# 通过政务文本语句构建低质量政务文本语句
import pandas as pd
import random
# 读取原始 CSV
df = pd.read_csv("D:\mcp_getting_started\data\政务文本语句.csv")
sentences = df["sentence"].tolist()
# 定义伪造规则
verbs = ["出台", "发布", "实施", "推动", "完成", "监督", "检查", "处理", "优化", "完善"]
def corrupt_sentence(sentence):
corruption_types = ['remove_verb', 'word_shuffle', 'fragment', 'repeat_part']
corruption = random.choice(corruption_types)
if corruption == 'remove_verb':
for verb in verbs:
if verb in sentence:
return sentence.replace(verb, '')
elif corruption == 'word_shuffle':
words = list(sentence)
random.shuffle(words)
return ''.join(words[:len(sentence)//2])
elif corruption == 'fragment':
if len(sentence) > 20:
return sentence[:random.randint(5, len(sentence)//2)]
else:
return sentence
elif corruption == 'repeat_part':
if len(sentence) > 15:
part = sentence[:random.randint(5, 10)]
return part + part + sentence[random.randint(len(part), len(sentence)):]
else:
return sentence
return sentence
# 生成低质量句子
low_quality_sentences = [corrupt_sentence(s) for s in sentences]
# 保存为 CSV
df["low_quality_sentence"] = low_quality_sentences
df.to_csv("政务主谓宾句子_低质量版本_1000条.csv", index=False, encoding="utf-8-sig")
