Ryen W. White

Chirag Shah, Ryen W. White, Reid Andersen et al.

Log data can reveal valuable information about how users interact with Web search services, what they want, and how satisfied they are. However, analyzing user intents in log data is not easy, especially for emerging forms of Web search such as AI-driven chat. To understand user intents from log data, we need a way to label them with meaningful categories that capture their diversity and dynamics. Existing methods rely on manual or machine-learned labeling, which are either expensive or inflexible for large and dynamic datasets. We propose a novel solution using large language models (LLMs), which can generate rich and relevant concepts, descriptions, and examples for user intents. However, using LLMs to generate a user intent taxonomy and apply it for log analysis can be problematic for two main reasons: (1) such a taxonomy is not externally validated; and (2) there may be an undesirable feedback loop. To address this, we propose a new methodology with human experts and assessors to verify the quality of the LLM-generated taxonomy. We also present an end-to-end pipeline that uses an LLM with human-in-the-loop to produce, refine, and apply labels for user intent analysis in log data. We demonstrate its effectiveness by uncovering new insights into user intents from search and chat logs from the Microsoft Bing commercial search engine. The proposed work's novelty stems from the method for generating purpose-driven user intent taxonomies with strong validation. This method not only helps remove methodological and practical bottlenecks from intent-focused research, but also provides a new framework for generating, validating, and applying other kinds of taxonomies in a scalable and adaptable way with reasonable human effort.

Dong-Ho Lee, Jay Pujara, Mohit Sewak et al.

Although large language models (LLMs) have advanced the state-of-the-art in NLP significantly, deploying them for downstream applications is still challenging due to cost, responsiveness, control, or concerns around privacy and security. As such, trainable models are still the preferred option in some cases. However, these models still require human-labeled data for optimal performance, which is expensive and time-consuming to obtain. In order to address this issue, several techniques to reduce human effort involve labeling or generating data using LLMs. Although these methods are effective for certain applications, in practice they encounter difficulties in real-world scenarios. Labeling data requires careful data selection, while generating data necessitates task-specific prompt engineering. In this paper, we propose a unified data creation pipeline that requires only a single formatting example, and which is applicable to a broad range of tasks, including traditionally problematic ones with semantically devoid label spaces. In our experiments we demonstrate that instruction-following LLMs are highly cost-effective data creators, and that models trained with these data exhibit performance better than those trained with human-labeled data (by up to 17.5%) on out-of-distribution evaluation, while maintaining comparable performance on in-distribution tasks. These results have important implications for the robustness of NLP systems deployed in the real-world.

Ryen W. White

As many of us in the information retrieval (IR) research community know and appreciate, search is far from being a solved problem. Millions of people struggle with tasks on search engines every day. Often, their struggles relate to the intrinsic complexity of their task and the failure of search systems to fully understand the task and serve relevant results. The task motivates the search, creating the gap/problematic situation that searchers attempt to bridge/resolve and drives search behavior as they work through different task facets. Complex search tasks require more than support for rudimentary fact finding or re-finding. Research on methods to support complex tasks includes work on generating query and website suggestions, personalizing and contextualizing search, and developing new search experiences, including those that span time and space. The recent emergence of generative artificial intelligence (AI) and the arrival of assistive agents, based on this technology, has the potential to offer further assistance to searchers, especially those engaged in complex tasks. There are profound implications from these advances for the design of intelligent systems and for the future of search itself. This article, based on a keynote by the author at the 2023 ACM SIGIR Conference, explores these issues and how AI agents are advancing the frontier of search system capabilities, with a special focus on information interaction and complex task completion.

Harshita Chopra, Krishna Kant Chintalapudi, Suman Nath et al.

Long-horizon personalization requires dialogue assistants to retrieve user-specific facts from extended interaction histories. In practice, many relevant facts often have low semanticsimilarity to the query under dense retrieval. Standard Retrieval-Augmented Generation (RAG) and GraphRAG systems are still largely retrospective: they rely on embedding similarity to the query or on fixed graph traversals, so they often miss facts that matter for the user's needs but lie far from the query in embedding space. Inspired by prospection, the human ability to use imagined futures as cues for recall, we introduce Prospection-Guided Retrieval (PGR), which decouples retrieval from how memories are stored. Given a user query, PGR first expands the goal into a short Tree-of-Thought (ToT) or linear chain of plausible next steps, and uses these steps as retrieval probes rather than relying on the original query alone. The facts retrieved by these probes are then used to personalize the next round of prospection, enabling PGR to uncover additional memories that become relevant only after the simulation is grounded in the user's history. We also introduce MemoryQuest, a challenging multi-session benchmark in which each query is annotated with 3--5 dated reference facts subject to a low query-reference similarity constraint. Across 1,625 queries spanning 185 user profiles from 3 publicly available datasets, PGR-TOT substantially improves retrieval, including nearly 3x recall on MemoryQuest over the strongest baseline. In pairwise LLM-as-judge comparisons against baselines, PGR-generated responses are preferred on 89--98% of queries, with blinded human annotations on held-out subsets showing the same trend. Overall, the results demonstrate that explicit prospection yields large gains in long-horizon retrieval and response quality relative to similarity-only baselines.

Chirag Shah, Ryen W. White

In the midst of the growing integration of Artificial Intelligence (AI) into various aspects of our lives, agents are experiencing a resurgence. These autonomous programs that act on behalf of humans are neither new nor exclusive to the mainstream AI movement. By exploring past incarnations of agents, we can understand what has been done previously, what worked, and more importantly, what did not pan out and why. This understanding lets us to examine what distinguishes the current focus on agents. While generative AI is appealing, this technology alone is insufficient to make new generations of agents more successful. To make the current wave of agents effective and sustainable, we envision an ecosystem that includes not only agents but also Sims, which represent user preferences and behaviors, as well as Assistants, which directly interact with the user and coordinate the execution of user tasks with the help of the agents.

Jiqun Liu, Jamshed Karimnazarov, Ryen W. White

Functional fixedness, a cognitive bias that restricts users' interactions with a new system or tool to expected or familiar ways, limits the full potential of Large Language Model (LLM)-enabled chat search, especially in complex and exploratory tasks. To investigate its impact, we conducted a crowdsourcing study with 450 participants, each completing one of six decision-making tasks spanning public safety, diet and health management, sustainability, and AI ethics. Participants engaged in a multi-prompt conversation with ChatGPT to address the task, allowing us to compare pre-chat intent-based expectations with observed interactions. We found that: 1) Several aspects of pre-chat expectations are closely associated with users' prior experiences with ChatGPT, search engines, and virtual assistants; 2) Prior system experience shapes language use and prompting behavior. Frequent ChatGPT users reduced deictic terms and hedge words and frequently adjusted prompts. Users with rich search experience maintained structured, less-conversational queries with minimal modifications. Users of virtual assistants favored directive, command-like prompts, reinforcing functional fixedness; 3) When the system failed to meet expectations, participants generated more detailed prompts with increased linguistic diversity, reflecting adaptive shifts. These findings suggest that while preconceived expectations constrain early interactions, unmet expectations can motivate behavioral adaptation. With appropriate system support, this may promote broader exploration of LLM capabilities. This work also introduces a typology for user intents in chat search and highlights the importance of mitigating functional fixedness to support more creative and analytical use of LLMs.

Chirag Shah, Ryen W. White

The emergence of generative artificial intelligence (GenAI) is transforming information interaction. For decades, search engines such as Google and Bing have been the primary means of locating relevant information for the general population. They have provided search results in the same standard format (the so-called "10 blue links"). The recent ability to chat via natural language with AI-based agents and have GenAI automatically synthesize answers in real-time (grounded in top-ranked results) is changing how people interact with and consume information at massive scale. These two information interaction modalities (traditional search and AI-powered chat) coexist in current search engines, either loosely coupled (e.g., as separate options/tabs) or tightly coupled (e.g., integrated as a chat answer embedded directly within a traditional search result page). We believe that the existence of these two different modalities, and potentially many others, is creating an opportunity to re-imagine the search experience, capitalize on the strengths of many modalities, and develop systems and strategies to support seamless flow between them. We refer to these as panmodal experiences. Unlike monomodal experiences, where only one modality is available and/or used for the task at hand, panmodal experiences make multiple modalities available to users (multimodal), directly support transitions between modalities (crossmodal), and seamlessly combine modalities to tailor task assistance (transmodal). While our focus is search and chat, with learnings from insights from a survey of over 100 individuals who have recently performed common tasks on these two modalities, we also present a more general vision for the future of information interaction using multiple modalities and the emergent capabilities of GenAI.

Jin Huang, Silviu Cucerzan, Sujay Kumar Jauhar et al. · microsoft-research

Large language models (LLMs) have demonstrated significant potential to accelerate scientific discovery as valuable tools for analyzing data, generating hypotheses, and supporting innovative approaches in various scientific fields. In this work, we investigate how LLMs can handle the transition from conceptual research ideas to well-structured research plans. Effective research planning not only supports scientists in advancing their research but also represents a crucial capability for the development of autonomous research agents. Despite its importance, the field lacks a systematic understanding of LLMs' research planning capability. To rigorously measure this capability, we introduce the Idea2Plan task and Idea2Plan Bench, a benchmark built from 200 ICML 2025 Spotlight and Oral papers released after major LLM training cutoffs. Each benchmark instance includes a research idea and a grading rubric capturing the key components of valid plans. We further propose Idea2Plan JudgeEval, a complementary benchmark to assess the reliability of LLM-based judges against expert annotations. Experimental results show that GPT-5 and GPT-5-mini achieve the strongest performance on the benchmark, though substantial headroom remains for future improvement. Our study provides new insights into LLMs' capability for research planning and lay the groundwork for future progress.

Siddharth Suri, Scott Counts, Leijie Wang et al.

Until recently, search engines were the predominant method for people to access online information. The recent emergence of large language models (LLMs) has given machines new capabilities such as the ability to generate new digital artifacts like text, images, code etc., resulting in a new tool, a generative search engine, which combines the capabilities of LLMs with a traditional search engine. Through the empirical analysis of Bing Copilot (Bing Chat), one of the first publicly available generative search engines, we analyze the types and complexity of tasks that people use Bing Copilot for compared to Bing Search. Findings indicate that people use the generative search engine for more knowledge work tasks that are higher in cognitive complexity than were commonly done with a traditional search engine.

Sujay Kumar Jauhar, Nirupama Chandrasekaran, Michael Gamon et al.

Tasks are a fundamental unit of work in the daily lives of people, who are increasingly using digital means to keep track of, organize, triage and act on them. These digital tools -- such as task management applications -- provide a unique opportunity to study and understand tasks and their connection to the real world, and through intelligent assistance, help people be more productive. By logging signals such as text, timestamp information, and social connectivity graphs, an increasingly rich and detailed picture of how tasks are created and organized, what makes them important, and who acts on them, can be progressively developed. Yet the context around actual task completion remains fuzzy, due to the basic disconnect between actions taken in the real world and telemetry recorded in the digital world. Thus, in this paper we compile and release a novel, real-life, large-scale dataset called MS-LaTTE that captures two core aspects of the context surrounding task completion: location and time. We describe our annotation framework and conduct a number of analyses on the data that were collected, demonstrating that it captures intuitive contextual properties for common tasks. Finally, we test the dataset on the two problems of predicting spatial and temporal task co-occurrence, concluding that predictors for co-location and co-time are both learnable, with a BERT fine-tuned model outperforming several other baselines. The MS-LaTTE dataset provides an opportunity to tackle many new modeling challenges in contextual task understanding and we hope that its release will spur future research in task intelligence more broadly.

Yonchanok Khaokaew, Mohammad Saiedur Rahaman, Ryen W. White et al.

App usage prediction is important for smartphone system optimization to enhance user experience. Existing modeling approaches utilize historical app usage logs along with a wide range of semantic information to predict the app usage; however, they are only effective in certain scenarios and cannot be generalized across different situations. This paper address this problem by developing a model called Contextual and Semantic Embedding model for App Usage Prediction (CoSEM) for app usage prediction that leverages integration of 1) semantic information embedding and 2) contextual information embedding based on historical app usage of individuals. Extensive experiments show that the combination of semantic information and history app usage information enables our model to outperform the baselines on three real-world datasets, achieving an MRR score over 0.55,0.57,0.86 and Hit rate scores of more than 0.71, 0.75, and 0.95, respectively.

Kristina Gligorić, Ryen W. White, Emre Kıcıman et al.

Nutrition is a key determinant of long-term health, and social influence has long been theorized to be a key determinant of nutrition. It has been difficult to quantify the postulated role of social influence on nutrition using traditional methods such as surveys, due to the typically small scale and short duration of studies. To overcome these limitations, we leverage a novel source of data: logs of 38 million food purchases made over an 8-year period on the Ecole Polytechnique Federale de Lausanne (EPFL) university campus, linked to anonymized individuals via the smartcards used to make on-campus purchases. In a longitudinal observational study, we ask: How is a person's food choice affected by eating with someone else whose own food choice is healthy vs. unhealthy? To estimate causal effects from the passively observed log data, we control confounds in a matched quasi-experimental design: we identify focal users who at first do not have any regular eating partners but then start eating with a fixed partner regularly, and we match focal users into comparison pairs such that paired users are nearly identical with respect to covariates measured before acquiring the partner, where the two focal users' new eating partners diverge in the healthiness of their respective food choice. A difference-in-differences analysis of the paired data yields clear evidence of social influence: focal users acquiring a healthy-eating partner change their habits significantly more toward healthy foods than focal users acquiring an unhealthy-eating partner. We further identify foods whose purchase frequency is impacted significantly by the eating partner's healthiness of food choice. Beyond the main results, the work demonstrates the utility of passively sensed food purchase logs for deriving insights, with the potential of informing the design of public health interventions and food offerings.

Jina Suh, Eric Horvitz, Ryen W. White et al.

Most work to date on mitigating the COVID-19 pandemic is focused urgently on biomedicine and epidemiology. Yet, pandemic-related policy decisions cannot be made on health information alone. Decisions need to consider the broader impacts on people and their needs. Quantifying human needs across the population is challenging as it requires high geo-temporal granularity, high coverage across the population, and appropriate adjustment for seasonal and other external effects. Here, we propose a computational methodology, building on Maslow's hierarchy of needs, that can capture a holistic view of relative changes in needs following the pandemic through a difference-in-differences approach that corrects for seasonality and volume variations. We apply this approach to characterize changes in human needs across physiological, socioeconomic, and psychological realms in the US, based on more than 35 billion search interactions spanning over 36,000 ZIP codes over a period of 14 months. The analyses reveal that the expression of basic human needs has increased exponentially while higher-level aspirations declined during the pandemic in comparison to the pre-pandemic period. In exploring the timing and variations in statewide policies, we find that the durations of shelter-in-place mandates have influenced social and emotional needs significantly. We demonstrate that potential barriers to addressing critical needs, such as support for unemployment and domestic violence, can be identified through web search interactions. Our approach and results suggest that population-scale monitoring of shifts in human needs can inform policies and recovery efforts for current and anticipated needs.

Ziming Li, Julia Kiseleva, Alekh Agarwal et al.

Effective optimization is essential for real-world interactive systems to provide a satisfactory user experience in response to changing user behavior. However, it is often challenging to find an objective to optimize for interactive systems (e.g., policy learning in task-oriented dialog systems). Generally, such objectives are manually crafted and rarely capture complex user needs in an accurate manner. We propose an approach that infers the objective directly from observed user interactions. These inferences can be made regardless of prior knowledge and across different types of user behavior. We introduce Interactive System Optimizer (ISO), a novel algorithm that uses these inferred objectives for optimization. Our main contribution is a new general principled approach to optimizing interactive systems using data-driven objectives. We demonstrate the high effectiveness of ISO over several simulations.

Maartje ter Hoeve, Robert Sim, Elnaz Nouri et al.

The role of conversational assistants has become more prevalent in helping people increase their productivity. Document-centered assistance, for example to help an individual quickly review a document, has seen less significant progress, even though it has the potential to tremendously increase a user's productivity. This type of document-centered assistance is the focus of this paper. Our contributions are three-fold: (1) We first present a survey to understand the space of document-centered assistance and the capabilities people expect in this scenario. (2) We investigate the types of queries that users will pose while seeking assistance with documents, and show that document-centered questions form the majority of these queries. (3) We present a set of initial machine learned models that show that (a) we can accurately detect document-centered questions, and (b) we can build reasonably accurate models for answering such questions. These positive results are encouraging, and suggest that even greater results may be attained with continued study of this interesting and novel problem space. Our findings have implications for the design of intelligent systems to support task completion via natural interactions with documents.

George Philipp, Ryen W. White

The Web is an important resource for understanding and diagnosing medical conditions. Based on exposure to online content, people may develop undue health concerns, believing that common and benign symptoms are explained by serious illnesses. In this paper, we investigate potential strategies to mine queries and searcher histories for clues that could help search engines choose the most appropriate information to present in response to exploratory medical queries. To do this, we performed a longitudinal study of health search behavior using the logs of a popular search engine. We found that query variations which might appear innocuous (e.g. "bad headache" vs "severe headache") may hold valuable information about the searcher which could be used by search engines to improve performance. Furthermore, we investigated how medically concerned users respond differently to search engine result pages (SERPs) and find that their disposition for clicking on concerning pages is pronounced, potentially leading to a self-reinforcement of concern. Finally, we studied to which degree variations in the SERP impact future search and real-world health-seeking behavior and obtained some surprising results (e.g., viewing concerning pages may lead to a short-term reduction of real-world health seeking).

Tim Althoff, Eric Horvitz, Ryen W. White et al.

Human cognitive performance is critical to productivity, learning, and accident avoidance. Cognitive performance varies throughout each day and is in part driven by intrinsic, near 24-hour circadian rhythms. Prior research on the impact of sleep and circadian rhythms on cognitive performance has typically been restricted to small-scale laboratory-based studies that do not capture the variability of real-world conditions, such as environmental factors, motivation, and sleep patterns in real-world settings. Given these limitations, leading sleep researchers have called for larger in situ monitoring of sleep and performance. We present the largest study to date on the impact of objectively measured real-world sleep on performance enabled through a reframing of everyday interactions with a web search engine as a series of performance tasks. Our analysis includes 3 million nights of sleep and 75 million interaction tasks. We measure cognitive performance through the speed of keystroke and click interactions on a web search engine and correlate them to wearable device-defined sleep measures over time. We demonstrate that real-world performance varies throughout the day and is influenced by both circadian rhythms, chronotype (morning/evening preference), and prior sleep duration and timing. We develop a statistical model that operationalizes a large body of work on sleep and performance and demonstrates that our estimates of circadian rhythms, homeostatic sleep drive, and sleep inertia align with expectations from laboratory-based sleep studies. Further, we quantify the impact of insufficient sleep on real-world performance and show that two consecutive nights with less than six hours of sleep are associated with decreases in performance which last for a period of six days. This work demonstrates the feasibility of using online interactions for large-scale physiological sensing.

Tim Althoff, Ryen W. White, Eric Horvitz

Physical activity helps people maintain a healthy weight and reduces the risk for several chronic diseases. Although this knowledge is widely recognized, adults and children in many countries around the world do not get recommended amounts of physical activity. While many interventions are found to be ineffective at increasing physical activity or reaching inactive populations, there have been anecdotal reports of increased physical activity due to novel mobile games that embed game play in the physical world. The most recent and salient example of such a game is Pokémon Go, which has reportedly reached tens of millions of users in the US and worldwide. We study the effect of Pokémon Go on physical activity through a combination of signals from large-scale corpora of wearable sensor data and search engine logs for 32 thousand users over a period of three months. Pokémon Go players are identified through search engine queries and activity is measured through accelerometry. We find that Pokémon Go leads to significant increases in physical activity over a period of 30 days, with particularly engaged users (i.e., those making multiple search queries for details about game usage) increasing their activity by 1473 steps a day on average, a more than 25% increase compared to their prior activity level ($p<10^{-15}$). In the short time span of the study, we estimate that Pokémon Go has added a total of 144 billion steps to US physical activity. Furthermore, Pokémon Go has been able to increase physical activity across men and women of all ages, weight status, and prior activity levels showing this form of game leads to increases in physical activity with significant implications for public health. We find that Pokémon Go is able to reach low activity populations while all four leading mobile health apps studied in this work largely draw from an already very active population.

Robert West, Ryen W. White, Eric Horvitz

Nutrition is a key factor in people's overall health. Hence, understanding the nature and dynamics of population-wide dietary preferences over time and space can be valuable in public health. To date, studies have leveraged small samples of participants via food intake logs or treatment data. We propose a complementary source of population data on nutrition obtained via Web logs. Our main contribution is a spatiotemporal analysis of population-wide dietary preferences through the lens of logs gathered by a widely distributed Web-browser add-on, using the access volume of recipes that users seek via search as a proxy for actual food consumption. We discover that variation in dietary preferences as expressed via recipe access has two main periodic components, one yearly and the other weekly, and that there exist characteristic regional differences in terms of diet within the United States. In a second study, we identify users who show evidence of having made an acute decision to lose weight. We characterize the shifts in interests that they express in their search queries and focus on changes in their recipe queries in particular. Last, we correlate nutritional time series obtained from recipe queries with time-aligned data on hospital admissions, aimed at understanding how behavioral data captured in Web logs might be harnessed to identify potential relationships between diet and acute health problems. In this preliminary study, we focus on patterns of sodium identified in recipes over time and patterns of admission for congestive heart failure, a chronic illness that can be exacerbated by increases in sodium intake.