Extraction of Product Evaluation Factors with a Convolutional Neural Network and Transfer Learning

Earlier studies have indicated that decision-making by a project development team can be improved throughout the design and development process by understanding the key factors that affect customers evaluations of a new product. Aspect extraction could thus be a useful tool for identifying important attributes when evaluating products or services. Aspect extraction based on deep convolutional neural networks has recently been suggested, demonstrating state-of-the-art performance when applied to a customer review of electronic devices. However, this approach is unsuited to the rapidly evolving smartphone industry, which involves a wide range of product lines. Whereas the previous approach required significant amounts of data labeling for each product, we propose a variant of that approach that includes transfer learning. We also propose a novel approach for transferring the architecture sequentially within the product group. The results indicate that the principal key feature of each product is extracted effectively by the proposed method without having to re-train the entire convolutional neural network model. Furthermore, the proposed method performs better than the previous method for each product line.