L ​ i ​ k ​ e ​ T ​ h ​ i ​ s ! LikeThis! Empowering App Users to Submit UI Improvement Suggestions Instead of Complaints Conference: 2026 IEEE/ACM 48th International Conference on Software Engineering; April 12–18, 2026; Rio de Janeiro, Brazil 2026 IEEE/ACM 48th International Conference on Software Engineering (ICSE ’26), April 12–18, 2026, Rio de Janeiro, Brazil DOI: 10.1145/3744916.3787834 ISBN: 979-8-4007-2025-3/2026/04 CCS: Software and its engineering Requirements analysis CCS: Human-centered computing Graphical user interfaces

Jialiang Wei, Ali Ebrahimi Pourasad, Walid Maalej · 2026

Read L ​ i ​ k ​ e ​ T ​ h ​ i ​ s ! LikeThis! Empowering App Users to Submit UI Improvement Suggestions Instead of Complaints Conference: 2026 IEEE/ACM 48th International Conference on Software Engineering; April 12–18, 2026; Rio de Janeiro, Brazil 2026 IEEE/ACM 48th International Conference on Software Engineering (ICSE ’26), April 12–18, 2026, Rio de Janeiro, Brazil DOI: 10.1145/3744916.3787834 ISBN: 979-8-4007-2025-3/2026/04 CCS: Software and its engineering Requirements analysis CCS: Human-centered computing Graphical user interfaces by Jialiang Wei, Ali Ebrahimi Pourasad, Walid Maalej (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-04245.txt.

Introduction

Wei and others present LikeThis!, empowering app users to submit UI gain suggestions instead of complaints. User feedback is crucial for mobile app evolution. However, research suggests users often submit uninformative, vague, or destructive feedback. To evaluate LikeThis!, they first ran a model benchmarking study using a public set of carefully critiqued UI designs. An intermediate step generates a answer specification before sketching the design, key to effective gain.

Catalog record

CREATORJialiang Wei, Ali Ebrahimi Pourasad, Walid Maalej
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE65136
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.04245

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

L ​ i ​ k ​ e ​ T ​ h ​ i ​ s ! LikeThis! Empowering App Users to Submit UI Improvement Suggestions Instead of Complaints Conference: 2026 IEEE/ACM 48th International Conference on Software Engineering; April 12–18, 2026; Rio de Janeiro, Brazil 2026 IEEE/ACM 48th International Conference on Software Engineering (ICSE ’26), April 12–18, 2026, Rio de Janeiro, Brazil DOI: 10.1145/3744916.3787834 ISBN: 979-8-4007-2025-3/2026/04 CCS: Software and its engineering Requirements analysis CCS: Human-centered computing Graphical user interfaces Jialiang Wei, Ali Ebrahimi Pourasad, Walid Maalej Abstract . User feedback is crucial for the evolution of mobile apps. However, research suggests that users tend to submit uninformative, vague, or destructive feedback. Unlike recent AI4SE approaches that focus on generating code and other development artifacts, our work aims at empowering users to submit better and more constructive UI feedback with concrete suggestions on how to improve the app. We propose L ​ i ​ k ​ e ​ T ​ h ​ i ​ s ! LikeThis! , a GenAI-based approach that takes a user comment with the corresponding screenshot to immediately generate multiple improvement alternatives, from which the user can easily choose their preferred option. To evaluate L ​ i ​ k ​ e ​ T ​ h ​ i ​ s ! LikeThis! , we first conducted a model benchmarking study based on a public dataset of carefully critiqued UI designs. The results show that GPT-Image-1 significantly outperformed three other state-of-the-art image generation models in improving the designs to address UI issues while keeping the fidelity and without introducing new issues. An intermediate step in L ​ i ​ k ​ e ​ T ​ h ​ i ​ s ! LikeThis! to generate a solution specification before changing the design was key to achieving effective improvement. Second, we conducted a user study with 10 production apps, where 15 users used L ​ i ​ k ​ e ​ T ​ h ​ i ​ s ! LikeThis! to submit their feedback on encountered issues. Later, the developers of the apps assessed the understandability and actionability of the feedback with and without generated improvements. The results show that our approach helps generate better feedback from both user and developer perspectives, paving the way for AI-assisted user-developer collaboration. 1. Introduction In the highly competitive app market, user satisfaction is a prerequisite for app success ( 51 ; 40 ; 62 ) . User feedback thus represents a crucial direct channel for developers to understand user needs, identify and fix bugs, or gather suggestions for enhancements ( 26 ) . App stores enable users to easily submit feedback as a text review and a 1-5 star rating. However, this might sometimes be insufficient to convey the specific thoughts and ideas of users. Research has shown that, while most reviews tend to be destructive or uninformative, many users are motivated to contribute in-depth, constructive feedback ( 62 ; 25 ; 37 ) . Their ability to do so is, however, often constrained by language barriers in text communication, including mismatches in concepts and vocabulary between users and developers ( 85 ; 24 ) . Even verbose user feedback can be too vague for developers to understand or might lack key details to act upon ( 49 ; 50 ) . Visual feedback pointing to the user interface (UI) helps communicate user needs and thoughts more precisely ( 48 ) . Visual mockups can convey specific ideas, e.g. for a layout change or a new button with a clarity that is difficult to achieve with text alone. While their benefits in requirements and design processes are well-documented, it remains challenging for users to easily create or edit UI mockups ( 59 ; 39 ) . Powerful design tools, such as Figma ( 18 ) and Sketch ( 66 ) , are built for professional designers and developers, not for average users, who simply want to easily and quickly share their feedback. Beyond text and code generation ( 33 ) , recent advance in Generative AI (GenAI) have made the creation of images ( 41 ; 29 ) from text prompts very easy. Researchers have already investigated the generation of UI from text prompts, but so far have focused only on generating new UIs instead of editing existing ones ( 76 ; 74 ) . Code-based UI generation ( 45 ; 83 ) is restricted to specific UI technologies and requires the comprehension of code—thus targeting rather developers and professional designers. To the best of our knowledge, no work has investigated how to enable users to edit existing UIs while sharing in-situ feedback. To address this gap, we propose LikeThis! , the first GenAI-based approach to empower users to directly submit improvement suggestions that address UI-related issues they encounter. LikeThis! takes a brief user comment and a screenshot as input to first generate three alternative design specifications that address the issue encountered (“Suggestion Generation”). Then, our approach edits the original UI image according to those specifications, generating three UIs that address the issue. These images are then presented to the user, who can finally choose and submit the one that best fits their idea. We conducted two extensive studies to evaluate our approach: the first focuses on GenAI models’ capabilities for UI improvement and the second on how effective and helpful is our approach for users and developers. Our research questions are as follows: RQ1:: How well can GenAI improve a UI that includes an issue? : RQ1.1: How do state-of-the-art image generation models perform in generating UI improvements addressing specific issues? : RQ1.2: Does the use of masking to point to a problematic area in the UI improve the results? : RQ1.3: What is the impact of the intermediate “Suggestion Generation” step of LikeThis! on the results? RQ2:: Does our approach help create better user feedback? : RQ2.1: From the user perspective , how accurate are the generated suggestions and how helpful is the whole approach? : RQ2.2: From the developer perspective , does LikeThis! improve the received feedback, particularly in terms of understandability and actionability? To answer RQ1, we conducted a benchmarking study based on a public dataset of carefully evaluated mobile UIs ( 14 ) . We found that GPT-Image-1 ( 60 ) clearly outperforms three other state-of-the-art image generation models in addressing UI issues, while maintaining high design fidelity and rarely introducing new issues. Flux ( 38 ) , Gemini ( 21 ) , and Bagel ( 12 ) are able to address the issue, at least partly, only in 25-53% of the evaluated screens. When evaluating the impact of masking (i.e., highlighting the concerned region in the UI using a binary image with black and white), we observed that it only works well when the problematic area is small. For large areas, masking often negatively impacts fidelity and robustness to new issues. We also found that our intermediate “Suggestion Generation” step was crucial for achieving top performance. To answer RQ2, we asked 15 users to use 10 production apps, identify issues, and submit feedback using LikeThis! . Participants agreed that 85.5% of generated suggestions were mostly or very accurate. The corresponding app developers assessed the submitted feedback, confirming that our

Read the complete text