Local Decision Pitfalls in Interactive Machine Learning: An Investigation into Feature Selection in Sentiment Analysis
The DOI matches its Crossref record; 66 of the 67 references it cites were found in a registry or an index.
Abstract
Tools for Interactive Machine Learning (IML) enable end users to update models in a “rapid, focused, and incremental”—yet local—manner. In this work, we study the question of local decision making in an IML context around feature selection for a sentiment classification task. Specifically, we characterize the utility of interactive feature selection through a combination of human-subjects experiments and computational simulations. We find that, in expectation, interactive modification fails to improve model performance and may hamper generalization due to overfitting. We examine how these trends are affected by the dataset, learning algorithm, and the training set size. Across these factors we observe consistent generalization issues. Our results suggest that rapid iterations with IML systems can be dangerous if they encourage local actions divorced from global context, degrading overall model performance. We conclude by discussing the implications of our feature selection results to the broader area of IML systems and research.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Citations
Cites
Scikit-learn: Machine Learning in Python
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Müller, A., Nothman, J., Louppe, G., et al. (2012) · arXiv · doi:10.48550/arxiv.1201.0490 · Cited by 71,016 · cited 1×
“We run the simulations across three commonly used classification algorithms: Logistic Regression, SVM, and random forest, using the implementations within Scikit-Learn [46].”
MethodJudgment under Uncertainty: Heuristics and Biases
Tversky, A., & Kahneman, D. (1975) · Utility, Probability, and Human Decision Making · doi:10.1007/978-94-010-1834-0_8 · Cited by 22,650
Individual Comparisons by Ranking Methods
Wilcoxon, F. (1945) · Biometrics Bulletin · doi:10.2307/3001968 · Cited by 15,354
Random effects structure for confirmatory hypothesis testing: Keep it maximal
Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013) · Journal of Memory and Language · doi:10.1016/j.jml.2012.11.001 · Cited by 10,328 · cited 1×
“We also included a maximal random effects structure [8], with a per-subject random inter-cept (capturing individual decision thresholds) and per-subject random slopes for all fixed effects (capturing varied sensitivities to those factors).”
MethodScikit-learn: Machine Learning in Python
(2011) · Journal of Machine Learning Research · doi:10.5555/1953048.2078195 · Cited by 8,539 · cited 1×
“We run the simulations across three commonly used classification algorithms: Logistic Regression, SVM, and random forest, using the implementations within Scikit-Learn [46].”
MethodThe Kolmogorov-Smirnov Test for Goodness of Fit
Massey, F. J. (1951) · Journal of the American Statistical Association · doi:10.1080/01621459.1951.10500769 · Cited by 7,210
Learning Word Vectors for Sentiment Analysis
Maas, A. L., Daly, R. E., Pham, P. T., Huang, D., Ng, A. Y., & Potts, C. (2011) · Cited by 6,960
Supervised Machine Learning: A Review of Classification Techniques
Kotsiantis, S. B. (2007) · Informatica · Cited by 6,007
Social psychology: handbook of basic principles
(1997) · Choice Reviews Online · doi:10.5860/choice.34-3595 · Cited by 4,895
The Kolmogorov-Smirnov Test for Goodness of Fit
Massey, F. J. (1951) · Journal of the American Statistical Association · doi:10.2307/2280095 · Cited by 3,690 · cited 1×
“In addition, where applicable, we examine the differences between the Δ F1 distribution resulting from different conditions using the Kolmogorov–Smirnov test [40].”
Method
Research works printed but not linked: 7
- Tolga Bolukbasi , Kai-Wei Chang , James Y. Zou , Venkatesh Saligrama , and Adam T . Kalai . 2016 . Man is to computer programmer as woman is to homemaker? Debiasing word embeddings. In Advances in Neural Information Processing Systems . 4349--4357. Tolga Bolukbasi, Kai-Wei Chang, James Y. Zou, Venkatesh Saligrama, and Adam T. Kalai. 2016. Man is to computer programmer as woman is to homemaker? Debiasing word embeddings. In Advances in Neural Information Processing Systems. 4349--4357.
- Jessica Hullman . 2013 . How prior knowledge affects the processing of visualized data . In Proceedings of the ACM CHI 2013, Many People Many Eyes Workshop. Jessica Hullman. 2013. How prior knowledge affects the processing of visualized data. In Proceedings of the ACM CHI 2013, Many People Many Eyes Workshop.
- Rensis Likert . 1932. A technique for the measurement of attitudes . In Archives of Psychology . The Science Press , New York . Rensis Likert. 1932. A technique for the measurement of attitudes. In Archives of Psychology. The Science Press, New York.
- 10.5555/2002472.2002491 · doi:10.5555/2002472.2002491
- Vangelis Metsis , Ion Androutsopoulos , and Georgios Paliouras . 2006 . Spam filtering with naive Bayes—which naive Bayes? In Proceedings of the 3rd Conference on Email and Anti-Spam (CEAS’06) , Vol. 17 . 28--69. Vangelis Metsis, Ion Androutsopoulos, and Georgios Paliouras. 2006. Spam filtering with naive Bayes—which naive Bayes? In Proceedings of the 3rd Conference on Email and Anti-Spam (CEAS’06), Vol. 17. 28--69.
- Xin Rong and Eytan Adar . 2016 . Visual tools for debugging neural language models . In Proceedings of ICML Workshop on Visualization for Deep Learning. Xin Rong and Eytan Adar. 2016. Visual tools for debugging neural language models. In Proceedings of ICML Workshop on Visualization for Deep Learning.
- Simone Stumpf , Adrian Bussone , and Dympna O ’sullivan . 2016 . Explanations considered harmful? User interactions with machine learning systems . In Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems (CHI’16) . Simone Stumpf, Adrian Bussone, and Dympna O ’sullivan. 2016. Explanations considered harmful? User interactions with machine learning systems. In Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems (CHI’16).
Cited by
Smith-Renner, A., Fan, R., Birchfield, M., Wu, T., Boyd-Graber, J., Weld, D. S., & Findlater, L. (2020) · Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3313831.3376624 · Cited by 112
Self-citationLiu, H., Lai, V., & Tan, C. (2021) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3479552 · Cited by 111
“Similar interactive tools have been used to support data scientists in debugging machine learning models and improving model performance [24, 26, 69].”
MethodExplainable Active Learning (XAL): Toward AI Explanations as Interfaces for Machine Teachers
Ghai, B., Liao, Q. V., Zhang, Y., Bellamy, R., & Mueller, K. (2021) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3432934 · Cited by 106
“For example, some have warned against “local decision pitfalls” [97] of human feedback in iML that overly focuses on modifying a subset of model features, commonly resulting in an overfitted model that fails to generalize.”
BackgroundImproving Human-AI Collaboration With Descriptions of AI Behavior
Cabrera, Á. A., Perer, A., & Hong, J. I. (2023) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3579612 · Cited by 87
“For this work, we used behavior descriptions stating the model accuracy for subgroups of the data for which the model performs significantly worse than average [10, 18, 44, 52, 62].”
MethodHuman-AI Collaboration for UX Evaluation: Effects of Explanation and Synchronization
Fan, M., Yang, X., Yu, T., Liao, Q. V., & Zhao, J. (2022) · Proceedings of the ACM on Human-Computer Interaction · doi:10.1145/3512943 · Cited by 81
Facilitating Knowledge Sharing from Domain Experts to Data Scientists for Building NLP Models
Park, S., Wang, A. Y., Kawas, B., Liao, Q. V., Piorkowski, D., & Danilevsky, M. (2021) · 26th International Conference on Intelligent User Interfaces · doi:10.1145/3397481.3450637 · Cited by 53
“There are known issues with letting ML novices directly adjust models [74], such as lacking generalization or over-fitting [84].”
BackgroundLeveraging explanations in interactive machine learning: An overview
Teso, S., Alkan, Ö., Stammer, W., & Daly, E. (2023) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2023.1066049 · Cited by 106
“This, in turn, complicates acquiring high-quality supervision and allocating trust [182, 122].”
BackgroundExplanation-Based Human Debugging of NLP Models: A Survey
Lertvittayakumjorn, P., & Toni, F. (2021) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00440 · Cited by 91
“…papers, we started from some pivotal EBHD work, e.g., (Kulesza et al., 2015; Ribeiro et al., 2016; Teso and Kersting, 2019), and added EBHD papers citing or being cited by the pivotal work, e.g., (Stumpf et al., 2009; Kulesza et al., 2010; Lertvittayakumjorn et al., 2020; Yao et al., 2021).”
BackgroundMethodInfluentialSalimzadeh, S., He, G., & Gadiraju, U. (2024) · Proceedings of the CHI Conference on Human Factors in Computing Systems · doi:10.1145/3613904.3641905 · Cited by 41
“In addition, the attributes of AI systems can enhance decision-making outcomes [68], which include aspects such as predictions generated by AI [66, 76, 99], information about AI predictions or AI systems themselves [13, 70, 118, 136], and interventions that impact cognitive processes [65, 99, 105].”
BackgroundFIND: Human-in-the-Loop Debugging Deep Text Classifiers
Lertvittayakumjorn, P., Specia, L., & Toni, F. (2020) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2020.emnlp-main.24 · Cited by 55
Designing Interactive Transfer Learning Tools for ML Non-Experts
Mishra, S., & Rzeszotarski, J. M. (2021) · Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3411764.3445096 · Cited by 32
“Several user studies have been conducted to investigate how individuals and teams build ML models [57], select features [55], iterate over data [24], interpret results and explanations [3, 36], and integrate their domain expertise [8, 10].”
BackgroundScatterShot: Interactive In-context Example Curation for Text Transformation
Wu, T., Shen, H., Weld, D. S., Heer, J., & Ribeiro, M. T. (2023) · ACM International Conference on Intelligent User Interfaces (IUI) · doi:10.1145/3581641.3584059 · Cited by 49
“Since LLMs have some capability to generalize individual examples more broadly to its semantically similar ones, we believe labeling in in-context learning would be more effective, and we use ScatterShot to reactivate labeling-based IML for in-context learning.”
MethodInfluentialSelf-citationModel Sketching: Centering Concepts in Early-Stage Machine Learning Model Design
Lam, M. S., Ma, Z., Li, A., Freitas, I., Wang, D., Landay, J. A., & Bernstein, M. S. (2023) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3544548.3581290 · Cited by 33
“Our sketch model results mirror prior literature on the common challenge of overfitting tendencies in IML [11, 64], especially given that we did not take steps to explicitly steer users away from overfitting behavior.”
ResultInfluentialAssessment of supervised classifiers for the task of detecting messages with suicidal ideation
Caicedo, R. W. A., Gomez, J. M., & Sasieta, H. A. M. (2020) · Heliyon · doi:10.1016/j.heliyon.2020.e04412 · Cited by 23
Experimental Results on Customer Reviews Using Lexicon-Based Word Polarity Identification Method
Abdalgader, K., & Shibli, A. A. (2020) · IEEE Access · doi:10.1109/access.2020.3028260 · Cited by 25
“Various polarity identification methods have been proposed in literature to determine and classify the sentiment orientation in a document-level text [16]–[18], sentence-level text [19], and word, or feature level [20].”
BackgroundOn Interactive Machine Learning and the Potential of Cognitive FeedbackPreprint
Michael, C. J., Acklin, D., & Scheuerman, J. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2003.10365 · Cited by 20
Understanding User Sensemaking in Machine Learning Fairness Assessment Systems
Gu, Z., Yan, J. N., & Rzeszotarski, J. M. (2021) · International World Wide Web Conference (WWW) · doi:10.1145/3442381.3450092 · Cited by 20
“Recent projects have explored how to mitigate cognitive biases [58, 69] and promote a more complete understanding of ML [30, 75].”
BackgroundDeepa, D. (2021) · Turkish Journal of Computer and Mathematics Education (TURCOMAT) · doi:10.17762/turcomat.v12i7.3055 · Cited by 13
EBBE-Text: Explaining Neural Networks by Exploring Text Classification Decision Boundaries
Delaforge, A., Azé, J., Bringay, S., Mollévi, C., Sallaberry, A., & Servajean, M. (2022) · IEEE Transactions on Visualization and Computer Graphics · doi:10.1109/tvcg.2022.3184247 · Cited by 17
“actions of users after having used an active learning method involving post-hoc explanation tools could have a positive [46] or negative [47] impact on model performance.”
MethodJin, N., Wen, L., & Xie, K. (2022) · Computational Intelligence and Neuroscience · doi:10.1155/2022/2538896 · Cited by 20
“studied local decision-making in feature selection of emotion classification task and analyzed the influence of interactive machine learning tools on feature selection results [22].”
BackgroundBootstrapping semi-supervised annotation method for potential suicidal messages
Caicedo, R. W. A., Gomez, J. M., & Sasieta, H. A. M. (2022) · Internet Interventions · doi:10.1016/j.invent.2022.100519 · Cited by 12
“…out are usually focused on a single data source: from Twitter (Jashinsky et al., 2014; Mowery, 2017; O ’ Dea et al., 2015; Purver et al., 2012; Wu et al., 2019), Weibo (Huang et al., 2014; Zhang et al., 2015), Netlog (Desmet and Hoste, 2018), other microblogs (Guan et al., 2015), and more…”
BackgroundLIMEADE: From AI Explanations to Advice Taking
Lee, B. C. G., Downey, D., Lo, K., & Weld, D. S. (2023) · ACM Transactions on Interactive Intelligent Systems · doi:10.1145/3589345 · Cited by 14
Self-citationInterpretation of Sentiment Analysis with Human-in-the-Loop
Yeruva, V. K., Chandrashekar, M., Lee, Y., Rydberg-Cox, J. A., Blanton, V., & Oyler, N. A. (2020) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata50022.2020.9378221 · Cited by 12
“[1], [2] It is essential to realize the advantages of human feedback and the pressing need for manual annotation to improve machine learning performance.”
BackgroundInteractive Machine Learning for Embodied Interaction Design: A tool and methodology
Plant, N., Hilton, C., Gillies, M. F. P., Fiebrink, R. A., Perry, P., Díaz, C. G., Gibson, R., Martelli, B., & Zbyszyński, M. (2021) · International Conference on Tangible, Embedded, and Embodied Interaction (TEI) · doi:10.1145/3430524.3442703 · Cited by 9
“It has also been found that iteration without this grounding knowledge does not lead to better models[Wu et al. 2019].”
BackgroundWang, L., Yurechko, K., Dani, P., Chen, Q. Z., & Zhang, A. X. (2025) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3706598.3713691 · Cited by 25
Human Expectations and Perceptions of Learning in Machine Teaching
Mishra, S., & Rzeszotarski, J. M. (2023) · ACM Conference on User Modeling, Adaptation and Personalization (UMAP) · doi:10.1145/3565472.3595612 · Cited by 8
Antar, A. D., Molaei, S., Chen, Y.-Y., Lee, M. L., & Banović, N. (2024) · ACM Symposium on User Interface Software and Technology (UIST) · doi:10.1145/3654777.3676323 · Cited by 11
FIND: Human-in-the-Loop Debugging Deep Text ClassifiersPreprint
Lertvittayakumjorn, P., Specia, L., & Toni, F. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.04987 · Cited by 16
“However, such local improvements for individual predictions could add up to inferior overall performance (Wu et al., 2019).”
BackgroundMulti-objective Hybrid Optimization-based Feature Selection for Sentiment Analysis
Zingade, D. S., Deshmukh, R. K., & Kadam, D. B. (2023) · International Conference for Emerging Technology (INCET) · doi:10.1109/incet57972.2023.10170147 · Cited by 2
“Wrapper-based feature selection algorithms outperform filter-based techniques, as proven in [16].”
MethodMuralidhar, R., & Hulipalled, V. R. (2023) · Foundations of Computing and Decision Sciences · doi:10.2478/fcds-2023-0008 · Cited by 4
Interpretation of Sentiment Analysis in Aeschylus's Greek Tragedy
2020, T. 2. I. C. O. C. L. (2020) · Underline Science Inc. · doi:10.48448/37z4-1g96 · Cited by 2
Yu, J. (2022) · Lecture notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering · doi:10.1007/978-3-030-94551-0_12 · Cited by 1
Sentiment Analysis using Multi-objective Optimization-based Feature Selection Approach
Zingade, D. S., Deshmukh, R. K., & Kadam, D. B. (2023) · International Conference for Emerging Technology (INCET) · doi:10.1109/incet57972.2023.10169912 · Cited by 1
“In [16], wrapper-based feature selection algorithms have been demonstrated to outperform filter-based strategies.”
MethodWu, T. (2021) · CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA) · doi:10.1145/3411763.3443423
“sentiment analysis model by adding or removing individual word features for 50 iterations as they saw ft [35], I observed that such local feature selection (a technique supported and advocated for in IML research) failed to improve models on average, and sometimes hampered generalization on the test set.”
BackgroundMethodSelf-citationDavis, C. W. H. (2022) · PDXScholar (Portland State University) · doi:10.15760/etd.7878 · Cited by 1
Analysing Language-Based Cyber Threats: An Examination of Cybersecurity for Chinese Text
Mills, B. (2024) · The Pinnacle A Journal by Scholar-Practitioners · doi:10.61643/c78533
Orbit: A Framework for Designing and Evaluating Multi-objective Rankers
Yang, C., Xiao, T., Shavlovsky, M., Kästner, C., & Wu, T. (2025) · ACM International Conference on Intelligent User Interfaces (IUI) · doi:10.1145/3708359.3712092 · Cited by 1
Self-citationInteractive Data Mining with Machine Learning: User-Centric Approaches and Tools
Deepthi, P., Bhavani, M., RubenRaju, K., Sindhuja, B., Madhavi, A., & Nandini, S. (2025) · Lecture notes in electrical engineering · doi:10.1007/978-981-95-0269-1_178
Defining Biases in Visual Data Analysis
Wall, E. (2026) · Synthesis lectures on visualization · doi:10.1007/978-3-032-09307-3_3
Metadata
- Authors
- Wu, T., Weld, D. S., & Heer, J.
- Affiliations
- University of Washington (US)
- Venue
- ACM Transactions on Computer-Human Interaction
- Published
- 2019-06-17
- Volume
- 26
- Issue
- 4
- Pages
- 1-27
- ISSN
- 1073-0516 (ISSN-L)
- 1557-7325
- OpenAlex
- W2949832227
- Cited by
- 39
- References
- 67
- Citations by year
- 2026: 1 · 2025: 3 · 2024: 3 · 2023: 9 · 2022: 6 · 2021: 9 · 2020: 8
- FWCI
- 2.62
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- ACM Transactions on Computer-Human Interaction (publishedVersion)
- Funders
- Office of Naval Research (N00014-18-1-2193)
- Washington Research Foundation (Cable Professorship)
- Moore Foundation (Data-Driven Discovery Investigator program)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Algorithms
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Physical Sciences › Computer Science › Artificial Intelligence › Sentiment Analysis and Opinion Mining
- Keywords
- interactive machine learning, local decision-making, feature selection, sentiment classification, overfitting, model generalization, human-in-the-loop learning, training data size, learning algorithm comparison
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.