Aaron Halfaker

Yixin Liu, Budhaditya Deb, Milagro Teruel et al. · microsoft-research

Despite the recent progress in language generation models, their outputs may not always meet user expectations. In this work, we study whether informational feedback in natural language can be leveraged to improve generation quality and user preference alignment. To this end, we consider factual consistency in summarization, the quality that the summary should only contain information supported by the input documents, as the user-expected preference. We collect a high-quality dataset, DeFacto, containing human demonstrations and informational natural language feedback consisting of corrective instructions, edited summaries, and explanations with respect to the factual consistency of the summary. Using our dataset, we study three natural language generation tasks: (1) editing a summary by following the human feedback, (2) generating human feedback for editing the original summary, and (3) revising the initial summary to correct factual errors by generating both the human feedback and edited summary. We show that DeFacto can provide factually consistent human-edited summaries and further insights into summarization factual consistency thanks to its informational natural language feedback. We further demonstrate that fine-tuned language models can leverage our dataset to improve the summary factual consistency, while large language models lack the zero-shot learning ability in our proposed tasks that require controllable text generation.

Sumit Asthana, Sagih Hilleli, Pengcheng He et al. · microsoft-research

Meetings play a critical infrastructural role in coordinating work. The recent surge of hybrid and remote meetings in computer-mediated spaces has led to new problems (e.g., more time spent in less engaging meetings) and new opportunities (e.g., automated transcription/captioning and recap support). Advances in dialogue summarization offer the potential for improving post-meeting experiences, but fixed-length summaries often fail to meet diverse needs, such as quick overviews or detailed insights. To address these gaps, we use cognitive science and discourse theories to conceptualize two recap designs: important highlights and a structured, hierarchical minutes view, targeting complementary recap needs. We operationalize these representations into high-fidelity prototypes using dialogue summarization. Finally, we evaluate the representations' effectiveness with seven users in the context of their work meetings at Microsoft. Our results show both recap types are valuable in different contexts, enabling collaboration through discussions and consensus-building. Exploring the meaning of users adding, editing, and deleting from recaps suggests varying alignment for using these actions to improve AI-recap. Our design implications, such as incorporating organizational artifacts (e.g., linking presentations) in recaps and personalizing context, advance the discourse of effective recap designs for organizational work and support past results from cognition studies.

Abhinav Java, Ashmit Khandelwal, Sukruta Midigeshi et al.

Information tasks such as writing surveys or analytical reports require complex search and reasoning, and have recently been grouped under the umbrella of \textit{deep research} -- a term also adopted by recent models targeting these capabilities. Despite growing interest, the scope of the deep research task remains underdefined and its distinction from other reasoning-intensive problems is poorly understood. In this paper, we propose a formal characterization of the deep research (DR) task and introduce a benchmark to evaluate the performance of DR systems. We argue that the core defining feature of deep research is not the production of lengthy report-style outputs, but rather the high fan-out over concepts required during the search process, i.e., broad and reasoning-intensive exploration. To enable objective evaluation, we define DR using an intermediate output representation that encodes key claims uncovered during search-separating the reasoning challenge from surface-level report generation. Based on this formulation, we propose a diverse, challenging benchmark LiveDRBench with 100 challenging tasks over scientific topics (e.g., datasets, materials discovery, prior art search) and public interest events (e.g., flight incidents, movie awards). Across state-of-the-art DR systems, F1 score ranges between 0.02 and 0.72 for any sub-category. OpenAI's model performs the best with an overall F1 score of 0.55. Analysis of reasoning traces reveals the distribution over the number of referenced sources, branching, and backtracking events executed by current DR systems, motivating future directions for improving their search mechanisms and grounding capabilities. The benchmark is available at https://github.com/microsoft/LiveDRBench.

Tzu-Sheng Kuo, Aaron Halfaker, Zirui Cheng et al.

AI tools are increasingly deployed in community contexts. However, datasets used to evaluate AI are typically created by developers and annotators outside a given community, which can yield misleading conclusions about AI performance. How might we empower communities to drive the intentional design and curation of evaluation datasets for AI that impacts them? We investigate this question on Wikipedia, an online community with multiple AI-based content moderation tools deployed. We introduce Wikibench, a system that enables communities to collaboratively curate AI evaluation datasets, while navigating ambiguities and differences in perspective through discussion. A field study on Wikipedia shows that datasets curated using Wikibench can effectively capture community consensus, disagreement, and uncertainty. Furthermore, study participants used Wikibench to shape the overall data curation process, including refining label definitions, determining data inclusion criteria, and authoring data statements. Based on our findings, we propose future directions for systems that support community-driven data curation.

Nathan TeBlunthuis, Benjamin Mako Hill, Aaron Halfaker

Online community moderators often rely on social signals such as whether or not a user has an account or a profile page as clues that users may cause problems. Reliance on these clues can lead to "overprofiling'' bias when moderators focus on these signals but overlook the misbehavior of others. We propose that algorithmic flagging systems deployed to improve the efficiency of moderation work can also make moderation actions more fair to these users by reducing reliance on social signals and making norm violations by everyone else more visible. We analyze moderator behavior in Wikipedia as mediated by RCFilters, a system which displays social signals and algorithmic flags, and estimate the causal effect of being flagged on moderator actions. We show that algorithmically flagged edits are reverted more often, especially those by established editors with positive social signals, and that flagging decreases the likelihood that moderation actions will be undone. Our results suggest that algorithmic flagging systems can lead to increased fairness in some contexts but that the relationship is complex and contingent.

C. Estelle Smith, Bowen Yu, Anjali Srivastava et al.

On Wikipedia, sophisticated algorithmic tools are used to assess the quality of edits and take corrective actions. However, algorithms can fail to solve the problems they were designed for if they conflict with the values of communities who use them. In this study, we take a Value-Sensitive Algorithm Design approach to understanding a community-created and -maintained machine learning-based algorithm called the Objective Revision Evaluation System (ORES)---a quality prediction system used in numerous Wikipedia applications and contexts. Five major values converged across stakeholder groups that ORES (and its dependent applications) should: (1) reduce the effort of community maintenance, (2) maintain human judgement as the final authority, (3) support differing peoples' differing workflows, (4) encourage positive engagement with diverse editor groups, and (5) establish trustworthiness of people and algorithms within the community. We reveal tensions between these values and discuss implications for future research to improve algorithms like ORES.

Aaron Halfaker, R. Stuart Geiger

Algorithmic systems---from rule-based bots to machine learning classifiers---have a long history of supporting the essential work of content moderation and other curation work in peer production projects. From counter-vandalism to task routing, basic machine prediction has allowed open knowledge projects like Wikipedia to scale to the largest encyclopedia in the world, while maintaining quality and consistency. However, conversations about how quality control should work and what role algorithms should play have generally been led by the expert engineers who have the skills and resources to develop and modify these complex algorithmic systems. In this paper, we describe ORES: an algorithmic scoring service that supports real-time scoring of wiki edits using multiple independent classifiers trained on different datasets. ORES decouples several activities that have typically all been performed by engineers: choosing or curating training data, building models to serve predictions, auditing predictions, and developing interfaces or automated agents that act on those predictions. This meta-algorithmic system was designed to open up socio-technical conversations about algorithms in Wikipedia to a broader set of participants. In this paper, we discuss the theoretical mechanisms of social change ORES enables and detail case studies in participatory machine learning around ORES from the 5 years since its deployment.

Isaac Johnson, Allen Yilun Lin, Toby Jia-Jun Li et al.

Wikipedia articles about places, OpenStreetMap features, and other forms of peer-produced content have become critical sources of geographic knowledge for humans and intelligent technologies. In this paper, we explore the effectiveness of the peer production model across the rural/urban divide, a divide that has been shown to be an important factor in many online social systems. We find that in both Wikipedia and OpenStreetMap, peer-produced content about rural areas is of systematically lower quality, is less likely to have been produced by contributors who focus on the local area, and is more likely to have been generated by automated software agents (i.e. bots). We then codify the systemic challenges inherent to characterizing rural phenomena through peer production and discuss potential solutions.

Christoph Kinkeldey, Claudia Müller-Birn, Tom Gülenman et al.

Machine learning systems are ubiquitous in various kinds of digital applications and have a huge impact on our everyday life. But a lack of explainability and interpretability of such systems hinders meaningful participation by people, especially by those without a technical background. Interactive visual interfaces (e.g., providing means for manipulating parameters in the user interface) can help tackle this challenge. In this paper we present PreCall, an interactive visual interface for ORES, a machine learning-based web service for Wikimedia projects such as Wikipedia. While ORES can be used for a number of settings, it can be challenging to translate requirements from the application domain into formal parameter sets needed to configure the ORES models. Assisting Wikipedia editors in finding damaging edits, for example, can be realized at various stages of automatization, which might impact the precision of the applied model. Our prototype PreCall attempts to close this translation gap by interactively visualizing the relationship between major model metrics (recall, precision, false positive rate) and a parameter (the threshold between valuable and damaging edits). Furthermore, PreCall visualizes the probable results for the current model configuration to improve the human's understanding of the relationship between metrics and outcome when using ORES. We describe PreCall's components and present a use case that highlights the benefits of our approach. Finally, we pose further research questions we would like to discuss during the workshop.

R. Stuart Geiger, Aaron Halfaker

This paper replicates, extends, and refutes conclusions made in a study published in PLoS ONE ("Even Good Bots Fight"), which claimed to identify substantial levels of conflict between automated software agents (or bots) in Wikipedia using purely quantitative methods. By applying an integrative mixed-methods approach drawing on trace ethnography, we place these alleged cases of bot-bot conflict into context and arrive at a better understanding of these interactions. We found that overwhelmingly, the interactions previously characterized as problematic instances of conflict are typically better characterized as routine, productive, even collaborative work. These results challenge past work and show the importance of qualitative/quantitative collaboration. In our paper, we present quantitative metrics and qualitative heuristics for operationalizing bot-bot conflict. We give thick descriptions of kinds of events that present as bot-bot reverts, helping distinguish conflict from non-conflict. We computationally classify these kinds of events through patterns in edit summaries. By interpreting found/trace data in the socio-technical contexts in which people give that data meaning, we gain more from quantitative measurements, drawing deeper understandings about the governance of algorithmic systems in Wikipedia. We have also released our data collection, processing, and analysis pipeline, to facilitate computational reproducibility of our findings and to help other researchers interested in conducting similar mixed-method scholarship in other platforms and contexts.

Amir Sarabadani, Aaron Halfaker, Dario Taraborelli

Wikidata, like Wikipedia, is a knowledge base that anyone can edit. This open collaboration model is powerful in that it reduces barriers to participation and allows a large number of people to contribute. However, it exposes the knowledge base to the risk of vandalism and low-quality contributions. In this work, we build on past work detecting vandalism in Wikipedia to detect vandalism in Wikidata. This work is novel in that identifying damaging changes in a structured knowledge-base requires substantially different feature engineering work than in a text-based wiki like Wikipedia. We also discuss the utility of these classifiers for reducing the overall workload of vandalism patrollers in Wikidata. We describe a machine classification strategy that is able to catch 89% of vandalism while reducing patrollers' workload by 98%, by drawing lightly from contextual features of an edit and heavily from the characteristics of the user making the edit.

Aaron Halfaker, Os Keyes, Daniel Kluver et al.

Session identification is a common strategy used to develop metrics for web analytics and behavioral analyses of user-facing systems. Past work has argued that session identification strategies based on an inactivity threshold is inherently arbitrary or advocated that thresholds be set at about 30 minutes. In this work, we demonstrate a strong regularity in the temporal rhythms of user initiated events across several different domains of online activity (incl. video gaming, search, page views and volunteer contributions). We describe a methodology for identifying clusters of user activity and argue that regularity with which these activity clusters appear implies a good rule-of-thumb inactivity threshold of about 1 hour. We conclude with implications that these temporal rhythms may have for system design based on our observations and theories of goal-directed human activity.