An Effective Model based on CNN - LSTM for E-Commerce Product Title Classification

The number of products posted by sellers on each e-commerce system is often very large, up to millions of products and can be many millions of times more. Therefore, product title quality control is especially important for each system. An efficient model based on the two most popular deep learning models today, CNN (Convolutional Neural Network) and LSTM (Long Short-Term Memory) for the product title classification problem is presented in the paper. Determines whether a product title is conciseness or not. The model will use many feature groups as input, then they will be extracted again through CNN and LSTM. The combination of features generated from CNN and LSTM will form the most efficient representation vector for each title. In the experiment, we use the Vietnamese product title data set. The results have shown the superior efficiency of the proposed model compared to the base models, such as SVM, Random Forrest, and XGBoost.