Prediction of next product title by simple deletion of the last word

Meisaku Suzuki, Ryo Koyama, Shin Ishiguro, Shunya Imada, Yuya Kunimoto

Keywords: KDD Cup, Multilingual Session Recommendation

Abstract: This paper presents the 9th rank method in KDD Cup 2023 task 3. In the Product Title Prediction Task of KDD Cup 2023 task 3, we improved the BLEU score by removing words in the second half of the product title, as opposed to the baseline method that uses the last product title as the predicted title, which may contain useless words. Since the second term of the BLEU formula penalizes product titles with long word counts, we improved BLEU by optimizing the second term of BLEU by removing words in the second half of the product title. The result was a 0.00268 improvement in BLEU over the baseline method, resulting in a final BLEU of 0.26821, which placed 9th on the Leader Board.

Submission Number: 7