A New Taxonomy of Web Search: A User-Centered Framework for Search Intent in the AI Era

Paper Title

A New Taxonomy of Web Search: A User-Centered Framework for Search Intent in the AI Era

Publication Info

• Topic area: Redefining user intent frameworks for modern web search and AI chatbot interactions.
• Keywords: search intent, AI chatbots, user-centered framework, taxonomy, information retrieval, generative AI, search engines, Broder’s taxonomy, task-centered, user behavior.

Background and Problem

• Problem / challenge: Existing search intent frameworks, particularly Broder’s taxonomy, are outdated and fail to capture the complexities of modern search behaviors, especially with the integration of AI chatbots and generative AI features.
• Significance: Misaligned frameworks can lead to flawed research, poorly designed systems, and user dissatisfaction, limiting the ability to develop adaptive and personalized search experiences.
• Motivation and related work: Broder’s taxonomy, developed in the early 2000s, remains widely used due to its simplicity, but it is increasingly inadequate for capturing multi-turn interactions, evolving intents, and AI-enabled tasks. While alternative models exist, they are often too conceptual or complex for practical application. This paper builds on prior work in query-based, task-based, and exploratory search models to propose a more user-centered approach.

Solution

• Proposed approach: A new task-centered taxonomy that categorizes user intent into three high-level categories: knowledge-seeking, guidance-seeking, and output-seeking, applicable to both search engines and AI chatbots.
• Novelty:
  1. Reconceptualization of search intent from a resource-centered to a task-centered perspective.
  2. Introduction of a session-based approach to capture evolving and overlapping intents.
  3. Differentiation between knowledge-seeking and guidance-seeking intents.
  4. Inclusion of new intent types to address AI-specific use cases, such as creative and generative tasks.
• Procedure and key techniques:
  • Conducted a survey where participants reflected on their last three search engine and AI chatbot interactions.
  • Applied Broder’s taxonomy deductively to identify its limitations.
  • Used inductive coding to develop a new taxonomy based on user goals and motivations.
  • Validated the new taxonomy by reclassifying sessions and comparing results against Broder’s framework.

Results

• Concrete findings:
  • Informational intent dominated under Broder’s taxonomy (68% of cases), but many sessions were ambiguous or unclassifiable.
  • The proposed taxonomy revealed a more balanced distribution: 43% knowledge-seeking, 33% guidance-seeking, and 24% output-seeking in search sessions; 36%, 40%, and 24% respectively in chatbot sessions.
• Advantage over baselines:
  • The new taxonomy addresses Broder’s limitations by focusing on user tasks rather than resource types, distinguishing between overlapping intents, and accommodating AI-specific behaviors.
• Experiments / evaluation:
  • Dataset: 86 participants, 500 sessions (247 search, 253 chatbot).
  • Methods: Deductive coding using Broder’s taxonomy, followed by inductive coding to develop and validate the new framework.
  • Metrics: Distribution of intents, clarity of classification, and ability to capture evolving user goals.
• Limitations and future work:
  • Sampling bias due to reliance on Prolific participants and self-reported data.
  • Ambiguity in defining session boundaries, especially for chatbots.
  • Need for broader validation across diverse datasets and user populations.
  • Future research should explore collaborative and socially motivated search behaviors, as well as the impact of emerging AI features.

Summary

This paper critiques the continued reliance on Broder’s taxonomy for search intent classification, arguing that it no longer reflects the complexities of modern search behaviors and AI-enabled interactions. Through a survey-based study, the authors propose a new task-centered taxonomy with three categories: knowledge-seeking, guidance-seeking, and output-seeking. The framework overcomes Broder’s limitations by focusing on user goals, accommodating evolving intents, and addressing AI-specific use cases. The findings highlight the need for updated frameworks to better understand and design for contemporary information retrieval systems.