User Modeling and User Profiling: A Comprehensive Survey

📝 Paper Summary

User modeling User profiling

This comprehensive survey provides a historical overview, a novel taxonomy, and updated encyclopedic definitions for user modeling and profiling, highlighting the shift toward implicit, deep learning-based, and ethical approaches.

Core Problem

Despite decades of research, the field of user modeling suffers from ambiguous core terminology, a lack of a unified taxonomy encompassing recent deep learning trends, and outdated surveys that miss the post-2019 explosion of ethical and neural approaches.

Why it matters:

Ambiguous definitions of 'user model' vs. 'user profile' confuse researchers and hinder standardized communication across sub-fields like recommender systems and HCI.
Rapid advancements in deep learning (transformers, GNNs) and ethical AI (fairness, privacy) have rendered existing surveys from 2019 and earlier insufficient for current practitioners.
The lack of a general taxonomy fragments the understanding of how diverse techniques—from early stereotypes to modern LLMs—fit into a cohesive user modeling framework.

Concrete Example: Previous literature often uses 'user model' and 'user profile' interchangeably or relies on outdated definitions from the 1990s that do not account for modern embedding-based representations. This paper rectifies this by proposing distinct, encyclopedic definitions.

Key Novelty

Unified User Modeling Taxonomy & Encyclopedic Redefinition

Proposes two novel encyclopedic definitions to clearly distinguish 'user modeling' (the process) from 'user profiling' (the representation), resolving long-standing terminological ambiguity.
Constructs a comprehensive taxonomy that categorizes the field's evolution from static stereotypes to dynamic, deep learning-based, and 'beyond-accuracy' (ethical/privacy-preserving) approaches.

Breakthrough Assessment

7/10

A highly necessary consolidation of a fragmented field. While it is a survey rather than a new algorithm, its structured taxonomy and clear definitions provide a foundational resource for future research.

⚙️ Technical Details

Problem Definition

Setting: Survey and taxonomy construction for the field of user modeling and profiling.

Inputs: Academic literature from late 1970s to present.

Outputs: Historical overview, novel taxonomy, standardized definitions, and future research directions.

Pipeline Flow

Historical Analysis (1970s-2020s)
Terminological Definition
Taxonomy Creation
Trend Analysis (Deep Learning, Ethics, Graphs)
Application Domain Review

System Modules

Historical Analysis

Trace evolution from stereotypes (1979) to semantic web (2000s) to deep learning (2010s+)

Model or implementation: Literature Review

Taxonomy Creation

Categorize active research topics including recent trends

Model or implementation: Hierarchical Classification

Trend Analysis

Detail specific modern techniques like GNNs, Transformers, and Federated Learning

Model or implementation: Technical Review

Novel Architectural Elements

Proposal of two novel encyclopedic definitions for 'User Modeling' and 'User Profiling' to resolve ambiguity
Creation of a comprehensive taxonomy encompassing both traditional methods and modern Deep Learning/Ethical trends

Comparison to Prior Work

vs. Eke et al. (2019): Updates the field with post-2019 advancements in Deep Learning, GNNs, and ethical concerns like Fairness and Explainability
vs. Cena et al. (2019): Provides a broader general taxonomy rather than focusing solely on ubiquitous/wearable contexts
vs. Li and Zhao (2020): Covers the entire field history and definitions, not just representation learning algorithms
+ 1 more
vs. Tan and Jiang (2023): Offers a holistic view of the field's history and diverse techniques, rather than focusing exclusively on LLMs

Limitations

As a survey, it does not propose a new algorithmic method or provide experimental results
The breadth of the survey may limit the depth of technical detail provided for individual deep learning architectures compared to specialized reviews

Reproducibility

Not applicable — this is a survey paper and does not provide code or experiments to reproduce.

📊 Experiments & Results

Evaluation Setup

Qualitative literature survey and taxonomy construction

Metrics:

Statistical methodology: Not applicable

Main Takeaways

Paradigm shift from Explicit to Implicit: Modern user profiling relies almost exclusively on implicit behavioral data or 'pseudo-explicit' public data rather than surveys.
Deep Learning dominance: The field has moved from statistical and semantic models to Deep Learning architectures (RNNs, LSTMs, Transformers, GNNs) for capturing complex user behaviors.
Beyond-accuracy importance: Evaluation now prioritizes ethical dimensions (Fairness, Privacy, Explainability) alongside traditional accuracy metrics.
Universal Representation gap: Despite progress, a truly generalized, cross-domain user model remains an open challenge (Universal User Representation).

📚 Prerequisite Knowledge

Prerequisites

Basic understanding of Recommender Systems (RS) and Information Retrieval (IR)
Familiarity with Machine Learning concepts (Deep Learning, Graph Neural Networks)
Awareness of ethical AI concepts (Fairness, Privacy, Explainability)

Key Terms

User Modeling: The process of identifying and inferring specific user characteristics, interests, and behaviors from data to create a conceptual understanding of the user.

User Profiling: The application of user modeling to construct a specific, often digital, representation (profile) of a user, used to tailor services or content.

Stereotypes: Predefined sets of characteristics associated with user groups, used in early systems (like Grundy) to infer preferences for new users based on minimal initial data.

Implicit User Profiling: Constructing user profiles based on observed behaviors (clicks, dwell time) rather than asking users for explicit feedback.

Beyond-accuracy: Evaluation perspectives that prioritize metrics other than prediction precision, such as fairness, privacy, diversity, and explainability.

Expert Finding: Identifying individuals within a dataset who possess specific knowledge or expertise on a topic.

Expert Profiling: Creating a comprehensive profile of an individual's skills and knowledge based on their associated documents and activities.

Adaptive Hypermedia: Systems that tailor the presentation of hyperlinks and content to the user's goals and knowledge level.

Graph Neural Networks (GNNs): Deep learning models designed to process data structured as graphs, increasingly used to model social connections and user-item interactions.

Federated User Modeling: A privacy-preserving approach where user models are trained across decentralized devices holding local data samples, without exchanging them.