Conference Agenda (All times are shown in Eastern Daylight Time)

Generative information retrieval (GenIR) systems represent a paradigm shift in interpreting and addressing user needs, yet their effectiveness remains constrained by limited interaction design capabilities for accurate intent understanding. Recognizing the connection between search intentions and task characteristics, this study identifies two categories of task characteristics that can inform the interaction design of GenIR systems: (a) general task characteristics, which consistently influence search behaviors across diverse contexts and user groups, and (b) specific task characteristics, which are particularly influential in certain contexts or among specific user groups. To identify these characteristics, the study quantitatively synthesizes existing research through a meta-analysis of 22 experimental studies. The findings suggest that task difficulty, task urgency, and task sources are key general task characteristics. However, task goal specificity is more influential in learning contexts, while task complexity predominates in daily life and healthcare contexts, as well as for middle school students. Based on these findings, this study develops a task-aware interaction design framework that strategically guides users in articulating both general and specific task characteristics through iterative dialogues. This design optimizes intent understanding by achieving: (1) operational efficiency through the analysis of general characteristics, and (2) personalized adaptation through the interpretation of specific characteristics.

People shopping online often abandon their shopping sessions because they feel overwhelmed or insufficiently supported during product searches. We instructed 31 participants to perform two goal-directed product searches online, simulating real-world scenarios for two product types: search products (laptops) and experience products (jackets). Through observation and think-aloud protocols, we captured user behavior across browser tabs and online resources, enabling us to develop a novel annotation scheme for product search that captures resources used, views seen, and actions taken. Qualitative analysis of these annotated sessions revealed nine distinct product search strategies, which participants often combined and applied at different stages of their search sessions. For each strategy, we describe similarities and differences between search and experience products and identify common strategy combinations across product types. Finally, by mapping these findings to established information-seeking models, we offer insights that can inform the design of more effective and supportive e-commerce platforms.

Expectation confirmation in data retrieval systems remains a critical yet understudied issue. To bridge this gap, this paper proposes a multi-dimensional framework for investigating the confirmation formation in the context of data retrieval, drawing upon the Expectation-Confirmation Theory and expectancy-value theory. Furthermore, it introduces a deep-learning model called Multiple-Expert System based on Bayesian Neural Networks (MES-BNN) to predict multi-dimensional confirmation by analyzing users’ search behavior data. The findings reveal the multi-dimensional and context-dependent nature of confirmation. In the data retrieval context, users discern gaps between their search experiences and expectations in terms of task, cognition, and emotion, collectively forming a three-dimensional confirmation. This three-dimensional confirmation can be predicted through search behavior data mining. Furthermore, the MES-BNN model demonstrates its effectiveness in mining small-scale behavioral data, enabling automatic and accurate prediction of the three-dimensional confirmation and contributes to advancing data analytical approaches employed in user-oriented retrieval studies in the intelligent age.