It is a common case that there are huge news reports with different topics in a website (even in a food-related website), food safety news reports are needed to be selected for further analysis, and this is a text (or document) classification problem. In this paper, we propose a food safety document classification method using LSTM (long and short-term memory)-based ensemble learning. Firstly, due to the high cost of human-annotation, the food safety document corpus includes only one-class samples, and the food safety document classification based on such a corpus is a one-class classification problem. We propose an automatic corpus expansion approach which uses a large number of unlabeled news reported online as negative samples (the documents that are not related with food safety), and our food safety document corpus becomes a binary-class-based corpus that has both positive samples and negative samples. Secondly, our automatic corpus expansion brings the following two problems to document corpus: data noise and data unbalance. We choose an ensemble learning method which is based on LSTM(Long Short-Term Memory) for our document classification. Overall, the document classification method based on the LSTM-based ensemble learning method can automatically detect food safety documents from websites with outstanding performances.