[1N4-GS-13] AI application: Machine learning and application (1)

[1N4-GS-13-04] Recommendation method for new apparel products using CNN and siamese network

〇Teruaki Nakahara¹, Yukinobu Hamuro², Takanobu Nakahara³, Masakazu Nakamoto² (1. BRANSHES Co,.Ltd., 2. Kwansei Gakuin University, 3. Senshu University)

Keywords:Convolutional Neural Network, Siamese Network, Quantification of preferences, Feature vector, Recommendation

Product recommendation systems for apparel sales often use methods by using the action history data, such as collaborative filtering, or purchase history data and browsing history of online shopping sites, but those methods based on the sales history in the past. It was impossible to recommend the products that not included in the data. Therefore, in those researches, we tried to model the strength of the relationship between different products, by incorporating the features of product images to the customer’s purchase history data. It makes possible to recommend the new products based on the customer's preference of product images, that cannot be recommended by the conventional methods. The recommendation system constructed in this time, the feature vector of the product image is calculated using “Convolutional Neural Network (CNN)”, and used “Siamese Network” for modeling the strength of the relationship between different product images. Our proposed method is able to capture the customer’s preference of the product image.