Temporal Multiple-convolutional Network for Commodity Classification of Online Retail Platform Data
Changhong Zhong, Lifen Jiang, Yan Liang, Huazhi Sun, Chunmei Ma · 2020
With the development and popularization of online shopping, a large number of commodities on the online retail platform should be managed through the multilevel category system, in which there are thousands of categories. Automatic commodity classification has become an issue. Classifying commodities according to their text titles can improve the efficiency of online retail platforms. The text titles of commodities contain the information of commodities, but the text titles seldom follow the grammar rules and the text length varies greatly. We propose a classification model, Temporal Multiple-Convolutional Network (TMN), which combines Temporal Convolutional Network(TCN) model and Multiple-Convolutional Neural Network(MCNN) model. The TCN model is firstly used to model sequence data based on both word embedding and character embedding respectively, and generate two output vectors. Then, the MCNN models are used to extract features from the abstract vectors. Finally, output layer employs softmax function to classify commodities. We conduct experiment on the online retail platform dataset provided by Inspur Group Co., Ltd. The results show that the TMN has a commodity classification accuracy of 84.2%, which is superior to that of state-of-the-art models.