The First Workshop on Ubiquitous and Multi-domain User Modeling (UMUM2021),
Held in Conjunction with EAI MobiQuitous 2021

Date: 8th November, 2021
Venue: Fully online (On-site available; Beppu, Japan * dependent on COVID-19 situation)

Deep Learning (DL) has recently been widely applied to user modeling for real-world applications such as context-aware recommendation or location-aware services. However, many studies focus on building DL models only in one or few domains and applying them to identical domains. Thus, how to build ubiquitous and multi-domain user models and increase their applicability in real-world contexts remain an open problem. Here, we refer to “ubiquitous” as the capability to adapt to, and therefore exist in diverse contexts, and “multi-domain” as the capability of training using data from multiple domains.

We think that there are several challenges towards realizing ubiquitous and multi-domain user modeling, as follows:

  • Real-world environment is a mixed environment in which many users and many objects are interacting in various ways. How to learn diverse contexts from user-object interaction data obtained from smartphones, IoT platforms, customer bases, and so on, is a challenge.
  • Collecting personal data from multiple domains and aggregating them in one place become increasingly difficult due to privacy concerns or huge data volume. How to model different user behaviors in each individual domain, integrate the models to enhance every prediction and, consequently, improve modeling performance is another challenge.
  • The relationships between multiple domains may change from time to time. Therefore, it will be a challenge to detect and adapt to such changes so that the performance of the models do not degrade.