Next Product Title Generation in E-commerce: Rule-based Methods and Autoencoder Model

Junya Miyamoto, Shinnosuke Hirano, Soma Makino, Kazuya Uekado, Xinyi Lao

Keywords: Session-based recommendations, Multilingual recommendation systems, New Cold Start product title generation, KDD CUP

Abstract: This paper primarily presents our approach and results for Task 3: Next Product Title Generation in Amazon KDD Cup 2023. Our approach attempted to combine rule-based methods with an auto encoder model. Initially, we set the baseline using the titles of the last item, and then we investigated a rule-based method to generate the next product title based on common words in the titles of products the customer recently interacted with. Subsequently, we experimented with an auto encoder model trained with LSTM, incorporating it with the rule-based approach to generate the next product title. The result demonstrated that the rule-based method achieved the best score. Our approach achieved a 10th place ranking in Task3 in Amazon KDD Cup 2023 competition.

Submission Number: 14