Why Recommender Systems? 

 Research in recommender systems has been ongoing for quite some time, yet there remains a wealth of aspects yet to be explored. We identify four compelling reasons to focus our efforts on recommender systems

• Volume and Variety of Data: The core of a recommender system is to better understand user behaviors and intentions. The increasing volume of data enhances our ability to model these behaviors. Additionally, as web system interfaces become more sophisticated, we can capture more nuanced and multifaceted user data, such as clicks, add-to-cart actions, favorites, and purchases.

• Advances in High Performance Computing: With the progression of Moore's Law, computing chips are becoming faster, and GPUs are now a staple in computing. This evolution provides the necessary processing power for complex tasks.

• Deep Learning-Based Model Architecture: The combination of the first two factors enables the development of increasingly sophisticated neural network architectures. These architectures leverage the growing data volume and diverse user behaviors, opening numerous possibilities in designing model architectures.

• Societal Impact: The growth of platform companies that provide products and services online underscores the significance of even a 1% improvement in accuracy, which can have substantial business implications. Our interest lies in research areas with significant business impacts, making the field of recommender systems, with its potential for advanced model architectures, an attractive domain for our focus.

By collaborating with various industrial partners, we are seeking efficient and effective ways for user behavior modeling with deep learning.   

• News platform :   

• Mobility platform: 

• E-commerce platform: 

• Job matching platform

• Medical platform

As of Sep 2023, we have also started to work on how large language models can be leveraged for recommender systems. As for pretraining phase of LLMs, we are part of the teams in the LLM-JP project (https://llm-jp.nii.ac.jp/).

E-Commerce Recommendation  

 Lookalike models are based on the assumption that user similarity plays an important role towards product selling and enhancing the existing advertising campaigns from a very large user base. Challenges associated to these models reside on the heterogeneity of the user base and its sparsity. In this work, we propose a novel framework that unifies the customers different behaviors or features such as demographics, buying behaviors on different platforms, customer loyalty behaviors and build a lookalike model to improve customer targeting for Rakuten Group, Inc. Extensive experiments on real e-commerce and travel datasets demonstrate the effectiveness of our proposed lookalike model for user targeting task.

• Md. Mostafizur Rahman, Daisuke Kikuta, Satyen Abrol, Yu Hirate, Toyotaro Suzumura, Pablo Loyola, Takuma Ebisu, Manoj Kondapaka: Exploring 360-Degree View of Customers for Lookalike Modeling. SIGIR 2023: 3400-3404 (SIGIR '23: Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval) 

News Recommendation   

 Precisely recommending candidate news articles to users has always been a core challenge for personalized news recommendation systems. Most recent works primarily focus on using advanced natural language processing techniques to extract semantic information from rich textual data, employing content-based methods derived from local historical news. However, this approach lacks a global perspective, failing to account for users’ hidden motivations and behaviors beyond semantic information. To address this challenge, we propose a novel model called GLORY (Global-LOcal news Recommendation sYstem), which combines global representations learned from other users with local representations to enhance personalized recommendation systems. We accomplish this by constructing a Global-aware Historical News Encoder, which includes a global news graph and employs gated graph neural networks to enrich news representations, thereby fusing historical news representations by a historical news aggregator. Similarly, we extend this approach to a Global Candidate News Encoder, utilizing a global entity graph and a candidate news aggregator to enhance candidate news representation. Evaluation results on two public news datasets demonstrate that our method outperforms existing approaches. Furthermore, our model offers more diverse recommendations.
Representative Papers
Going Beyond Local: Global Graph-Enhanced Personalized News Recommendations, RecSys 2023,  Best Full Paper Runner-Up / Best Student Paper Award 
https://arxiv.org/abs/2307.06576

Job Recommendation 

We have been working on job recommendations and matching problems by collaborating with industrial partners. You can check our Japanese article (Please use Google Translate if needed ) to explain the overview of the project. https://note.com/utokyo_itc/n/nde9d29bec8d4.   For technical details, please check out our published paper in the AAAI 2024 workshop as follows.

Waki Satoshi, Toyotaro Suzumura and Hiroki Kanezashi, Optimizing Matching Markets: A Comprehensive Approach Using Graph Neural Networks and Reinforcement Learning, AAAI 2024 EcoSys Workshop (Workshop on Recommendation Ecosystems: Modeling, Optimization and Incentive Design), Feb 2024, Accepted and To Appear

Abstract : This study introduces a groundbreaking recommendation system designed for matching markets, such as job placement and online dating, which goes beyond the traditional focus on individual user preferences. Traditional Reciprocal Rec- ommendation Systems in these markets often fail to consider the overall market dynamics, leading to a narrow focus on specific popular choices and neglecting the diversity of user needs. To address this, our approach conceptualizes the mar- ket as a network, utilizing Graph Neural Networks to analyze the intricate connections within this network. We also incor- porate Reinforcement Learning to optimize outcomes for the entire market, not just individual users. Furthermore, to ad- dress the issue of sparse user-item interactions in matching markets, our approach incorporates a novel graph data aug- mentation technique. This method enriches the network by adding labeled edges, enhancing the market’s representation. This augmentation facilitates more effective and varied rec- ommendations, leading to a noticeable increase in successful matches in various market scenarios, as evidenced by our of- fline experiments with both synthetic and real-world data.

Location Recommendation for Mobility 

We have been working on representation learning in the mobility area by collaborating with the Toyota corporation. Here are some publications related to this topic.  

Spatio-Temporal Meta-Graph Learning for Traffic Forecasting, AAAI 2023
https://ojs.aaai.org/index.php/AAAI/article/view/25976

• • Traffic forecasting as a canonical task of multivariate time series forecasting has been a significant research topic in AI community. To address the spatio-temporal heterogeneity and non-stationarity implied in the traffic stream, in this study, we propose Spatio-Temporal Meta-Graph Learning as a novel Graph Structure Learning mechanism on spatio-temporal data. Specifically, we implement this idea into Meta-Graph Convolutional Recurrent Network (MegaCRN) by plugging the Meta-Graph Learner powered by a Meta-Node Bank into GCRN encoder-decoder. We conduct a comprehensive evaluation on two benchmark datasets (i.e., METR-LA and PEMS-BAY) and a new large-scale traffic speed dataset called EXPY-TKY that covers 1843 expressway road links in Tokyo. Our model outperformed the state-of-the-arts on all three datasets. Besides, through a series of qualitative evaluations, we demonstrate that our model can explicitly disentangle the road links and time slots with different patterns and be robustly adaptive to any anomalous traffic situations. Codes and datasets are available at https://github.com/deepkashiwa20/MegaCRN.

Revisiting Mobility Modeling with Graph: A Graph Transformer Model for Next Point-of-Interest Recommendation, ACM SIGSPATIAL 2023 

• • Xiaohang Xu (The University of Tokyo), Toyotaro Suzumura (The University of Tokyo), Jiawei Yong (Toyota Motor Corporation), Masatoshi Hanai (The University of Tokyo), Chuang Yang (The University of Tokyo), Hiroki Kanezashi (The University of Tokyo), Renhe Jiang (The University of Tokyo), Shintaro Fukushima (Toyota Motor Corporation)