Main Research Area
Statistical Analysis of Network
Web-log Data / Process Data Analysis
Unstructured Data(text, conversation dialog) Analysis
Bayesian Adaptive Clinical Trial
Current Ongoing Research Projects
Analysis of Log Process Data
(1) This study focuses on analyzing behavioral transition patterns among users across countries using log data.
A hierarchical multi-state survival model is employed to estimate transition probabilities between states, which are then analyzed using network-based statistical methods to uncover transition patterns.
(2) This study focuses on enhanced item response model that incorporates log data from computer-based testing.
The current research involves embedding methods and aims to statistically analyze the resulting embedding vectors.
Statistical Analysis of Clickstream Data
This study introduces a metric space to measure similarity in clickstream data.
It explores how user behavior patterns converge to an underlying distribution, enabling hypothesis testing for pattern comparison.
Network Analysis on Political Science
The research focuses on discovering the inherent characteristics of legislators and bills, considering Congressional Roll-Call data as network data.
Hyperbolic Latent Space Joint Model for Reconstruction Phylogenetic Tree
The aim of this research is to develop a novel model for phylogenetic tree reconstruction based on latent space model.
Especially, we expand the latent space joint model (LSJM) into hyperbolic space in order to handle the latent position of the leaf nodes in a tree network properly.
We anticipate that this model can be a new alternative model of conventional Bayesian phylogenetics, for instance, BEAST or MrBayes, which has some shortcomings such as convergence instability.
Zero-inflated Latent Space Item Response Model for Count Data with an Application to Single-cell RNA Data
This study proposes a novel dimension reduction method for scRNA-seq data using a Zero-Inflated Negative Binomial model to address its inherent high dimensionality and sparsity.
It automatically identifies the optimal dimensions with a Spike-and-Slab Indian Buffet Prior, thereby capturing the key characteristics of the data.
Hierarchical Mediation Analysis Using Bayesian SVD
This study aims to develop a hierarchical extension of network mediation analysis that can estimate both lower and upper level mediation effect.
Estimating Social Influence and Selection Using Network Mediation Model
This study focuses on developing a mediation model that can interpret social influence and social selection by using social network as a mediator.
A Hybrid Approach to Estimating Social Selection
Combining Graph Attention Networks and Latent Space Model: This study proposes a novel methodology for detecting social selection from item-response data.
Using projections of a bipartite network, it estimates edge weights by Graph Attention which indicate the strength of social selection.
Causal Mediation Model with Item Responses as a Mediator
The research focuses on utilizing item response data as a mediator in causal mediation models.
Bayesian Singular Value Decomposition (BSVD) is employed to effectively manage the data.
Current efforts are dedicated to refining the BSVD methodology.
Hierarchical Latent Space Item Response Model Using Bayesian SVD
This study focuses on developing a LSIRM that reflects a hierarchical structure.
Doubly Layered Latent Space Item-Response Model
This study focuses on devloping a model-based clustering in Item-response data.Â
Dynamic Latent Space Network Model for Node Addition/Deletion
This study focuses on developing a latent space network model
to represent evolving networks, allowing for the addition and deletion of nodes over time.
Interaction Pattern Labeling Using Prompt Engineering
This research focuses on labeling complex interaction patterns
within student dialogue data using prompt engineering techniques.
Signed Network Analysis
This study focuses on developing a statistical method for analyzing signed networks, where relationships between nodes can be either positive or negative.
Time-to-event BOIN Design Incorporating Low-grade Toxicity Information
This study aims to develop a novel Phase I clinical trial methodology that incorporates low-grade toxicity into the TITE-BOIN(Time-to-Event Bayesian Optimal INterval) design.
By accounting for both dose-limiting and low-grade toxicities, the study seeks to refine dose escalation strategies and enhance patient safety.
Decomposition of Interaction Effects in LSM
This study employs matrix factorization within Latent Space Models (LSM) to analyze network data.
By adapting matrix decomposition techniques from covariance regression, the research aims to separate a multicomplex network data matrix into a component that is explained by covariates and a component that remains unexplained.
This methodology facilitates the separation of social selection effects from pure interaction effects.
Bayesian Dynamic Differential Network
This study proposes a Bayesian framework to analyze dynamic changes in protein-protein interactions using data from Cancer Perturbed Proteomics Atlas (CPPA) project.
The proposed model incorporates both binary treatment status and continuous time as covariates to characterize covariance structures, enabling the construction of time-specific differential networks.
By applying functional clustering, the study identifies distinct temporal patterns of interaction changes, providing network-level insights into drug mechanisms of action.
Global Value Chain Network Analysis
This study analyzes temporal patterns in international trade using network modeling of WITS global value chain data.
The research implements a Dynamic AMEN (DAME) model, extending the static framework to incorporate time-varying network structures.
A key objective is identifying countries' positions within global value chains and tracking positional shifts over time.
This approach reveals changing national roles and evolving interdependencies within global production networks.
Functional Clustering of Schools Based on Attraction Parameters Estimated from Item Response Matrices Using Inhomogeneous Point Processes
This study proposes a novel approach for comparing and analyzing multiple latent spaces estimated from different item response datasets measuring the same items, by recognizing these estimated latent spaces as point processes.
By applying attraction-repulsion point process models to the estimated latent spaces, we estimate attraction rate functions and use them as a basis for clustering multiple latent spaces.
This methodology provides a systematic framework for analyzing and comparing latent spaces across different item response datasets.