Conversational Information Seeking

Conversational information seeking (CIS) is concerned with a sequence of interactions between one or more users and an information system. Interactions in CIS are primarily based on natural language dialogue, while they may include other types of interactions, such as click, touch, and body gestures. This monograph provides a thorough overview of CIS definitions, applications, interactions, interfaces, design, implementation, and evaluation. This monograph views CIS applications as including conversational search, conversational question answering, and conversational recommendation. Our aim is to provide an overview of past research related to CIS, introduce the current state-of-the-art in CIS, highlight the challenges still being faced in the community. and suggest future directions.

💡 Research Summary

Conversational Information Seeking (CIS) is an emerging research area that focuses on multi‑turn natural language interactions between users and information systems. This monograph provides a comprehensive overview of CIS, covering definitions, system architectures, interaction modalities, core tasks, evaluation methods, and open research challenges.

The authors begin by positioning CIS within the broader information retrieval (IR) landscape, emphasizing that recent advances in machine learning—particularly large‑scale language models—and the proliferation of voice‑enabled devices have accelerated the development of systems capable of understanding and generating conversational language. Unlike traditional IR, which typically handles isolated queries, CIS systems must maintain context over multiple turns, model user intent, and adapt their behavior based on both system‑initiated and user‑initiated actions.

A high‑level architecture is presented, consisting of four main components: (1) conversational interfaces and result presentation, (2) dialogue flow tracking and state management, (3) next‑utterance generation, and (4) initiative control. The interface layer supports a variety of modalities—text chat, spoken dialogue, live‑chat support, and chatbot widgets—each with distinct design considerations. Result presentation has evolved from classic search result lists to speech bubbles, multimodal cards, and hybrid visual‑audio layouts, reflecting the need to convey information effectively across devices and contexts.

Understanding the dialogue is tackled through several modeling approaches. Turn‑level state representations capture salience, while history‑based models (e.g., graph‑structured conversation histories) and discourse‑structure analyses (e.g., dialogue trees, intent flows) provide richer context. Language‑understanding tasks such as turn salience detection, query expansion, query rewriting, and entity detection/linking are discussed, with both unsupervised (clustering, topic modeling) and supervised (sequence labeling, span prediction) methods highlighted. Long‑term and multi‑session conversations require persistent user profiles and dynamic intent tracking, which the authors argue remain open problems.

Response generation and ranking are examined in depth. Early work focused on short answer selection, but recent systems employ transformer‑based conversational QA models, open‑retrieval pipelines that combine document retrieval with generative models, and knowledge‑graph‑driven QA. For longer answers, ranking mechanisms incorporate summarization and re‑generation techniques. Procedural and task‑oriented ranking leverage reinforcement learning to optimize multi‑step information‑seeking goals. The chapter also discusses how conversational recommendation integrates preference elicitation with dialogue flow, enabling systems to suggest items while simultaneously refining user models.

Mixed‑initiative interaction receives special attention. The monograph categorizes system‑initiated conversations, clarification question generation, preference elicitation, and feedback loops. Clarification strategies include template‑based slot filling, sequence editing, seq2seq generation, and utility‑maximization models; each is evaluated for relevance, diversity, and user burden. Preference elicitation is framed as an iterative dialogue where user choices continuously update a latent preference vector, guiding subsequent recommendations.

Evaluation is split into offline and online paradigms. Offline evaluation relies on publicly available conversational datasets (single‑turn, multi‑turn, multimodal) and simulated users, with metrics such as accuracy, MRR, NDCG for individual components and dialogue length, task success rate, and user satisfaction for end‑to‑end performance. Online evaluation includes lab studies, crowdsourced experiments, and real‑world deployments, highlighting challenges in constructing realistic test collections for highly personalized, adaptive conversations. The authors note the difficulty of obtaining reliable human judgments for multi‑turn interactions and propose the use of high‑fidelity simulators and hybrid human‑in‑the‑loop methods.

The concluding chapter outlines five major research directions: (1) achieving human‑level natural language understanding and long‑range coherence, (2) standardizing evaluation frameworks and benchmarks, (3) expanding beyond text and speech to multimodal interaction (vision, gestures, AR/VR), (4) incorporating privacy, fairness, and safety considerations into CIS system design, and (5) developing domain‑specific CIS solutions for areas such as healthcare, legal assistance, and education. The authors advocate for interdisciplinary collaboration, large‑scale user log analysis, and open‑source sharing of tools and datasets to advance the field.

Overall, this monograph serves as a definitive reference for researchers and practitioners, summarizing the state‑of‑the‑art in conversational information seeking, identifying persistent challenges, and charting a roadmap for future innovations.