Abstract: The automatic mining of billions of product specification information in Web has important application value in many fields such as e-commerce market analysis, commodity recommendation, after-sales service and so on. But the current methods of specification extraction don’t effectively solve the balance between manual annotation workload, scalability and accuracy. This paper proposes the Title Seed Automatic Extract（TSAE） method, using unsupervised learning method, using the page title as seed, combining with statistical characteristics, natural and machine semantics, it achieves higher accuracy while reducing the workload, enhancing the scalability. The experimental results show that the TSAE method has better automatic extraction precision while providing good performance and expansibility, can support the massive data processing, has good practical value.

Key words: information extraction, automatic extraction, product specification, e-commerce

摘要： Web中数十亿的商品规格信息的自动挖掘，对电子商务领域的市场分析、商品推荐、售后服务等诸多领域有重要的应用价值。但目前的商品规格信息抽取方法尚未有效解决人工标注工作量、扩展性和准确率之间的平衡问题，提出一种商品网页规格信息自动抽取方法TSAE（Title Seed Automatic Extract），采用无监督的学习方法，以网页标题为种子，结合统计特征、自然语义和机器语义，在减少工作量、提升扩展性的同时，达到了较高的准确率。实验表明，TSAE方法在提供更好的自动化抽取效果的同时，具备良好的性能和扩展性，能够支撑海量数据处理，具有良好的实用价值。

关键词: 信息抽取, 自动抽取, 商品规格信息, 电子商务