Candidate Reranking Solution using Many Variant Features for KDDCup2023, Team YAMALEX Solution

Hiroki Yamamoto, Takashi Sasaki, Shin Higuchi, Tomonori Fujiwara, Shun Yoshioka

Keywords: Re-ranking, Recommendation

TL;DR: Team YAMALEX KDDCup2023 solution, recommend system of Candidate/Re-ranking using many variant features(graph embedding, neural network, co-matrix, BPR).

Abstract:

It is essential for e-commerce stores to model customers’ shopping intentions which directly lead to user experience and engagement.
Although there is an increasing interest in utilizing session data to predict what the user will purchase next, there has not been many studies on session-based recommendation using real-world multi-lingual and imbalanced scenarios. In the Amazon KDD Cup 2023,Amazon presented the "Multilingual Shopping Session Dataset" with millions of user sessions from six different locales, namely:English, German, Japanese, French, Italian, and Spanish. The dataset introduces imbalance by having fewer data for French, Italian, andSpanish compared to other locales.
In this paper, we present our approach to this challenge using two stage approach to generate the candidates.

Submission Number: 13