Matthew Lease

William Rudman, Abhishek Divekar, Kanishk Jain et al. · amazon-science

Fitting quantitative models to data is a central step in scientific workflows, yet it remains one of the least automated. Recent agent-based systems leverage language and vision-language models (VLMs) to iteratively propose and refine statistical models, but these systems struggle on more challenging modeling tasks. To address these limitations, we introduce VESTA: Visual Exploration with Statistical Tool Agents, a framework that equips VLMs with a dynamically growing exploration toolkit to guide model refinement through data transformations, hypothesis-driven visualizations, and robust statistical tests. Unlike prior systems that rely on iterative critique alone, VESTA actively explores data before and during refinement by selecting or creating diagnostic tools, which accumulate in the model's context and can be reused later. We evaluate VESTA against established baselines in three toolkit configurations: no tools, static expert-written tools, and dynamic model-written tools. To support this evaluation, we introduce DAWN (Dataset for Automated Workflows and Numerical Modeling), a benchmark targeting distribution fitting and time series modeling with varying difficulty tiers, and culminating in real-world astronomy tasks including modeling initial mass functions and gravitational-wave chirp signals. We find that VESTA's dynamic tool creation outperforms prior agentic pipelines, with the largest gains on complex and domain-specific tasks. We further show that dynamically generated tools are substantially more sophisticated than those produced by existing visual tool-creation systems, covering more diagnostic categories per function and strongly preferring visual outputs that the VLM critic can reason over directly.

Luis Oala, Manil Maskey, Lilith Bat-Leah et al. · mit

Drawing from discussions at the inaugural DMLR workshop at ICML 2023 and meetings prior, in this report we outline the relevance of community engagement and infrastructure development for the creation of next-generation public datasets that will advance machine learning science. We chart a path forward as a collective effort to sustain the creation and maintenance of these datasets and methods towards positive scientific, societal and business impact.

Anubrata Das, Chitrank Gupta, Venelin Kovatchev et al.

We present ProtoTEx, a novel white-box NLP classification architecture based on prototype networks. ProtoTEx faithfully explains model decisions based on prototype tensors that encode latent clusters of training examples. At inference time, classification decisions are based on the distances between the input text and the prototype tensors, explained via the training examples most similar to the most influential prototypes. We also describe a novel interleaved training algorithm that effectively handles classes characterized by the absence of indicative features. On a propaganda detection task, ProtoTEx accuracy matches BART-large and exceeds BERT-large with the added benefit of providing faithful explanations. A user study also shows that prototype-based explanations help non-experts to better recognize propaganda in online news.

Alexander Braylan, Omar Alonso, Matthew Lease

When annotators label data, a key metric for quality assurance is inter-annotator agreement (IAA): the extent to which annotators agree on their labels. Though many IAA measures exist for simple categorical and ordinal labeling tasks, relatively little work has considered more complex labeling tasks, such as structured, multi-object, and free-text annotations. Krippendorff's alpha, best known for use with simpler labeling tasks, does have a distance-based formulation with broader applicability, but little work has studied its efficacy and consistency across complex annotation tasks. We investigate the design and evaluation of IAA measures for complex annotation tasks, with evaluation spanning seven diverse tasks: image bounding boxes, image keypoints, text sequence tagging, ranked lists, free text translations, numeric vectors, and syntax trees. We identify the difficulty of interpretability and the complexity of choosing a distance function as key obstacles in applying Krippendorff's alpha generally across these tasks. We propose two novel, more interpretable measures, showing they yield more consistent IAA measures across tasks and annotation distance functions.

Soumyajit Gupta, Sooyong Lee, Maria De-Arteaga et al.

Algorithmic bias often arises as a result of differential subgroup validity, in which predictive relationships vary across groups. For example, in toxic language detection, comments targeting different demographic groups can vary markedly across groups. In such settings, trained models can be dominated by the relationships that best fit the majority group, leading to disparate performance. We propose framing toxicity detection as multi-task learning (MTL), allowing a model to specialize on the relationships that are relevant to each demographic group while also leveraging shared properties across groups. With toxicity detection, each task corresponds to identifying toxicity against a particular demographic group. However, traditional MTL requires labels for all tasks to be present for every data point. To address this, we propose Conditional MTL (CondMTL), wherein only training examples relevant to the given demographic group are considered by the loss function. This lets us learn group specific representations in each branch which are not cross contaminated by irrelevant labels. Results on synthetic and real data show that using CondMTL improves predictive recall over various baselines in general and for the minority demographic group in particular, while having similar overall accuracy.

Venelin Kovatchev, Trina Chatterjee, Venkata S Govindarajan et al.

Developing methods to adversarially challenge NLP systems is a promising avenue for improving both model performance and interpretability. Here, we describe the approach of the team "longhorns" on Task 1 of the The First Workshop on Dynamic Adversarial Data Collection (DADC), which asked teams to manually fool a model on an Extractive Question Answering task. Our team finished first, with a model error rate of 62%. We advocate for a systematic, linguistically informed approach to formulating adversarial questions, and we describe the results of our pilot experiments, as well as our official submission.

Anubrata Das, Houjiang Liu, Venelin Kovatchev et al.

Misinformation threatens modern society by promoting distrust in science, changing narratives in public health, heightening social polarization, and disrupting democratic elections and financial markets, among a myriad of other societal harms. To address this, a growing cadre of professional fact-checkers and journalists provide high-quality investigations into purported facts. However, these largely manual efforts have struggled to match the enormous scale of the problem. In response, a growing body of Natural Language Processing (NLP) technologies have been proposed for more scalable fact-checking. Despite tremendous growth in such research, however, practical adoption of NLP technologies for fact-checking still remains in its infancy today. In this work, we review the capabilities and limitations of the current NLP technologies for fact-checking. Our particular focus is to further chart the design space for how these technologies can be harnessed and refined in order to better meet the needs of human fact-checkers. To do so, we review key aspects of NLP-based fact-checking: task formulation, dataset construction, modeling, and human-centered strategies, such as explainable models and human-in-the-loop approaches. Next, we review the efficacy of applying NLP-based fact-checking tools to assist human fact-checkers. We recommend that future research include collaboration with fact-checker stakeholders early on in NLP research, as well as incorporation of human-centered design practices in model development, in order to further guide technology development for human use and practical adoption. Finally, we advocate for more research on benchmark development supporting extrinsic evaluation of human-centered fact-checking technologies.

Joydeep Biswas, Sheila Schoepp, Gautham Vasan et al.

Scientific peer review faces mounting strain as submission volumes surge, making it increasingly difficult to sustain review quality, consistency, and timeliness. Recent advances in AI have led the community to consider its use in peer review, yet a key unresolved question is whether AI can generate technically sound reviews at real-world conference scale. Here we report the first large-scale field deployment of AI-assisted peer review: every main-track submission at AAAI-26 received one clearly identified AI review from a state-of-the-art system. The system combined frontier models, tool use, and safeguards in a multi-stage process to generate reviews for all 22,977 full-review papers in less than a day. A large-scale survey of AAAI-26 authors and program committee members showed that participants not only found AI reviews useful, but actually preferred them to human reviews on key dimensions such as technical accuracy and research suggestions. We also introduce a novel benchmark and find that our system substantially outperforms a simple LLM-generated review baseline at detecting a variety of scientific weaknesses. Together, these results show that state-of-the-art AI methods can already make meaningful contributions to scientific peer review at conference scale, opening a path toward the next generation of synergistic human-AI teaming for evaluating research.

Soumyajit Gupta, Venelin Kovatchev, Anubrata Das et al.

Optimizing NLP models for fairness poses many challenges. Lack of differentiable fairness measures prevents gradient-based loss training or requires surrogate losses that diverge from the true metric of interest. In addition, competing objectives (e.g., accuracy vs. fairness) often require making trade-offs based on stakeholder preferences, but stakeholders may not know their preferences before seeing system performance under different trade-off settings. To address these challenges, we begin by formulating a differentiable version of a popular fairness measure, Accuracy Parity, to provide balanced accuracy across demographic groups. Next, we show how model-agnostic, HyperNetwork optimization can efficiently train arbitrary NLP model architectures to learn Pareto-optimal trade-offs between competing metrics. Focusing on the task of toxic language detection, we show the generality and efficacy of our methods across two datasets, three neural architectures, and three fairness losses.

Houjiang Liu, Anubrata Das, Alexander Boltz et al.

While many Natural Language Processing (NLP) techniques have been proposed for fact-checking, both academic research and fact-checking organizations report limited adoption of such NLP work due to poor alignment with fact-checker practices, values, and needs. To address this, we investigate a co-design method, Matchmaking for AI, to enable fact-checkers, designers, and NLP researchers to collaboratively identify what fact-checker needs should be addressed by technology, and to brainstorm ideas for potential solutions. Co-design sessions we conducted with 22 professional fact-checkers yielded a set of 11 design ideas that offer a "north star", integrating fact-checker criteria into novel NLP design concepts. These concepts range from pre-bunking misinformation, efficient and personalized monitoring misinformation, proactively reducing fact-checker potential biases, and collaborative writing fact-check reports. Our work provides new insights into both human-centered fact-checking research and practice and AI co-design research.

Yiheng Su, Junyi Jessy Li, Matthew Lease

Can we preserve the accuracy of neural models while also providing faithful explanations of model decisions to training data? We propose a "wrapper box'' pipeline: training a neural model as usual and then using its learned feature representation in classic, interpretable models to perform prediction. Across seven language models of varying sizes, including four large language models (LLMs), two datasets at different scales, three classic models, and four evaluation metrics, we first show that the predictive performance of wrapper classic models is largely comparable to the original neural models. Because classic models are transparent, each model decision is determined by a known set of training examples that can be directly shown to users. Our pipeline thus preserves the predictive performance of neural language models while faithfully attributing classic model decisions to training data. Among other use cases, such attribution enables model decisions to be contested based on responsible training instances. Compared to prior work, our approach achieves higher coverage and correctness in identifying which training data to remove to change a model decision. To reproduce findings, our source code is online at: https://github.com/SamSoup/WrapperBox.

Soumyajit Gupta, Maria De-Arteaga, Matthew Lease

Target-group detection is the task of detecting which group(s) a piece of content is ``directed at or about''. Applications include targeted marketing, content recommendation, and group-specific content assessment. Key challenges include: 1) that a single post may target multiple groups; and 2) ensuring consistent detection accuracy across groups for fairness. In this work, we investigate fairness implications of target-group detection in the context of toxicity detection, where the perceived harm of a social media post often depends on which group(s) it targets. Because toxicity is highly contextual, language that appears benign in general can be harmful when targeting specific demographic groups. We show our {\em fairness-aware multi-group target detection} approach both reduces bias across groups and shows strong predictive performance, surpassing existing fairness-aware baselines. To enable reproducibility and spur future work, we share our code online.

Md Mustafizur Rahman, Mucahid Kutlu, Tamer Elsayed et al.

To create a new IR test collection at low cost, it is valuable to carefully select which documents merit human relevance judgments. Shared task campaigns such as NIST TREC pool document rankings from many participating systems (and often interactive runs as well) in order to identify the most likely relevant documents for human judging. However, if one's primary goal is merely to build a test collection, it would be useful to be able to do so without needing to run an entire shared task. Toward this end, we investigate multiple active learning strategies which, without reliance on system rankings: 1) select which documents human assessors should judge; and 2) automatically classify the relevance of additional unjudged documents. To assess our approach, we report experiments on five TREC collections with varying scarcity of relevant documents. We report labeling accuracy achieved, as well as rank correlation when evaluating participant systems based upon these labels vs.\ full pool judgments. Results show the effectiveness of our approach, and we further analyze how varying relevance scarcity across collections impacts our findings. To support reproducibility and follow-on work, we have shared our code online: https://github.com/mdmustafizurrahman/ICTIR_AL_TestCollection_2020/.

Terrence Neumann, Sooyong Lee, Maria De-Arteaga et al.

The pervasive spread of misinformation and disinformation poses a significant threat to society. Professional fact-checkers play a key role in addressing this threat, but the vast scale of the problem forces them to prioritize their limited resources. This prioritization may consider a range of factors, such as varying risks of harm posed to specific groups of people. In this work, we investigate potential implications of using a large language model (LLM) to facilitate such prioritization. Because fact-checking impacts a wide range of diverse segments of society, it is important that diverse views are represented in the claim prioritization process. This paper examines whether a LLM can reflect the views of various groups when assessing the harms of misinformation, focusing on gender as a primary variable. We pose two central questions: (1) To what extent do prompts with explicit gender references reflect gender differences in opinion in the United States on topics of social relevance? and (2) To what extent do gender-neutral prompts align with gendered viewpoints on those topics? To analyze these questions, we present the TopicMisinfo dataset, containing 160 fact-checked claims from diverse topics, supplemented by nearly 1600 human annotations with subjective perceptions and annotator demographics. Analyzing responses to gender-specific and neutral prompts, we find that GPT 3.5-Turbo reflects empirically observed gender differences in opinion but amplifies the extent of these differences. These findings illuminate AI's complex role in moderating online communication, with implications for fact-checkers, algorithm designers, and the use of crowd-workers as annotators. We also release the TopicMisinfo dataset to support continuing research in the community.

Sanjana Gautam, Houjiang Liu, Yujin Choi et al.

In the early stages of scientific research, researchers rely on core scholarly judgments to identify relevant literature, assess credible evidence, and determine which directions merit pursuit. As AI tools become increasingly integrated into these early-stage workflows, the scholarly judgments that were once transparent and attributable to individual researchers become obscured, raising critical Responsible AI (RAI) concerns around accountability, transparency, and trust. Yet how these three dimensions manifest in real-time, in-situ scholarly practice remains largely unexplored. To address this gap, we conducted a think-aloud study with 15 researchers to examine how they used AI tools powered by large language models (LLMs) across early-stage research tasks, including literature exploration, synthesis, and research ideation. Our key findings address the tripartite constructs of accountability, transparency, and trust. First, the confident tone of AI outputs misrepresents epistemic uncertainty, making it more difficult for researchers (who are ultimately accountable) to identify which outputs require the greatest scrutiny. Second, opaque retrieval and content construction make provenance difficult to establish for transparency. Third, trust in AI is fragile, context-dependent, and easily eroded. In response, participant researchers were seen to develop compensatory strategies to restore scholarly judgment under uncertainty. Overall, our findings serve to contextualize AI-mediated research as a RAI problem grounded in lived researcher experience and motivate attention to deliberate AI integration that preserves accountability, supports transparency, and fosters informed trust.

Venelin Kovatchev, Matthew Lease

In this paper we present an exploratory research on quantifying the impact that data distribution has on the performance and evaluation of NLP models. We propose an automated framework that measures the data point distribution across 6 different dimensions: ambiguity, difficulty, discriminability, length, noise, and perplexity. We use disproportional stratified sampling to measure how much the data distribution affects absolute (Acc/F1) and relative (Rank) model performance. We experiment on 2 different datasets (SQUAD and MNLI) and test a total of 135 different models (125 on SQUAD and 10 on MNLI). We demonstrate that without explicit control of the data distribution, standard evaluation frameworks are inconsistent and unreliable. We find that the impact of the data is statistically significant and is often larger than the impact of changing the metric. In a second set of experiments, we demonstrate that the impact of data on evaluation is not just observable, but also predictable. We propose to use benchmark transparency as a method for comparing datasets and quantifying the similarity between them. We find that the ``dataset similarity vector'' can be used to predict how well a model generalizes out of distribution.

Alexander Braylan, Madalyn Marabella, Omar Alonso et al.

Human annotations are vital to supervised learning, yet annotators often disagree on the correct label, especially as annotation tasks increase in complexity. A strategy to improve label quality is to ask multiple annotators to label the same item and aggregate their labels. Many aggregation models have been proposed for categorical or numerical annotation tasks, but far less work has considered more complex annotation tasks involving open-ended, multivariate, or structured responses. While a variety of bespoke models have been proposed for specific tasks, our work is the first to introduce aggregation methods that generalize across many diverse complex tasks, including sequence labeling, translation, syntactic parsing, ranking, bounding boxes, and keypoints. This generality is achieved by devising a task-agnostic method to model distances between labels rather than the labels themselves. This article extends our prior work with investigation of three new research questions. First, how do complex annotation properties impact aggregation accuracy? Second, how should a task owner navigate the many modeling choices to maximize aggregation accuracy? Finally, what diagnoses can verify that aggregation models are specified correctly for the given data? To understand how various factors impact accuracy and to inform model selection, we conduct simulation studies and experiments on real, complex datasets. Regarding testing, we introduce unit tests for aggregation models and present a suite of such tests to ensure that a given model is not mis-specified and exhibits expected behavior. Beyond investigating these research questions above, we discuss the foundational concept of annotation complexity, present a new aggregation model as a bridge between traditional models and our own, and contribute a new semi-supervised learning method for complex label aggregation that outperforms prior work.

Gauri Kambhatla, Matthew Lease, Ashwin Rajadesingan

To promote constructive discussion of controversial topics online, we propose automatic reframing of disagreeing responses to signal receptiveness to a preceding comment. Drawing on research from psychology, communications, and linguistics, we identify six strategies for reframing. We automatically reframe replies to comments according to each strategy, using a Reddit dataset. Through human-centered experiments, we find that the replies generated with our framework are perceived to be significantly more receptive than the original replies and a generic receptiveness baseline. We illustrate how transforming receptiveness, a particular social science construct, into a computational framework, can make LLM generations more aligned with human perceptions. We analyze and discuss the implications of our results, and highlight how a tool based on our framework might be used for more teachable and creative content moderation.

Gauri Kambhatla, Sanjana Gautam, Angela Zhang et al.

The ability to accurately align LLMs with human population groups on subjective questions would have great value. In this work, we show that use of simple supervision can greatly improve language model alignment with diverse population groups more consistently, as measured over three datasets spanning various topics. Beyond evaluating average alignment, we also report how alignment varies across specific groups. Our broad findings provide insights into the distributional alignment of LLMs with diverse population groups. By conducting evaluation over many LLMs and prompting strategies, along with open-sourcing our work, we provide a benchmark to stimulate future research.

Yiheng Su, Matthew Lease

We present RELISH (REgression with a Latent Iterative State Head), a novel, lightweight architecture designed for text regression with large language models. Rather than decoding numeric targets as text or aggregating multiple generated outputs, RELISH predicts scalar values directly from frozen LLM representations by iteratively refining a learned latent state through cross-attention over token-level representations, and then mapping the final state to a point estimate with a linear regressor. Across five datasets, four LLM backbones, and two LLM training regimes, RELISH consistently outperforms prior baselines from all three major LLM regression families, including autoregressive decoding, regression-aware inference, and existing predictive head methods. Despite these gains, RELISH remains highly parameter-efficient, requiring only 3.4-3.7M trainable parameters across frozen LLM backbones (only 0.01-0.04% additional overhead), far less than LoRA-based alternatives that grow with model size (0.26-0.42%).

Chi-Yang Hsu, Alexander Braylan, Yiheng Su et al.

We motivate and share a new benchmark for instance-level performance prediction of long-form generation tasks having multi-faceted, fine-grained quality metrics. Our task-, model- and metric-agnostic formulation predicts continuous evaluation metric scores given only black-box model inputs and outputs. Beyond predicting point estimates of metric scores, the benchmark also requires inferring prediction intervals to quantify uncertainty around point estimates. Evaluation spans 11 long-form datasets/tasks with multiple LLMs, baselines, and metrics per task. We show that scores can be effectively predicted across long-form generation tasks using as few as 16 training examples. Overall, we introduce a novel and useful task, a valuable benchmark to drive progress, and baselines ready for practical adoption today.

Vijay Keswani, L. Elisa Celis, Krishnaram Kenthapadi et al.

Automatically assigning tasks to people is challenging because human performance can vary across tasks for many reasons. This challenge is further compounded in real-life settings in which no oracle exists to assess the quality of human decisions and task assignments made. Instead, we find ourselves in a "closed" decision-making loop in which the same fallible human decisions we rely on in practice must also be used to guide task allocation. How can imperfect and potentially biased human decisions train an accurate allocation model? Our key insight is to exploit weak prior information on human-task similarity to bootstrap model training. We show that the use of such a weak prior can improve task allocation accuracy, even when human decision-makers are fallible and biased. We present both theoretical analysis and empirical evaluation over synthetic data and a social media toxicity detection task. Results demonstrate the efficacy of our approach.

Li Shi, Nilavra Bhattacharya, Anubrata Das et al.

We conducted a lab-based eye-tracking study to investigate how the interactivity of an AI-powered fact-checking system affects user interactions, such as dwell time, attention, and mental resources involved in using the system. A within-subject experiment was conducted, where participants used an interactive and a non-interactive version of a mock AI fact-checking system and rated their perceived correctness of COVID-19 related claims. We collected web-page interactions, eye-tracking data, and mental workload using NASA-TLX. We found that the presence of the affordance of interactively manipulating the AI system's prediction parameters affected users' dwell times, and eye-fixations on AOIs, but not mental workload. In the interactive system, participants spent the most time evaluating claims' correctness, followed by reading news. This promising result shows a positive role of interactivity in a mixed-initiative AI-powered system.

Vijay Keswani, Matthew Lease, Krishnaram Kenthapadi

In hybrid human-machine deferral frameworks, a classifier can defer uncertain cases to human decision-makers (who are often themselves fallible). Prior work on simultaneous training of such classifier and deferral models has typically assumed access to an oracle during training to obtain true class labels for training samples, but in practice there often is no such oracle. In contrast, we consider a "closed" decision-making pipeline in which the same fallible human decision-makers used in deferral also provide training labels. How can imperfect and biased human expert labels be used to train a fair and accurate deferral framework? Our key insight is that by exploiting weak prior information, we can match experts to input examples to ensure fairness and accuracy of the resulting deferral framework, even when imperfect and biased experts are used in place of ground truth labels. The efficacy of our approach is shown both by theoretical analysis and by evaluation on two tasks.

Vivek Krishna Pradhan, Mike Schaekermann, Matthew Lease

We propose a novel three-stage FIND-RESOLVE-LABEL workflow for crowdsourced annotation to reduce ambiguity in task instructions and thus improve annotation quality. Stage 1 (FIND) asks the crowd to find examples whose correct label seems ambiguous given task instructions. Workers are also asked to provide a short tag which describes the ambiguous concept embodied by the specific instance found. We compare collaborative vs. non-collaborative designs for this stage. In Stage 2 (RESOLVE), the requester selects one or more of these ambiguous examples to label (resolving ambiguity). The new label(s) are automatically injected back into task instructions in order to improve clarity. Finally, in Stage 3 (LABEL), workers perform the actual annotation using the revised guidelines with clarifying examples. We compare three designs for using these examples: examples only, tags only, or both. We report image labeling experiments over six task designs using Amazon's Mechanical Turk. Results show improved annotation accuracy and further insights regarding effective design for crowdsourced annotation tasks.

Danula Hettiachchi, Mark Sanderson, Jorge Goncalves et al.

This volume contains the position papers presented at CSCW 2021 Workshop - Investigating and Mitigating Biases in Crowdsourced Data, held online on 23rd October 2021, at the 24th ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW 2021). The workshop explored how specific crowdsourcing workflows, worker attributes, and work practices contribute to biases in data. The workshop also included discussions on research directions to mitigate labelling biases, particularly in a crowdsourced context, and the implications of such methods for the workers.

Lora Aroyo, Matthew Lease, Praveen Paritosh et al.

The efficacy of machine learning (ML) models depends on both algorithms and data. Training data defines what we want our models to learn, and testing data provides the means by which their empirical progress is measured. Benchmark datasets define the entire world within which models exist and operate, yet research continues to focus on critiquing and improving the algorithmic aspect of the models rather than critiquing and improving the data with which our models operate. If "data is the new oil," we are still missing work on the refineries by which the data itself could be optimized for more effective use.

Soumyajit Gupta, Gurpreet Singh, Raghu Bollapragada et al.

Multi-objective optimization (MOO) problems require balancing competing objectives, often under constraints. The Pareto optimal solution set defines all possible optimal trade-offs over such objectives. In this work, we present a novel method for Pareto-front learning: inducing the full Pareto manifold at train-time so users can pick any desired optimal trade-off point at run-time. Our key insight is to exploit Fritz-John Conditions for a novel guided double gradient descent strategy. Evaluation on synthetic benchmark problems allows us to vary MOO problem difficulty in controlled fashion and measure accuracy vs. known analytic solutions. We further test scalability and generalization in learning optimal neural model parameterizations for Multi-Task Learning (MTL) on image classification. Results show consistent improvement in accuracy and efficiency over prior MTL methods as well as techniques from operations research.

Prakhar Singh, Anubrata Das, Junyi Jessy Li et al.

Fact-checking is the process of evaluating the veracity of claims (i.e., purported facts). In this opinion piece, we raise an issue that has received little attention in prior work -- that some claims are far more difficult to fact-check than others. We discuss the implications this has for both practical fact-checking and research on automated fact-checking, including task formulation and dataset design. We report a manual analysis undertaken to explore factors underlying varying claim difficulty and identify several distinct types of difficulty. We motivate this new claim difficulty prediction task as beneficial to both automated fact-checking and practical fact-checking organizations.

Md Mustafizur Rahman, Dinesh Balakrishnan, Dhiraj Murthy et al.

Building a benchmark dataset for hate speech detection presents various challenges. Firstly, because hate speech is relatively rare, random sampling of tweets to annotate is very inefficient in finding hate speech. To address this, prior datasets often include only tweets matching known "hate words". However, restricting data to a pre-defined vocabulary may exclude portions of the real-world phenomenon we seek to model. A second challenge is that definitions of hate speech tend to be highly varying and subjective. Annotators having diverse prior notions of hate speech may not only disagree with one another but also struggle to conform to specified labeling guidelines. Our key insight is that the rarity and subjectivity of hate speech are akin to that of relevance in information retrieval (IR). This connection suggests that well-established methodologies for creating IR test collections can be usefully applied to create better benchmark datasets for hate speech. To intelligently and efficiently select which tweets to annotate, we apply standard IR techniques of {\em pooling} and {\em active learning}. To improve both consistency and value of annotations, we apply {\em task decomposition} and {\em annotator rationale} techniques. We share a new benchmark dataset for hate speech detection on Twitter that provides broader coverage of hate than prior datasets. We also show a dramatic drop in accuracy of existing detection models when tested on these broader forms of hate. Annotator rationales we collect not only justify labeling decisions but also enable future work opportunities for dual-supervision and/or explanation generation in modeling. Further details of our approach can be found in the supplementary materials.

Ruijiang Gao, Maytal Saar-Tsechansky, Maria De-Arteaga et al.

Human-machine complementarity is important when neither the algorithm nor the human yield dominant performance across all instances in a given domain. Most research on algorithmic decision-making solely centers on the algorithm's performance, while recent work that explores human-machine collaboration has framed the decision-making problems as classification tasks. In this paper, we first propose and then develop a solution for a novel human-machine collaboration problem in a bandit feedback setting. Our solution aims to exploit the human-machine complementarity to maximize decision rewards. We then extend our approach to settings with multiple human decision makers. We demonstrate the effectiveness of our proposed methods using both synthetic and real human responses, and find that our methods outperform both the algorithm and the human when they each make decisions on their own. We also show how personalized routing in the presence of multiple human decision-makers can further improve the human-machine team performance.

Danula Hettiachchi, Mike Schaekermann, Tristan McKinney et al.

We consider a class of variable effort human annotation tasks in which the number of labels required per item can greatly vary (e.g., finding all faces in an image, named entities in a text, bird calls in an audio recording, etc.). In such tasks, some items require far more effort than others to annotate. Furthermore, the per-item annotation effort is not known until after each item is annotated since determining the number of labels required is an implicit part of the annotation task itself. On an image bounding-box task with crowdsourced annotators, we show that annotator accuracy and recall consistently drop as effort increases. We hypothesize reasons for this drop and investigate a set of approaches to counteract it. Firstly, we benchmark on this task a set of general best-practice methods for quality crowdsourcing. Notably, only one of these methods actually improves quality: the use of visible gold questions that provide periodic feedback to workers on their accuracy as they work. Given these promising results, we then investigate and evaluate variants of the visible gold approach, yielding further improvement. Final results show a 7% improvement in bounding-box accuracy over the baseline. We discuss the generality of the visible gold approach and promising directions for future research.

Vijay Keswani, Matthew Lease, Krishnaram Kenthapadi

Machine learning models are often implemented in cohort with humans in the pipeline, with the model having an option to defer to a domain expert in cases where it has low confidence in its inference. Our goal is to design mechanisms for ensuring accuracy and fairness in such prediction systems that combine machine learning model inferences and domain expert predictions. Prior work on "deferral systems" in classification settings has focused on the setting of a pipeline with a single expert and aimed to accommodate the inaccuracies and biases of this expert to simultaneously learn an inference model and a deferral system. Our work extends this framework to settings where multiple experts are available, with each expert having their own domain of expertise and biases. We propose a framework that simultaneously learns a classifier and a deferral system, with the deferral system choosing to defer to one or more human experts in cases of input where the classifier has low confidence. We test our framework on a synthetic dataset and a content moderation dataset with biased synthetic experts, and show that it significantly improves the accuracy and fairness of the final predictions, compared to the baselines. We also collect crowdsourced labels for the content moderation task to construct a real-world dataset for the evaluation of hybrid machine-human frameworks and show that our proposed learning framework outperforms baselines on this real-world dataset as well.

Gurpreet Singh, Soumyajit Gupta, Matthew Lease et al.

Classification, recommendation, and ranking problems often involve competing goals with additional constraints (e.g., to satisfy fairness or diversity criteria). Such optimization problems are quite challenging, often involving non-convex functions along with considerations of user preferences in balancing trade-offs. Pareto solutions represent optimal frontiers for jointly optimizing multiple competing objectives. A major obstacle for frequently used linear-scalarization strategies is that the resulting optimization problem might not always converge to a global optimum. Furthermore, such methods only return one solution point per run. A Pareto solution set is a subset of all such global optima over multiple runs for different trade-off choices. Therefore, a Pareto front can only be guaranteed with multiple runs of the linear-scalarization problem, where all runs converge to their respective global optima. Consequently, extracting a Pareto front for practical problems is computationally intractable with substantial computational overheads, limited scalability, and reduced accuracy. We propose a robust, low cost, two-stage, hybrid neural Pareto optimization approach that is accurate and scales (compute space and time) with data dimensions, as well as number of functions and constraints. The first stage (neural network) efficiently extracts a weak Pareto front, using Fritz-John conditions as the discriminator, with no assumptions of convexity on the objectives or constraints. The second stage (efficient Pareto filter) extracts the strong Pareto optimal subset given the weak front from stage 1. Fritz-John conditions provide us with theoretical bounds on approximation error between the true and network extracted weak Pareto front. Numerical experiments demonstrates the accuracy and efficiency on a canonical set of benchmark problems and a fairness optimization task from prior works.

Md Mustafizur Rahman, Mucahid Kutlu, Matthew Lease

Research community evaluations in information retrieval, such as NIST's Text REtrieval Conference (TREC), build reusable test collections by pooling document rankings submitted by many teams. Naturally, the quality of the resulting test collection thus greatly depends on the number of participating teams and the quality of their submitted runs. In this work, we investigate: i) how the number of participants, coupled with other factors, affects the quality of a test collection; and ii) whether the quality of a test collection can be inferred prior to collecting relevance judgments from human assessors. Experiments conducted on six TREC collections illustrate how the number of teams interacts with various other factors to influence the resulting quality of test collections. We also show that the reusability of a test collection can be predicted with high accuracy when the same document collection is used for successive years in an evaluation campaign, as is common in TREC.

Prateek Chaudhry, Matthew Lease

Hate speech detection research has predominantly focused on purely content-based methods, without exploiting any additional context. We briefly critique pros and cons of this task formulation. We then investigate profiling users by their past utterances as an informative prior to better predict whether new utterances constitute hate speech. To evaluate this, we augment three Twitter hate speech datasets with additional timeline data, then embed this additional context into a strong baseline model. Promising results suggest merit for further investigation, though analysis is complicated by differences in annotation schemes and processes, as well as Twitter API limitations and data sharing policies.

Gurpreet Singh, Soumyajit Gupta, Matthew Lease et al.

In a Big Data setting computing the dominant SVD factors is restrictive due to the main memory requirements. Recently introduced streaming Randomized SVD schemes work under the restrictive assumption that the singular value spectrum of the data has exponential decay. This is seldom true for any practical data. Although these methods are claimed to be applicable to scientific computations due to associated tail-energy error bounds, the approximation errors in the singular vectors and values are high when the aforementioned assumption does not hold. Furthermore from a practical perspective, oversampling can still be memory intensive or worse can exceed the feature dimension of the data. To address these issues, we present Range-Net as an alternative to randomized SVD that satisfies the tail-energy lower bound given by Eckart-Young-Mirsky (EYM) theorem. Range-Net is a deterministic two-stage neural optimization approach with random initialization, where the main memory requirement depends explicitly on the feature dimension and desired rank, independent of the sample dimension. The data samples are read in a streaming setting with the network minimization problem converging to the desired rank-r approximation. Range-Net is fully interpretable where all the network outputs and weights have a specific meaning. We provide theoretical guarantees that Range-Net extracted SVD factors satisfy EYM tail-energy lower bound at machine precision. Our numerical experiments on real data at various scales confirms this bound. A comparison against the state of the art streaming Randomized SVD shows that Range-Net accuracy is better by six orders of magnitude while being memory efficient.

Gurpreet Singh, Soumyajit Gupta, Matthew Lease

Constrained Optimization solution algorithms are restricted to point based solutions. In practice, single or multiple objectives must be satisfied, wherein both the objective function and constraints can be non-convex resulting in multiple optimal solutions. Real world scenarios include intersecting surfaces as Implicit Functions, Hyperspectral Unmixing and Pareto Optimal fronts. Local or global convexification is a common workaround when faced with non-convex forms. However, such an approach is often restricted to a strict class of functions, deviation from which results in sub-optimal solution to the original problem. We present neural solutions for extracting optimal sets as approximate manifolds, where unmodified, non-convex objectives and constraints are defined as modeler guided, domain-informed $L_2$ loss function. This promotes interpretability since modelers can confirm the results against known analytical forms in their specific domains. We present synthetic and realistic cases to validate our approach and compare against known solvers for bench-marking in terms of accuracy and computational efficiency.

Anubrata Das, Matthew Lease

While search efficacy has been evaluated traditionally on the basis of result relevance, fairness of search has attracted recent attention. In this work, we define a notion of distributional fairness and provide a conceptual framework for evaluating search results based on it. As part of this, we formulate a set of axioms which an ideal evaluation framework should satisfy for distributional fairness. We show how existing TREC test collections can be repurposed to study fairness, and we measure potential data bias to inform test collection design for fair search. A set of analyses show metric divergence between relevance and fairness, and we describe a simple but flexible interpolation strategy for integrating relevance and fairness into a single metric for optimization and evaluation.

Anubrata Das, Kunjan Mehta, Matthew Lease

The effect of user bias in fact-checking has not been explored extensively from a user-experience perspective. We estimate the user bias as a function of the user's perceived reputation of the news sources (e.g., a user with liberal beliefs may tend to trust liberal sources). We build an interface to communicate the role of estimated user bias in the context of a fact-checking task. We also explore the utility of helping users visualize their detected level of bias. 80% of the users of our system find that the presence of an indicator for user bias is useful in judging the veracity of a political claim.

Mucahid Kutlu, Tyler McDonnell, Aashish Sheshadri et al.

Crowdsourcing offers an affordable and scalable means to collect relevance judgments for IR test collections. However, crowd assessors may show higher variance in judgment quality than trusted assessors. In this paper, we investigate how to effectively utilize both groups of assessors in partnership. We specifically investigate how agreement in judging is correlated with three factors: relevance category, document rankings, and topical variance. Based on this, we then propose two collaborative judging methods in which a portion of the document-topic pairs are assessed by in-house judges while the rest are assessed by crowd-workers. Experiments conducted on two TREC collections show encouraging results when we distribute work intelligently between our two groups of assessors.

Brandon Dang, Martin J. Riedl, Matthew Lease

Though detection systems have been developed to identify obscene content such as pornography and violence, artificial intelligence is simply not good enough to fully automate this task yet. Due to the need for manual verification, social media companies may hire internal reviewers, contract specialized workers from third parties, or outsource to online labor markets for the purpose of commercial content moderation. These content moderators are often fully exposed to extreme content and may suffer lasting psychological and emotional damage. In this work, we aim to alleviate this problem by investigating the following question: How can we reveal the minimum amount of information to a human reviewer such that an objectionable image can still be correctly identified? We design and conduct experiments in which blurred graphic and non-graphic images are filtered by human moderators on Amazon Mechanical Turk (AMT). We observe how obfuscation affects the moderation experience with respect to image classification accuracy, interface usability, and worker emotional well-being.

Mucahid Kutlu, Vivek Khetan, Matthew Lease

Because researchers typically do not have the time or space to present more than a few evaluation metrics in any published study, it can be difficult to assess relative effectiveness of prior methods for unreported metrics when baselining a new method or conducting a systematic meta-review. While sharing of study data would help alleviate this, recent attempts to encourage consistent sharing have been largely unsuccessful. Instead, we propose to enable relative comparisons with prior work across arbitrary metrics by predicting unreported metrics given one or more reported metrics. In addition, we further investigate prediction of high-cost evaluation measures using low-cost measures as a potential strategy for reducing evaluation cost. We begin by assessing the correlation between 23 IR metrics using 8 TREC test collections. Measuring prediction error wrt. R-square and Kendall's tau, we show that accurate prediction of MAP, P@10, and RBP can be achieved using only 2-3 other metrics. With regard to lowering evaluation cost, we show that RBP(p=0.95) can be predicted with high accuracy using measures with only evaluation depth of 30. Taken together, our findings provide a valuable proof-of-concept which we expect to spur follow-on work by others in proposing more sophisticated models for metric prediction.

Akash Mankar, Riddhi J. Shah, Matthew Lease

Entry-level crowd work is often reported to pay less than minimum wage. While this may be appropriate or even necessary, due to various legal, economic, and pragmatic factors, some Requesters and workers continue to question this status quo. To promote further discussion on the issue, we survey Requesters and workers whether they would support restricting tasks to require minimum wage pay. As a form of design activism, we confronted workers with this dilemma directly by posting a dummy Mechanical Turk task which told them that they could not work on it because it paid less than their local minimum wage, and we invited their feedback. Strikingly, for those workers expressing an opinion, two-thirds of Indians favored the policy while two-thirds of Americans opposed it. Though a majority of Requesters supported minimum wage pay, only 20\% would enforce it. To further empower Requesters, and to ensure that effort or ignorance are not barriers to change, we provide a simple public API to make it easy to find a worker's local minimum wage by his/her IP address.

Ye Zhang, Matthew Lease, Byron C. Wallace

A fundamental advantage of neural models for NLP is their ability to learn representations from scratch. However, in practice this often means ignoring existing external linguistic resources, e.g., WordNet or domain specific ontologies such as the Unified Medical Language System (UMLS). We propose a general, novel method for exploiting such resources via weight sharing. Prior work on weight sharing in neural networks has considered it largely as a means of model compression. In contrast, we treat weight sharing as a flexible mechanism for incorporating prior knowledge into neural models. We show that this approach consistently yields improved performance on classification tasks compared to baseline strategies that do not exploit weight sharing.

Mucahid Kutlu, Tamer Elsayed, Matthew Lease

While test collections provide the cornerstone for Cranfield-based evaluation of information retrieval (IR) systems, it has become practically infeasible to rely on traditional pooling techniques to construct test collections at the scale of today's massive document collections. In this paper, we propose a new intelligent topic selection method which reduces the number of search topics needed for reliable IR evaluation. To rigorously assess our method, we integrate previously disparate lines of research on intelligent topic selection and deep vs. shallow judging. While prior work on intelligent topic selection has never been evaluated against shallow judging baselines, prior work on deep vs. shallow judging has largely argued for shallowed judging, but assuming random topic selection. We argue that for evaluating any topic selection method, ultimately one must ask whether it is actually useful to select topics, or should one simply perform shallow judging over many topics? In seeking a rigorous answer to this over-arching question, we conduct a comprehensive investigation over a set of relevant factors never previously studied together 1) topic selection method 2) the effect of topic familiarity on human judging speed and 3) how different topic generation processes impact (i) budget utilization and (ii) the resultant quality of judgments. Experiments on NIST TREC Robust 2003 and Robust 2004 test collections show that not only can we reliably evaluate IR systems with fewer topics, but also that 1) when topics are intelligently selected, deep judging is often more cost-effective than shallow judging in evaluation reliability and 2) topic familiarity and topic generation costs greatly impact the evaluation cost vs. reliability trade-off. Our findings challenge conventional wisdom in showing that deep judging is often preferable to shallow judging when topics are selected intelligently.

Ye Zhang, Md Mustafizur Rahman, Alex Braylan et al.

A recent "third wave" of Neural Network (NN) approaches now delivers state-of-the-art performance in many machine learning tasks, spanning speech recognition, computer vision, and natural language processing. Because these modern NNs often comprise multiple interconnected layers, this new NN research is often referred to as deep learning. Stemming from this tide of NN work, a number of researchers have recently begun to investigate NN approaches to Information Retrieval (IR). While deep NNs have yet to achieve the same level of success in IR as seen in other areas, the recent surge of interest and work in NNs for IR suggest that this state of affairs may be quickly changing. In this work, we survey the current landscape of Neural IR research, paying special attention to the use of learned representations of queries and documents (i.e., neural embeddings). We highlight the successes of neural IR thus far, catalog obstacles to its wider adoption, and suggest potentially promising directions for future research.

Yalin Sun, Pengxiang Cheng, Shengwei Wang et al.

Information extraction is a critical step in the practice of conducting biomedical systematic literature reviews. Extracted structured data can be aggregated via methods such as statistical meta-analysis. Typically highly trained domain experts extract data for systematic reviews. The high expense of conducting biomedical systematic reviews has motivated researchers to explore lower cost methods that achieve similar rigor without compromising quality. Crowdsourcing represents one such promising approach. In this work-in-progress study, we designed a crowdsourcing task for biomedical information extraction. We briefly report the iterative design process and the results of two pilot testings. We found that giving more concrete examples in the task instruction can help workers better understand the task, especially for concepts that are abstract and confusing. We found a few workers completed most of the work, and our payment level appeared more attractive to workers from low-income countries. In the future, we will further evaluate our results with reference to gold standard extractions, thus assessing the feasibility of tasking crowd workers with extracting biomedical intervention information for systematic reviews.

Ye Zhang, Matthew Lease, Byron C. Wallace

We propose a new active learning (AL) method for text classification with convolutional neural networks (CNNs). In AL, one selects the instances to be manually labeled with the aim of maximizing model performance with minimal effort. Neural models capitalize on word embeddings as representations (features), tuning these to the task at hand. We argue that AL strategies for multi-layered neural models should focus on selecting instances that most affect the embedding space (i.e., induce discriminative word representations). This is in contrast to traditional AL approaches (e.g., entropy-based uncertainty sampling), which specify higher level objectives. We propose a simple approach for sentence classification that selects instances containing words whose embeddings are likely to be updated with the greatest magnitude, thereby rapidly learning discriminative, task-specific embeddings. We extend this approach to document classification by jointly considering: (1) the expected changes to the constituent word representations; and (2) the model's current overall uncertainty regarding the instance. The relative emphasis placed on these criteria is governed by a stochastic process that favors selecting instances likely to improve representations at the outset of learning, and then shifts toward general uncertainty sampling as AL progresses. Empirical results show that our method outperforms baseline AL approaches on both sentence and document classification tasks. We also show that, as expected, the method quickly learns discriminative word embeddings. To the best of our knowledge, this is the first work on AL addressing neural models for text classification.

Xi Zheng, Akanksha Bansal, Matthew Lease

We present the Bullseye system for scholarly search. Given a collection of research papers, Bullseye: 1) identifies relevant passages using any on-the-shelf algorithm; 2) automatically detects document structure and restricts retrieved passages to user-specifed sections; and 3) highlights those passages for each PDF document retrieved. We evaluate Bullseye with regard to three aspects: system effectiveness, user effectiveness, and user effort. In a system-blind evaluation, users were asked to compare passage retrieval using Bullseye vs. a baseline which ignores document structure, in regard to four types of graded assessments. Results show modest improvement in system effectiveness while both user effectiveness and user effort show substantial improvement. Users also report very strong demand for passage highlighting in scholarly search across both systems considered.