23_Mini_Project : ✧ Book-Recommendation ✧

• Usage:

  • Session
  • Conference
  • Mini projects
  • Study

• Period:

  • 2024.10 ~ 2024.12

• Team:

  • Team Manager: 윤왕규
  • Team Members: 김경민, 백다운, 오영민

---

Backgrounds

• We initiated this project to deepen our understanding of the ongoing advancements and emerging trends in the field of recommendation systems.
• We chose LightGCN and GLocal-K to develop models that precisely capture users’ inherent preferences while ensuring scalability and efficiency.
• As research increasingly explores integrating recommendation systems with large language models (LLMs), we plan to maintain an active interest and continue our exploration in this direction.

Data

• This is a large-scale Amazon Reviews dataset, collected in 2023 by McAuley Lab, and it includes rich features such as:
  • User Reviews (ratings, text, helpfulness votes, etc.)
  • Item Metadata (descriptions, price, raw image, etc.)
  • Links (user-item / bought together graphs)
• Link : https://amazon-reviews-2023.github.io

Models

• Taking into account that the actual rating matrix is sparse and side information is limited, we aimed to implement an accurate recommendation model tailored to each user’s preferences.
• To achieve this, we formed teams of two, studied the relevant research papers, and implemented the two models described below.

1) GLocal-K

• Paper : https://arxiv.org/abs/2108.12184
• Reference : https://paperswithcode.com/paper/glocal-k-global-and-local-kernels-for

2) LightGCN

• Paper : https://arxiv.org/abs/2002.02126
• Reference : https://paperswithcode.com/method/lightgcn

Metrics

• NDCG
• Recall
• Precision
• Map

Conclusion

• After performing EDA on the Amazon book dataset, we found that there were relatively few user reviews per individual item, despite the dataset containing a wide variety of different items.
• Since this was our first experience implementing a research paper, we encountered many difficulties, and it’s regrettable that we weren’t able to modify the model itself to better suit our dataset.
• We also regret not being able to implement an app or web interface to showcase the recommended results, which highlights a technical shortcoming in our approach.