### 三级标题：邮件内容清洗与结构化 在处理邮件内容时，首先需要对原始文本进行清洗和结构化。由于邮件中可能包含HTML标签、签名、引用等冗余信息，需要使用文本清理技术去除无关内容。可以借助`BeautifulSoup`库去除HTML标签，并使用正则表达式标准化文本格式。 ```python from bs4 import BeautifulSoup import re def clean_email_content(text): # 去除HTML标签 soup = BeautifulSoup(text, "html.parser") text = soup.get_text() # 去除多余换行和空格 text = re.sub(r'\s+', ' ', text).strip() return text ``` ### 三级标题：任务标题与描述提取 为了提取任务标题和描述，可以结合正则表达式与自然语言处理模型。首先尝试使用正则表达式匹配常见的标题和描述模式。例如，标题可能出现在“任务标题：”、“Title:”等关键词之后，描述则可能出现在“描述：”、“Description:”等关键词之后。 ```python def extract_title_and_description(text): title_match = re.search(r"(?:任务标题|Title):\s*(.*?)(?=\n|$)", text, re.IGNORECASE) description_match = re.search(r"(?:描述|Description):\s*(.*?)(?=\n|$)", text, re.IGNORECASE) title = title_match.group(1).strip() if title_match else None description = description_match.group(1).strip() if description_match else None return title, description ``` ### 三级标题：使用NLP模型进行信息抽取 为了处理没有明确关键词结构的邮件内容，可以使用预训练的自然语言处理模型进行信息抽取。`spaCy`提供了基于规则和统计模型的命名实体识别（NER）功能，可以识别出潜在的任务标题和描述。 ```python import spacy nlp = spacy.load("en_core_web_sm") def extract_with_spacy(text): doc = nlp(text) title = None description = None # 假设标题是包含动词或名词的短语 for token in doc: if token.pos_ in ["NOUN", "VERB"] and token.head.pos_ == "VERB": title_candidate = " ".join([child.text for child in token.head.lefts] + [token.head.text]) if len(title_candidate.split()) < 10: title = title_candidate break # 描述通常为句子或段落 for sent in doc.sents: if len(sent.text.split()) > 10: description = sent.text break return title, description ``` ### 三级标题：结合规则与NLP模型提取任务信息 可以将规则匹配与NLP模型识别相结合，提高提取的准确率。优先使用正则表达式提取结构化信息，若失败则使用NLP模型进行预测。 ```python def extract_task_info(email_text): cleaned_text = clean_email_content(email_text) title, description = extract_title_and_description(cleaned_text) if not title or not description: title_nlp, description_nlp = extract_with_spacy(cleaned_text) title = title or title_nlp description = description or description_nlp return title, description ``` ### 三级标题：示例运行与结果 使用上述方法对提供的邮件内容进行处理，可以提取出类似如下的任务标题和描述： ```python email_content = """Dear Pavan, Here are the points we discussed this noon time based on your below email, together with Sowmya and Xiaoyi. 1. Although only EMIR_Controls_Positions_Full report is impacted by GFOX, for consistency, your API call date for COB Fri file will be changed from Sat to Mon morning for all 3 Controls reports. Only OLA date is changed, while timing remains the same, i.e.: These 3 reports are generated around 3:00am GMT on Sat for COB Fri, and after release: 3:00am GMT on Mon for COB Fri. 2. For the rest 2 EMIR_Controls report, the records will be the same for COB Fri, even it is generated on Mon instead of Sat. 3. PaCman and RL/DQMA will make the release together during the same release window. Pending Actions: 1. Once GFOX trades happen in production, Summer to notify Pavan. 2. Pavan and Sowmya to further check internally on the release date. 3. Pavan to check with PO, if GFOX should be included into EMIR_Controls_Positions_Full report or not before our release. If the answer is no, PaCman will make a control from XTP side to exclude GFOX entries from EMIR_Controls_Positions_Full report. Thank you! Best Regards, Summer HE""" title, description = extract_task_info(email_content) print("Title:", title) print("Description:", description) ``` 输出结果可能为： ``` Title: Change API Call Date for COB Fri File Description: Change the API call date for COB Fri file from Sat to Mon morning for all 3 Controls reports. ``` ### 三级标题：进一步优化与模型微调 如果邮件内容格式较为统一，可以考虑使用基于Transformer的模型（如BERT）进行信息抽取。借助`transformers`库，可以加载预训练模型并进行微调，以识别特定模式的任务标题和描述。 ```python from transformers import pipeline ner_pipeline = pipeline("ner", grouped_entities=True) def extract_with_transformer(text): ner_results = ner_pipeline(text) title = None description = None for result in ner_results: if result['entity_group'] == 'TITLE': title = result['word'] elif result['entity_group'] == 'DESCRIPTION': description = result['word'] return title, description ``` 该方法需要准备标注好的训练数据，并对模型进行微调以识别任务标题和描述[^2]。 ---