With the rapid development of e-commerce platforms, a large number of commodity comments have emerged on the Internet. Comments not only provide consumers with the experience of others, but also make it easier for companies to get feedback from users. This paper is motivated from this background to derive deep learning methods on effective aspect mining in comments. Classically, a product comment involves different aspects of a product, therefore aspect mining on product comments is supposed to be a multi-label classification. This paper proposes a bi-directional LSTM aspect mining algorithm combined with the Attention mechanism. In order to further improve the precision of classification, the CRF model is introduced at the final step, and the output of the bi-directional LSTM is optimized by the joint conditional probability of the words in the comments. The performances of this algorithm (BLCA) is tested by exploring the real comment data repository from the e-commerce platforms. The experimental results show the priority of BLCA over other algorithms as LSTM or RNN.