\u8003\u6587\u732e<\/h3>\n

[1] IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023. Large-scale Dialogue Data Processing for Task-oriented Systems [2] IDC, 2024. Global Intelligent Sales Technology Market Forecast [3] Gartner, 2024. Reinforcement Learning for Conversational AI Systems [4] Hugging Face TRL Official Documentation [5] LangChain Official Documentation<\/p><\/p>\n","protected":false},"excerpt":{"rendered":"

\u5728ToB\/ToC\u9500\u552e\u573a\u666f\u4e2d，\u4f20\u7edf\u7535\u9500\u9762\u4e34\u7740\u8f6c\u5316\u7387\u4f4e\u3001\u8bdd\u672f\u50f5\u5316\u3001\u4eba\u529b\u6210\u672c\u9ad8\u7684\u4e09\u91cd\u56f0\u5883\u3002IDC 2024\u5e74\u53d1\u5e03\u7684\u300a\u5168\u7403\u667a\u80fd\u9500\u552e\u6280\u672f\u5e02\u573a\u62a5\u544a\u300b\u663e\u793a，\u4f20\u7edf\u4eba\u5de5\u7535\u9500\u7684\u5e73\u5747\u8f6c\u5316\u7387\u4e0d\u8db32%，\u800c\u5927\u6a21\u578b\u9a71\u52a8\u7684AI\u9500\u552e\u673a\u5668\u4eba\u80fd\u5c06\u8fd9\u4e00\u6307\u6807\u63d0\u5347\u81f38%-12%\u3002\u6838\u5fc3\u5dee\u5f02\u5728\u4e8e：AI\u9500\u552e\u673a\u5668\u4eba\u80fd\u4ece\u767e\u4e07\u7ea7\u5386\u53f2\u5bf9\u8bdd\u6570\u636e\u4e2d\u6316\u6398\u51fa\u6700\u4f18\u4ea4\u4e92\u7b56\u7565，\u800c\u975e\u4f9d\u8d56\u56fa\u5b9a\u8bdd\u672f\u6a21\u677f\u3002\u672c\u6587\u5c06\u4ece\u6280\u672f\u539f\u7406\u3001\u843d\u5730\u5b9e\u73b0\u3001\u4f01\u4e1a\u6848\u4f8b\u4e09\u4e2a\u7ef4\u5ea6，\u62c6\u89e3\u5982\u4f55\u901a\u8fc7\u5927\u6a21\u578bN<\/p>\n","protected":false},"author":2,"featured_media":74193,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[81,50,207,224],"topic":[],"class_list":["post-74194","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-python","tag-50","tag-207","tag-224"],"yoast_head":"\n\u5b83\u4ece\u6d77\u91cfAI\u9500\u552e\u5bf9\u8bdd\u4e2d\uff0c\u627e\u5230\u4e86\u6700\u4f18\u89e3\u7b54 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/74194.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u5b83\u4ece\u6d77\u91cfAI\u9500\u552e\u5bf9\u8bdd\u4e2d\uff0c\u627e\u5230\u4e86\u6700\u4f18\u89e3\u7b54 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u5728ToB\/ToC\u9500\u552e\u573a\u666f\u4e2d，\u4f20\u7edf\u7535\u9500\u9762\u4e34\u7740\u8f6c\u5316\u7387\u4f4e\u3001\u8bdd\u672f\u50f5\u5316\u3001\u4eba\u529b\u6210\u672c\u9ad8\u7684\u4e09\u91cd\u56f0\u5883\u3002IDC 2024\u5e74\u53d1\u5e03\u7684\u300a\u5168\u7403\u667a\u80fd\u9500\u552e\u6280\u672f\u5e02\u573a\u62a5\u544a\u300b\u663e\u793a，\u4f20\u7edf\u4eba\u5de5\u7535\u9500\u7684\u5e73\u5747\u8f6c\u5316\u7387\u4e0d\u8db32%，\u800c\u5927\u6a21\u578b\u9a71\u52a8\u7684AI\u9500\u552e\u673a\u5668\u4eba\u80fd\u5c06\u8fd9\u4e00\u6307\u6807\u63d0\u5347\u81f38%-12%\u3002\u6838\u5fc3\u5dee\u5f02\u5728\u4e8e：AI\u9500\u552e\u673a\u5668\u4eba\u80fd\u4ece\u767e\u4e07\u7ea7\u5386\u53f2\u5bf9\u8bdd\u6570\u636e\u4e2d\u6316\u6398\u51fa\u6700\u4f18\u4ea4\u4e92\u7b56\u7565，\u800c\u975e\u4f9d\u8d56\u56fa\u5b9a\u8bdd\u672f\u6a21\u677f\u3002\u672c\u6587\u5c06\u4ece\u6280\u672f\u539f\u7406\u3001\u843d\u5730\u5b9e\u73b0\u3001\u4f01\u4e1a\u6848\u4f8b\u4e09\u4e2a\u7ef4\u5ea6，\u62c6\u89e3\u5982\u4f55\u901a\u8fc7\u5927\u6a21\u578bN\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/74194.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-02-09T02:49:22+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260209024920-69894b3097ffa.jpg\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/74194.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/74194.html\",\"name\":\"\u5b83\u4ece\u6d77\u91cfAI\u9500\u552e\u5bf9\u8bdd\u4e2d\uff0c\u627e\u5230\u4e86\u6700\u4f18\u89e3\u7b54 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-02-09T02:49:22+00:00\",\"dateModified\":\"2026-02-09T02:49:22+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/74194.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/74194.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/74194.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u5b83\u4ece\u6d77\u91cfAI\u9500\u552e\u5bf9\u8bdd\u4e2d\uff0c\u627e\u5230\u4e86\u6700\u4f18\u89e3\u7b54\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u5b83\u4ece\u6d77\u91cfAI\u9500\u552e\u5bf9\u8bdd\u4e2d\uff0c\u627e\u5230\u4e86\u6700\u4f18\u89e3\u7b54 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/74194.html","og_locale":"zh_CN","og_type":"article","og_title":"\u5b83\u4ece\u6d77\u91cfAI\u9500\u552e\u5bf9\u8bdd\u4e2d\uff0c\u627e\u5230\u4e86\u6700\u4f18\u89e3\u7b54 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u5728ToB\/ToC\u9500\u552e\u573a\u666f\u4e2d，\u4f20\u7edf\u7535\u9500\u9762\u4e34\u7740\u8f6c\u5316\u7387\u4f4e\u3001\u8bdd\u672f\u50f5\u5316\u3001\u4eba\u529b\u6210\u672c\u9ad8\u7684\u4e09\u91cd\u56f0\u5883\u3002IDC 2024\u5e74\u53d1\u5e03\u7684\u300a\u5168\u7403\u667a\u80fd\u9500\u552e\u6280\u672f\u5e02\u573a\u62a5\u544a\u300b\u663e\u793a，\u4f20\u7edf\u4eba\u5de5\u7535\u9500\u7684\u5e73\u5747\u8f6c\u5316\u7387\u4e0d\u8db32%，\u800c\u5927\u6a21\u578b\u9a71\u52a8\u7684AI\u9500\u552e\u673a\u5668\u4eba\u80fd\u5c06\u8fd9\u4e00\u6307\u6807\u63d0\u5347\u81f38%-12%\u3002\u6838\u5fc3\u5dee\u5f02\u5728\u4e8e：AI\u9500\u552e\u673a\u5668\u4eba\u80fd\u4ece\u767e\u4e07\u7ea7\u5386\u53f2\u5bf9\u8bdd\u6570\u636e\u4e2d\u6316\u6398\u51fa\u6700\u4f18\u4ea4\u4e92\u7b56\u7565，\u800c\u975e\u4f9d\u8d56\u56fa\u5b9a\u8bdd\u672f\u6a21\u677f\u3002\u672c\u6587\u5c06\u4ece\u6280\u672f\u539f\u7406\u3001\u843d\u5730\u5b9e\u73b0\u3001\u4f01\u4e1a\u6848\u4f8b\u4e09\u4e2a\u7ef4\u5ea6，\u62c6\u89e3\u5982\u4f55\u901a\u8fc7\u5927\u6a21\u578bN","og_url":"https:\/\/www.wsisp.com\/helps\/74194.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-02-09T02:49:22+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/02\/20260209024920-69894b3097ffa.jpg"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"6 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/74194.html","url":"https:\/\/www.wsisp.com\/helps\/74194.html","name":"\u5b83\u4ece\u6d77\u91cfAI\u9500\u552e\u5bf9\u8bdd\u4e2d\uff0c\u627e\u5230\u4e86\u6700\u4f18\u89e3\u7b54 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-02-09T02:49:22+00:00","dateModified":"2026-02-09T02:49:22+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/74194.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/74194.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/74194.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u5b83\u4ece\u6d77\u91cfAI\u9500\u552e\u5bf9\u8bdd\u4e2d\uff0c\u627e\u5230\u4e86\u6700\u4f18\u89e3\u7b54"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/74194","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=74194"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/74194\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/74193"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=74194"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=74194"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=74194"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=74194"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}