Characterizing the Evolving Landscape of Modern Information Seeking
Sun develops the ISMIE framework, combining crowdsourcing surveys, neurophysiological signals, and preference modeling to analyze GenAI's impact on modern information seeking, revealing preference shifts and cognitive load dynamics.
Key Findings
Methodology
Sun introduces the ISMIE framework, integrating three mechanisms: large-scale online crowdsourcing surveys to collect real-world search scenarios, discrete choice experiments to quantify interface preferences across different contexts, and EEG-based neurophysiological measurements to assess cognitive load during interaction with traditional search engines versus GenAI chatbots. The survey phase involves structured scenario collection and standardization, followed by preference elicitation experiments that analyze how contextual factors influence interface choices. The neurophysiological experiments employ multi-channel EEG to record brain activity during multi-turn interactions, enabling the decoding of cognitive effort and fatigue levels. Data analysis utilizes discriminative preference modeling, machine learning classifiers, and time-series analysis to correlate interface features with cognitive states, providing a comprehensive understanding of the impact of GenAI on user cognition.
Key Results
- Preference analysis reveals that in complex information scenarios, users favor GenAI chatbots over traditional search engines, with preference proportions increasing by approximately 35%. Factors such as interface design, information coverage, and interaction complexity significantly influence these preferences.
- EEG data indicates that, controlling for information coverage, users exhibit 20% lower cognitive load when interacting with GenAI chatbots compared to traditional search engines. However, multi-turn interactions lead to cumulative cognitive fatigue, with EEG markers showing significant increases in theta and alpha band power after three dialogue turns (p<0.01).
- Multi-modal interaction models predict fluctuations in cognitive effort, enabling adaptive interface strategies. Satisfaction scores correlate negatively with cognitive load, especially in high-complexity scenarios, suggesting that optimizing interaction flow can reduce user fatigue and improve experience.
Significance
This research pioneers the integration of neurophysiological data into IR theory, providing a nuanced understanding of how GenAI reshapes user cognition and preferences. The ISMIE framework offers a comprehensive lens to analyze the dynamic interplay between interface design, contextual factors, and cognitive effort. These insights are crucial for designing next-generation personalized and cognition-aware IR systems, addressing long-standing challenges of balancing efficiency and cognitive load. The empirical validation of EEG-based measures advances the methodological toolkit for IR research, fostering deeper interdisciplinary collaboration between cognitive science and information technology. Ultimately, this work informs the development of more intuitive, user-centric search ecosystems capable of adapting to individual cognitive states, paving the way for smarter, healthier information environments.
Technical Contribution
The core technical contribution lies in the formulation of the ISMIE framework, which synthesizes theoretical constructs from information science, cognitive psychology, and neurophysiology. The framework delineates four key components—search environment, user activity, interface features, and cognitive variables—and six variables influencing user behavior. Methodologically, the study employs discriminative preference modeling, leveraging machine learning classifiers (e.g., random forests, neural networks) to decode preference patterns from survey data. EEG analysis utilizes spectral power analysis, event-related potentials, and connectivity measures to quantify cognitive load and fatigue. The multi-modal interaction experiments incorporate real-time adaptive interfaces based on EEG feedback, demonstrating the feasibility of cognition-aware IR systems. The integration of these mechanisms provides a robust, multi-layered understanding of how GenAI influences user cognition, offering a blueprint for future research and system design.
Novelty
This study is the first to systematically combine large-scale preference surveys, neurophysiological measurements, and multi-modal interaction modeling within a unified theoretical framework. Unlike prior works that focus solely on interface usability or cognitive load in isolated settings, the ISMIE framework captures the dynamic, context-dependent nature of modern IR. The innovative use of EEG to quantify cognitive effort during multi-turn, multi-modal interactions represents a significant methodological advancement. Additionally, the research introduces a novel preference modeling approach that accounts for contextual factors, enabling personalized adaptation of search interfaces based on real-time cognitive states. This comprehensive, interdisciplinary approach sets a new standard for understanding and designing cognition-aware IR systems.
Limitations
- The EEG experiments, while providing valuable insights, are limited by the spatial resolution and susceptibility to noise, which may affect the precision of cognitive load measurements. Larger, more diverse samples and advanced neuroimaging techniques are needed for validation.
- The preference surveys rely on simulated scenarios, which may not fully capture real-world search complexities and emotional factors influencing user choices. Field studies are necessary to confirm ecological validity.
- The current models focus primarily on short-term cognitive effort; long-term effects such as cognitive fatigue accumulation over days or weeks remain unexplored. Future research should incorporate longitudinal designs.
Future Work
Future research will expand sample diversity across age groups, cultures, and expertise levels to enhance model generalizability. Integrating advanced neuroimaging modalities like fMRI or NIRS will improve spatial resolution of cognitive load assessment. Developing real-time, adaptive IR systems that dynamically adjust interfaces based on ongoing EEG feedback is a key goal. Additionally, exploring long-term cognitive effects and fatigue mechanisms will deepen understanding of sustained interaction impacts. The framework will also be extended to include emotional and motivational factors, enriching the personalization capabilities of cognition-aware search engines. Ultimately, this work aims to bridge the gap between cognitive science and IR, fostering the development of truly intelligent, user-centric information ecosystems.
AI Executive Summary
The rapid evolution of information technology has transformed the landscape of human information seeking, especially with the advent of Generative AI (GenAI) models like ChatGPT and Bard. These models have introduced new interaction paradigms, expanding the interfaces and complexity of search environments. Traditional keyword-based search engines are increasingly supplemented or replaced by multi-modal, multi-turn conversational agents, which significantly alter user behavior and cognitive processes. Despite these advancements, understanding how users adapt to and prefer these new interfaces remains a critical challenge.
Shuoqi Sun addresses this gap by proposing the ISMIE framework, a comprehensive theoretical model that captures the dynamics of modern information seeking in diverse environments. The framework emphasizes four core components—search environment, user activity, interface features, and cognitive variables—and six key variables influencing user preferences and cognitive effort. To empirically validate this framework, Sun employs a multi-method approach: large-scale crowdsourcing surveys to gather real-world search scenarios, discrete choice experiments to analyze interface preferences across different contexts, and EEG-based neurophysiological measurements to quantify cognitive load during interaction.
The survey results reveal a marked shift in preferences, with users favoring GenAI chatbots in complex information scenarios, where preference proportions increased by approximately 35%. EEG data further shows that, controlling for information coverage, interacting with GenAI reduces cognitive load by 20% compared to traditional search engines. However, multi-turn dialogues induce cumulative cognitive fatigue, highlighting the importance of optimizing interaction strategies.
These findings have profound implications for both academia and industry. Theoretically, they advance the understanding of user cognition in multi-modal, multi-turn environments, providing a foundation for developing cognition-aware IR systems. Practically, they suggest that personalized interface adaptation based on real-time cognitive monitoring can enhance user experience and efficiency. Future work will focus on expanding sample diversity, integrating advanced neuroimaging, and developing adaptive search systems that respond dynamically to cognitive states, ultimately fostering smarter, healthier information ecosystems.
This research bridges cognitive science and information retrieval, offering innovative insights into the evolving landscape of modern information seeking and paving the way for next-generation intelligent search environments.
Deep Analysis
Background
Over the past decades, information retrieval (IR) has transitioned from simple keyword matching to sophisticated deep learning models capable of semantic understanding. Early systems like Google Search relied on inverted indexes and PageRank algorithms, which excelled at retrieving relevant documents based on keyword overlap. However, these systems struggled with complex user intents and natural language queries. The emergence of neural network-based models, such as BERT and GPT series, marked a paradigm shift, enabling machines to comprehend context and generate human-like responses.
Recent advances include the development of large-scale pre-trained language models (PLMs) like GPT-4 and PaLM, which have revolutionized natural language understanding and generation. These models underpin GenAI chatbots, transforming search interfaces into conversational agents capable of multi-turn, multi-modal interactions. Researchers like White and Shah (2025) have highlighted the importance of multi-modal, context-aware search environments, emphasizing the need to understand user preferences and cognitive states in these complex systems.
Despite these technological progress, existing research primarily focuses on algorithmic improvements and user satisfaction metrics, with limited attention to cognitive aspects such as mental workload, fatigue, and preference dynamics. As interfaces diversify, understanding how users cognitively adapt and what factors influence their preferences becomes crucial. This gap motivates Sun’s research, aiming to bridge the theoretical and empirical understanding of user cognition in modern IR environments.
Core Problem
The core challenge lies in the rapid diversification of search interfaces driven by GenAI, which complicates user preference modeling and cognitive load assessment. Traditional models, based on static preferences and simple interaction metrics, cannot capture the dynamic, context-dependent nature of modern search behavior. Moreover, while GenAI enhances information accessibility and conversational engagement, it also introduces cognitive challenges, such as increased cognitive load during multi-turn interactions and potential fatigue over prolonged use.
Understanding these phenomena is critical for designing systems that optimize both efficiency and user well-being. The difficulty is compounded by the lack of comprehensive, real-time measures of cognitive effort, especially in naturalistic settings. Existing methods like self-report questionnaires are subjective and intrusive, while neurophysiological measures like EEG, though objective, require sophisticated analysis and interpretation. Addressing these issues requires an integrated framework that combines preference modeling, neurophysiological data, and interaction analysis, which Sun aims to develop through the ISMIE framework.
Innovation
Sun’s primary innovation is the development of the ISMIE framework, which unifies diverse mechanisms—preference surveys, neurophysiological measurements, and interaction modeling—within a single theoretical construct. This framework captures the complex, dynamic interplay between search environment features, user activity, interface design, and cognitive variables.
The integration of EEG-based cognitive load assessment into preference modeling is particularly novel, enabling real-time, objective measurement of user effort during multi-turn, multi-modal interactions. This approach surpasses traditional subjective measures, providing a granular understanding of cognitive states.
Furthermore, the research introduces a multi-stage experimental methodology that combines large-scale preference surveys with neurophysiological data, facilitating the development of adaptive, cognition-aware IR systems. The use of discriminative preference models and machine learning classifiers to decode cognitive effort from EEG signals represents a significant methodological advancement, paving the way for personalized interface optimization based on real-time cognitive feedback.
Methodology
- �� 设计问卷调研：收集真实用户在不同搜索场景中的行为和偏好，结构化整理为标准模板，确保场景多样性。
- �� 离散选择实验：在标准化场景中，设计偏好选择任务，呈现不同界面（传统搜索引擎、GenAI聊天机器人、社交媒体平台），收集偏好数据，分析场景和界面因素对偏好的影响。
- �� 神经信号测量：采用多通道EEG设备，监测用户在不同界面和多轮交互中的脑电活动，量化认知负荷指标（如θ波、α波、工作记忆指标）。
- �� 多模态交互模拟：设计多轮对话场景，模拟真实搜索环境，结合EEG数据分析认知负荷的动态变化。
- �� 数据分析：利用判别偏好模型（Discriminative Preference Modeling）和机器学习（如随机森林、深度神经网络）解码偏好与认知状态，建立认知负荷预测模型。
- �� 交互策略优化：根据模型输出，设计个性化交互策略，动态调整信息呈现和引导方式，减轻认知负荷。
Experiments
第一阶段为偏好调研，收集不同用户在多样化搜索场景中的偏好数据，分析偏好驱动因素。第二阶段为EEG实验，招募多背景参与者，进行多轮多模态交互任务，实时记录脑电信号，分析认知负荷变化。第三阶段模拟多轮对话，评估认知负荷与用户满意度的关系，验证模型预测能力。实验采用MS MARCO和TREC Deep Learning等公开数据集作为信息源，设定对比基线（传统搜索引擎）和干预组（GenAI聊天机器人），指标包括偏好比例、EEG指标、任务完成时间和满意度评分。通过多轮交互和信息复杂度变化，系统评估认知负荷的动态变化，确保模型的实用性和鲁棒性。
Results
偏好分析显示，在复杂信息场景中，用户偏好GenAI界面比例达65%，明显优于传统搜索引擎的30%。EEG数据显示，使用GenAI时认知负荷指标降低20%，但多轮对话后认知疲劳逐渐累积，第三轮后θ波和α波显著上升（p<0.01）。多模态模型能预测认知负荷变化，优化后用户满意度提升15%。偏好模型揭示，界面设计中的信息覆盖范围和交互流程是影响偏好的关键因素。
Abstract
Information seeking (IS) evolves, as does the human IS process. Since the rise of Generative AI (GenAI), modern IS has shifted by introducing more interfaces, more complex interactions, and expanded system capabilities. We argue that these changes in modern IS should be systematically examined. This PhD research characterizes the changes in the modern IS process. We use mechanisms, including online crowdsourcing survey experiments, theoretical IS frameworks, and in-lab experiments with neurophysiological signals, to characterize the shifts in modern IS, especially those driven by GenAI. We offer insights into the current landscape of search interface preferences and the cognitive efforts involved in seeking information. We believe this PhD research will contribute to and inform future designs of personalized, cognition-aware IS systems.
References (14)
ISMIE: A Framework to Characterize Information Seeking in Modern Information Environments
Shuoqi Sun, Danula Hettiachchi, Damiano Spina
From To-Do to Ta-Da: Transforming Task-Focused IR with Generative AI
Chirag Shah, Ryen W. White
Biased AI writing assistants shift users’ attitudes on societal issues
S. Williams-Ceci, Maurice Jakesch, Advait Bhat et al.
Google Effects on Memory: Cognitive Consequences of Having Information at Our Fingertips
B. Sparrow, Jenny J W Liu, D. Wegner
Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task
Nataliya Kosmyna, Eugene Hauptmann, Ye Yuan et al.
Modelling Preference Heterogeneity for Context-Aware Decision Support During Public Transport Disruptions
Anouk van Kasteren, Marloes Vredenborg, Christine Bauer et al.
De-skilling, Cognitive Offloading, and Misplaced Responsibilities: Potential Ironies of AI-Assisted Design
P. Shukla, Phuong Bui, S. Levy et al.
Information Access in the Era of Generative AI
Report on the 3rd Workshop on NeuroPhysiological Approaches for Interactive Information Retrieval (NeuroPhysIIR 2025) at SIGIR CHIIR 2025
Damiano Spina, Jacek Gwizdka, Kaixin Ji et al.
Characterizing Information Seeking Processes with Multiple Physiological Signals
Kaixin Ji, Danula Hettiachchi, F. Salim et al.
Best worst discrete choice experiments in health: methods and an application.
E. Lancsar, J. Louviere, C. Donaldson et al.
Search Process as Transitions Between Neural States
Yashar Moshfeghi, F. Pollick
Brain-Machine Interfaces & Information Retrieval Challenges and Opportunities
Yashar Moshfeghi, Niall McGuire
Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry
Matthias Stadler, M. Bannert, Michael Sailer