Vive la difference! Individualised interaction with users

IJCAI 1995

TL;DR

This paper argues for the critical role of user modeling in creating truly individualized and effective human-computer interactions. It proposes a framework for understanding user models, categorizing them into domain expertise, collective user models (stereotypes), and individual user models, and discusses the AI techniques needed to build and utilize them. The core contribution is a conceptualization of user modeling as a vital AI research area that bridges various subfields and offers solutions for managing the complexity of personalized systems.

What Problem Does This Paper Address?

The paper addresses the challenge of building computer systems that can effectively interact with individual users. Current systems often treat users as homogenous, failing to account for their diverse knowledge, preferences, and goals. This leads to suboptimal interactions, where systems might provide information users already know, omit crucial details, or fail to adapt to individual needs. The authors highlight that as we move towards an era of vast information mediated by machines, the ability to personalize interactions becomes paramount to prevent users from being overwhelmed. They frame user modeling as the solution, enabling systems to tailor their behavior, interpretation, and output to individual users.

Methods

The paper doesn't present a single novel algorithm but rather a conceptual framework for user modeling. It defines two main types of user models: user private (inaccessible to the machine) and user shared (accessible, like typed commands). Within the machine's private knowledge, it categorizes user models into three classes:

  • M<sub>domain</sub> (Domain Expertise Model): Represents knowledge about the domain itself, including goals and actions to achieve them.
  • M<sub>users, domain</sub> (Collective User Models/Stereotypes): Models the expertise of typical users within a domain, often categorized into groups like "beginner," "intermediate," and "guru." This is where stereotypes play a significant role.
  • M<sub>individual</sub> (Individual User Model): Models a specific user's unique knowledge, beliefs, goals, and progress. This model needs to handle inconsistencies and temporal changes.

The paper also discusses the AI techniques required for user modeling, including knowledge representation (handling noisy, changing data), automated reasoning, planning, machine learning, and natural language processing. It emphasizes the need for user control, access, and privacy in user model representations.

Datasets

Not discussed in the paper. The paper focuses on the conceptualization and architecture of user modeling systems, using examples like a Unix advisor and a customized hypertext newspaper, rather than empirical evaluation on specific datasets.

Key Insights and Ideas

  • User Modeling as a Unifying AI Domain: The paper posits user modeling as a fertile ground for AI research, demanding solutions from various subfields and, in turn, providing constraints and problems that can drive innovation in those fields.
  • The Three Pillars of User Models: Clearly distinguishing between domain knowledge, collective user knowledge (stereotypes), and individual user knowledge provides a structured way to think about what a system needs to know about its users and the world.
  • Stereotypes as Pragmatic Tools: Stereotypes are presented not as rigid classifications but as efficient ways to make initial assumptions about users, allowing systems to quickly adapt even with limited information.
  • The Importance of "User Private" vs. "User Shared": Recognizing what information is inherently private to the user versus what can be observed by the system is crucial for ethical and effective user modeling.
  • The Need for Dynamic and Robust Models: User models must be able to cope with incomplete, noisy, and changing information, reflecting the fluid nature of human knowledge and goals.

Key Results

The paper does not present quantitative results from a specific experiment. Instead, its key "results" are conceptual:

  • A clear taxonomy of user model types (M<sub>domain</sub>, M<sub>users, domain</sub>, M<sub>individual</sub>).
  • An architectural overview (Figure 2) of a system that uses user modeling for filtering and customization, demonstrating how these model types interact.
  • An argument for the necessity and benefits of user modeling in creating more effective and individualized human-computer interactions, particularly in information-rich environments.

Limitations and Future Work

  • Fragmentation of Research: The paper notes that user modeling research is fragmented across different AI communities, hindering the transfer of results.
  • Need for Interdisciplinary Cooperation: Exploiting individualization requires greater cooperation between AI subcommunities.
  • Ethical Considerations: While mentioned, the paper acknowledges that user control, access, and privacy are demanding aspects of knowledge representation for user models.
  • Future Work: The paper suggests that the complexity of modern interfaces and the increasing availability of information will make user modeling even more invaluable. It also points to the potential of machine learning for building collective user models and the need to combine the three classes of models effectively. The example system described points towards future work in customized hypermedia and filtering.

Summary updated

BibTeX
@inproceedings{1995vive,
  title = {{Vive la difference! Individualised interaction with users}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1995},
}
Figure