互联网产品结构化信息抽取技术

With the development of e-commerce and vertical search engine, structured information extraction for online products becomes a research focus in data mining, information retrieval and natural language processing. This paper, as an example in named product entity recognition in clothing, we analyzed the difficulties and characteristics of information extraction, adopted an active learning method based on conditional random fields, and we implemented the process of product entity extraction from Chinese webpages. With the use of data from a group buying e-commerce platform for the experiment, this method obtained the overall F1 value 0.80, including brand, color, material F1 value, 0.70, 0.89 and 0.72, respectively. The experiment shows that this method has achieved higher performance; meanwhile, it has effectively reduced the cost of data labeling.