A Multi-Label Classification Method Based On Textual Data Mining

Researches on Big data has been a heated discussion because the rapid development of Internet brings a large amount of valuable data. In this paper, we proposed an effective method to mining the textual data. To be more specific, a multi-label classification method based on textual data mining was designed. The design of this method is built on deep learning. It first adopts Bret pre-training model to express the text, and then be processed with text mining module. This module employs the self-attention mechanism to get the sentence embedding of the text. Meanwhile, convolution is adopted to gain the local semantic information. Then, keyless attention fusion, an effective method used in multi-modal fusion is utilized to aggerate the above two parts. Besides, the value of textual data for reviews about commodities engrossed our attention, as it can heavily help the buying decision of individuals. Thus, our research of textual data focused on the reviews of commodities. Each data has multiple attributes, and that is the reason we chose Multi-label classification in Natural Language Processing as the direction of our model design. We conducted an in-depth discussion of the combination of multi-label classification and the features of those textual data. Finally, we compare our model with other baseline models with evaluation metrics of precision@K, recall@K, F1_score@K and Ndcg@k. Our model gains the best performance in those metrics. With the help of our model, customers can gain precise and objective information about the goods and their purchase decision can be more sensible.