Chun Yong Chong

Daniel Hao Xian Yuen, Andrew Yong Chen Pang, Zhou Yang et al.

Recent years have witnessed wider adoption of Automated Speech Recognition (ASR) techniques in various domains. Consequently, evaluating and enhancing the quality of ASR systems is of great importance. This paper proposes ASDF, an Automated Speech Recognition Differential Testing Framework for testing ASR systems. ASDF extends an existing ASR testing tool, the CrossASR++, which synthesizes test cases from a text corpus. However, CrossASR++ fails to make use of the text corpus efficiently and provides limited information on how the failed test cases can improve ASR systems. To address these limitations, our tool incorporates two novel features: (1) a text transformation module to boost the number of generated test cases and uncover more errors in ASR systems and (2) a phonetic analysis module to identify on which phonemes the ASR system tend to produce errors. ASDF generates more high-quality test cases by applying various text transformation methods (e.g., change tense) to the texts in failed test cases. By doing so, ASDF can utilize a small text corpus to generate a large number of audio test cases, something which CrossASR++ is not capable of. In addition, ASDF implements more metrics to evaluate the performance of ASR systems from multiple perspectives. ASDF performs phonetic analysis on the identified failed test cases to identify the phonemes that ASR systems tend to transcribe incorrectly, providing useful information for developers to improve ASR systems. The demonstration video of our tool is made online at https://www.youtube.com/watch?v=DzVwfc3h9As. The implementation is available at https://github.com/danielyuenhx/asdf-differential-testing.

Shangyu Li, Juyong Jiang, Meibo Ren et al.

Transpilation, or code translation, aims to convert source code from one programming language (PL) to another. It is beneficial for many downstream applications, from modernizing large legacy codebases to augmenting data for low-resource PLs. Recent large language model (LLM)-based approaches have demonstrated immense potential for code translation. Among these approaches, training-based methods are particularly important because LLMs currently do not effectively adapt to domain-specific settings that suffer from a lack of knowledge without targeted training. This limitation is evident in transpilation tasks involving low-resource PLs. However, existing training-based approaches rely on a pairwise transpilation paradigm, making it impractical to support a diverse range of PLs. This limitation is particularly prominent for low-resource PLs due to a scarcity of training data. Furthermore, these methods suffer from suboptimal reinforcement learning (RL) reward formulations. To address these limitations, we propose CodePivot, a training framework that leverages Python as an intermediate representation (IR), augmented by a novel RL reward mechanism, Aggressive-Partial-Functional reward, to bootstrap the model's multilingual transpilation ability without requiring parallel corpora. Experiments involving 10 PLs show that the resulting 7B model, trained on Python-to-Others tasks, consistently improves performance across both general and low-resource PL-related transpilation tasks. It outperforms substantially larger mainstream models with hundreds of billions more parameters, such as Deepseek-R1 and Qwen3-235B-A22B-Instruct-2507, on Python-to-Others tasks and Others-to-All tasks, respectively. In addition, it outperforms its counterpart trained directly on Any-to-Any tasks on general transpilation tasks. The code and data are available at https://github.com/lishangyu-hkust/CodePivot.

Muxin Pu, Meng Yi Kuan, Nyee Thoang Lim et al.

Fairness of deepfake detectors in the presence of anomalies are not well investigated, especially if those anomalies are more prominent in either male or female subjects. The primary motivation for this work is to evaluate how deepfake detection model behaves under such anomalies. However, due to the black-box nature of deep learning (DL) and artificial intelligence (AI) systems, it is hard to predict the performance of a model when the input data is modified. Crucially, if this defect is not addressed properly, it will adversely affect the fairness of the model and result in discrimination of certain sub-population unintentionally. Therefore, the objective of this work is to adopt metamorphic testing to examine the reliability of the selected deepfake detection model, and how the transformation of input variation places influence on the output. We have chosen MesoInception-4, a state-of-the-art deepfake detection model, as the target model and makeup as the anomalies. Makeups are applied through utilizing the Dlib library to obtain the 68 facial landmarks prior to filling in the RGB values. Metamorphic relations are derived based on the notion that realistic perturbations of the input images, such as makeup, involving eyeliners, eyeshadows, blushes, and lipsticks (which are common cosmetic appearance) applied to male and female images, should not alter the output of the model by a huge margin. Furthermore, we narrow down the scope to focus on revealing potential gender biases in DL and AI systems. Specifically, we are interested to examine whether MesoInception-4 model produces unfair decisions, which should be considered as a consequence of robustness issues. The findings from our work have the potential to pave the way for new research directions in the quality assurance and fairness in DL and AI systems.

Muxin Pu, Chun Yong Chong, Mei Kuan Lim

Hand pose estimation (HPE) is a task that predicts and describes the hand poses from images or video frames. When HPE models estimate hand poses captured in a laboratory or under controlled environments, they normally deliver good performance. However, the real-world environment is complex, and various uncertainties may happen, which could degrade the performance of HPE models. For example, the hands could be occluded, the visibility of hands could be reduced by imperfect exposure rate, and the contour of hands prone to be blurred during fast hand movements. In this work, we adopt metamorphic testing to evaluate the robustness of HPE models and provide suggestions on the choice of HPE models for different applications. The robustness evaluation was conducted on four state-of-the-art models, namely MediaPipe hands, OpenPose, BodyHands, and NSRM hand. We found that on average more than 80\% of the hands could not be identified by BodyHands, and at least 50\% of hands could not be identified by MediaPipe hands when diagonal motion blur is introduced, while an average of more than 50\% of strongly underexposed hands could not be correctly estimated by NSRM hand. Similarly, applying occlusions on only four hand joints will also largely degrade the performance of these models. The experimental results show that occlusions, illumination variations, and motion blur are the main obstacles to the performance of existing HPE models. These findings may pave the way for researchers to improve the performance and robustness of hand pose estimation models and their applications.

Nyee Thoang Lim, Meng Yi Kuan, Muxin Pu et al.

Deepfakes utilise Artificial Intelligence (AI) techniques to create synthetic media where the likeness of one person is replaced with another. There are growing concerns that deepfakes can be maliciously used to create misleading and harmful digital contents. As deepfakes become more common, there is a dire need for deepfake detection technology to help spot deepfake media. Present deepfake detection models are able to achieve outstanding accuracy (>90%). However, most of them are limited to within-dataset scenario, where the same dataset is used for training and testing. Most models do not generalise well enough in cross-dataset scenario, where models are tested on unseen datasets from another source. Furthermore, state-of-the-art deepfake detection models rely on neural network-based classification models that are known to be vulnerable to adversarial attacks. Motivated by the need for a robust deepfake detection model, this study adapts metamorphic testing (MT) principles to help identify potential factors that could influence the robustness of the examined model, while overcoming the test oracle problem in this domain. Metamorphic testing is specifically chosen as the testing technique as it fits our demand to address learning-based system testing with probabilistic outcomes from largely black-box components, based on potentially large input domains. We performed our evaluations on MesoInception-4 and TwoStreamNet models, which are the state-of-the-art deepfake detection models. This study identified makeup application as an adversarial attack that could fool deepfake detectors. Our experimental results demonstrate that both the MesoInception-4 and TwoStreamNet models degrade in their performance by up to 30\% when the input data is perturbed with makeup.

Heng Ee Tay, Mei Kuan Lim, Chun Yong Chong

Conventional approaches to identify depression are not scalable, and the public has limited awareness of mental health, especially in developing countries. As evident by recent studies, social media has the potential to complement mental health screening on a greater scale. The vast amount of first-person narrative posts in chronological order can provide insights into one's thoughts, feelings, behavior, or mood for some time, enabling a better understanding of depression symptoms reflected in the online space. In this paper, we propose SERCNN, which improves the user representation by (1) stacking two pretrained embeddings from different domains and (2) reintroducing the embedding context to the MLP classifier. Our SERCNN shows great performance over state-of-the-art and other baselines, achieving 93.7% accuracy in a 5-fold cross-validation setting. Since not all users share the same level of online activity, we introduced the concept of a fixed observation window that quantifies the observation period in a predefined number of posts. With as minimal as 10 posts per user, SERCNN performed exceptionally well with an 87% accuracy, which is on par with the BERT model, while having 98% less in the number of parameters. Our findings open up a promising direction for detecting depression on social media with a smaller number of posts for inference, towards creating solutions for a cost-effective and timely intervention. We hope that our work can bring this research area closer to real-world adoption in existing clinical practice.

Muxin Pu, Xiao-Ming Wu, Mei Kuan Lim et al.

We present SignDPO, a novel multi-level Direct Preference Optimisation (DPO) framework designed to enhance the alignment of skeleton-based Sign Language Translation. While current skeleton-based models have made significant progress using Maximum Likelihood Estimation, they are primarily constrained by an imitation-based paradigm that lacks discriminative sensitivity to the fine-grained spatio-temporal nuances of sign language, often leading to semantic drift. To address this, SignDPO shifts the optimisation goal from simple sequence mimicry to structured preference alignment across spatial, temporal, and linguistic dimensions. Our framework involves three key designs. First, we introduce a hierarchical perturbation strategy to construct spatial and temporal non-preferred samples at both global and local granularities automatically. Second, we propose a self-guiding mechanism that leverages decoder cross-attention scores to identify and perturb semantically salient skeletal regions, forcing the model to distinguish genuine sign signals from structural distortions. Third, we establish an automated language-level preference generator by fine-tuning a dedicated perturbation model, capturing complex output-level failure modes without manual annotation. Extensive experiments on three widely adopted benchmarks, CSL-Daily, How2Sign, and OpenASL, demonstrate that SignDPO consistently outperforms state-of-the-art gloss-free methods and even rivals established gloss-based ones. Our results suggest that multi-level preference alignment is a powerful paradigm for bridging the gap between high-entropy skeletal trajectories and discrete linguistic semantics.

Ruiqi Wang, Xinchen Wang, Cuiyun Gao et al.

Large language models (LLMs) have been increasingly deployed in real-world software engineering, fostering the development of code evaluation metrics to study the quality of LLM-generated code. Conventional rule-based metrics merely score programs based on their surface-level similarities with reference programs instead of analyzing functionality and code quality in depth. To address this limitation, researchers have developed LLM-as-a-judge metrics, prompting LLMs to evaluate and score code, and curated various code evaluation benchmarks to validate their effectiveness. However, these benchmarks suffer from critical limitations, hindering reliable assessments of evaluation capability: Some feature coarse-grained binary labels, which reduce rich code behavior to a single bit of information, obscuring subtle errors. Others propose fine-grained but subjective, vaguely-defined evaluation criteria, introducing unreliability in manually-annotated scores, which is the ground-truth they rely on. Furthermore, they often use uncontrolled data synthesis methods, leading to unbalanced score distributions that poorly represent real-world code generation scenarios. To curate a diverse benchmark with programs of well-balanced distributions across various quality levels and streamline the manual annotation procedure, we propose AXIOM, a novel perturbation-based framework for synthesizing code evaluation benchmarks at scale. It reframes program scores as the refinement effort needed for deployment, consisting of two stages: (1) Rule-guided perturbation, which prompts LLMs to apply sequences of predefined perturbation rules to existing high-quality programs to modify their functionality and code quality, enabling us to precisely control each program's target score to achieve balanced score distributions. (2) Multisource quality calibration, which first selects a subset of...

Muxin Pu, Mei Kuan Lim, Chun Yong Chong et al.

Skeleton-based isolated sign language recognition (ISLR) demands fine-grained understanding of articulated motion across multiple spatial scales, from subtle finger movements to global body dynamics. Existing approaches typically rely on deep feed-forward architectures, which increase model capacity but lack mechanisms for recurrent refinement and structured representation. We propose LA-Sign, a looped transformer framework with geometry-aware alignment for ISLR. Instead of stacking deeper layers, LA-Sign derives its depth from recurrence, repeatedly revisiting latent representations to progressively refine motion understanding under shared parameters. To further regularise this refinement process, we present a geometry-aware contrastive objective that projects skeletal and textual features into an adaptive hyperbolic space, encouraging multi-scale semantic organisation. We study three looping designs and multiple geometric manifolds, demonstrating that encoder-decoder looping combined with adaptive Poincare alignment yields the strongest performance. Extensive experiments on WLASL and MSASL benchmarks show that LA-Sign achieves state-of-the-art results while using fewer unique layers, highlighting the effectiveness of recurrent latent refinement and geometry-aware representation learning for sign language recognition.

Alvin Jian Jia Tan, Chun Yong Chong, Aldeida Aleti

Maintenance of existing software requires a large amount of time for comprehending the source code. The architecture of a software, however, may not be clear to maintainers if up to date documentations are not available. Software clustering is often used as a remodularisation and architecture recovery technique to help recover a semantic representation of the software design. Due to the diverse domains, structure, and behaviour of software systems, the suitability of different clustering algorithms for different software systems are not investigated thoroughly. Research that introduce new clustering techniques usually validate their approaches on a specific domain, which might limit its generalisability. If the chosen test subjects could only represent a narrow perspective of the whole picture, researchers might risk not being able to address the external validity of their findings. This work aims to fill this gap by introducing a new approach, Explaining Software Clustering for Remodularisation, to evaluate the effectiveness of different software clustering approaches. This work focuses on hierarchical clustering and Bunch clustering algorithms and provides information about their suitability according to the features of the software, which as a consequence, enables the selection of the most optimum algorithm and configuration from our existing pool of choices for a particular software system. The proposed framework is tested on 30 open source software systems with varying sizes and domains, and demonstrates that it can characterise both the strengths and weaknesses of the analysed software clustering algorithms using software features extracted from the code. The proposed approach also provides a better understanding of the algorithms behaviour through the application of dimensionality reduction techniques.

Najam Nazar, Norman Chen, Chun Yong Chong

While constructing supervised learning models, we require labelled examples to build a corpus and train a machine learning model. However, most studies have built the labelled dataset manually, which in many occasions is a daunting task. To mitigate this problem, we have built an online tool called CodeLabeller. CodeLabeller is a web-based tool that aims to provide an efficient approach to handling the process of labelling source code files for supervised learning methods at scale by improving the data collection process throughout. CodeLabeller is tested by constructing a corpus of over a thousand source files obtained from a large collection of open source Java projects and labelling each Java source file with their respective design patterns and summaries. Twenty five experts in the field of software engineering participated in a usability evaluation of the tool using the standard User Experience Questionnaire online survey. The survey results demonstrate that the tool achieves the Good standard on hedonic and pragmatic quality standards, is easy to use and meets the needs of the annotating the corpus for supervised classifiers. Apart from assisting researchers in crowdsourcing a labelled dataset, the tool has practical applicability in software engineering education and assists in building expert ratings for software artefacts.

Ruiqi Wang, Jiyu Guo, Cuiyun Gao et al.

Recently, large language models (LLMs) have been deployed to tackle various software engineering (SE) tasks like code generation, significantly advancing the automation of SE tasks. However, assessing the quality of these LLM-generated code and text remains challenging. The commonly used Pass@k metric necessitates extensive unit tests and configured environments, demands a high labor cost, and is not suitable for evaluating LLM-generated text. Conventional metrics like BLEU, which measure only lexical rather than semantic similarity, have also come under scrutiny. In response, a new trend has emerged to employ LLMs for automated evaluation, known as LLM-as-a-judge. These LLM-as-a-judge methods are claimed to better mimic human assessment than conventional metrics without relying on high-quality reference answers. Nevertheless, their exact human alignment in SE tasks remains unexplored. In this paper, we empirically explore LLM-as-a-judge methods for evaluating SE tasks, focusing on their alignment with human judgments. We select seven LLM-as-a-judge methods that utilize general-purpose LLMs, alongside two LLMs specifically fine-tuned for evaluation. After generating and manually scoring LLM responses on three recent SE datasets of code translation, code generation, and code summarization, we then prompt these methods to evaluate each response. Finally, we compare the scores generated by these methods with human evaluation. The results indicate that output-based methods reach the highest Pearson correlation of 81.32 and 68.51 with human scores in code translation and generation, achieving near-human evaluation, noticeably outperforming ChrF++, one of the best conventional metrics, at 34.23 and 64.92. Such output-based methods prompt LLMs to output judgments directly, and exhibit more balanced score distributions that resemble human score patterns. Finally, we provide...

Wei Hung Pan, Ming Jie Chok, Jonathan Leong Shan Wong et al.

Educators are increasingly concerned about the usage of Large Language Models (LLMs) such as ChatGPT in programming education, particularly regarding the potential exploitation of imperfections in Artificial Intelligence Generated Content (AIGC) Detectors for academic misconduct. In this paper, we present an empirical study where the LLM is examined for its attempts to bypass detection by AIGC Detectors. This is achieved by generating code in response to a given question using different variants. We collected a dataset comprising 5,069 samples, with each sample consisting of a textual description of a coding problem and its corresponding human-written Python solution codes. These samples were obtained from various sources, including 80 from Quescol, 3,264 from Kaggle, and 1,725 from LeetCode. From the dataset, we created 13 sets of code problem variant prompts, which were used to instruct ChatGPT to generate the outputs. Subsequently, we assessed the performance of five AIGC detectors. Our results demonstrate that existing AIGC Detectors perform poorly in distinguishing between human-written code and AI-generated code.

Guodong Fan, Cuiyun Gao, Chun Yong Chong et al.

The rapid evolution of software services poses substantial challenges to the design and implementation of effective recommendation systems. Traditional service recommendation approaches often rely on static representations and historical usage data, which are insufficient for adapting to the dynamic and evolving nature of service ecosystems. Recently, large language models (LLMs) have shown strong potential to overcome these limitations by leveraging rich contextual understanding. However, their practical use faces two major challenges: outdated service facts and invalid or redundant services. To address these issues, we propose EVOREC, an evolution-aware framework for service recommendation that leverages model editing in a locate-then-edit paradigm to incorporate updated service facts without costly retraining efficiently. This allows the model to remain aligned with evolving service ecosystems. To address invalid service issues, we introduce a Finite Automata (FA)-based constrained decoding mechanism with deduplication, which enforces structural and semantic validity while eliminating repeated services. Experiments on real-world service datasets demonstrate that our framework consistently outperforms existing baselines, e.g., achieving an average relative improvement of 25.9% in Recall@5. Moreover, under evolving service scenarios, our approach outperforms model fine-tuning approaches by 22.3%, demonstrating strong adaptability to service evolution and providing a practical solution for service recommendation in dynamic ecosystems

Shuzheng Gao, Chaozheng Wang, Cuiyun Gao et al.

Test cases are essential for validating the reliability and quality of software applications. Recent studies have demonstrated the capability of Large Language Models (LLMs) to generate useful test cases for given source code. However, the existing work primarily relies on human-written plain prompts, which often leads to suboptimal results since the performance of LLMs can be highly influenced by the prompts. Moreover, these approaches use the same prompt for all LLMs, overlooking the fact that different LLMs might be best suited to different prompts. Given the wide variety of possible prompt formulations, automatically discovering the optimal prompt for each LLM presents a significant challenge. Although there are methods on automated prompt optimization in the natural language processing field, they are hard to produce effective prompts for the test case generation task. First, the methods iteratively optimize prompts by simply combining and mutating existing ones without proper guidance, resulting in prompts that lack diversity and tend to repeat the same errors in the generated test cases. Second, the prompts are generally lack of domain contextual knowledge, limiting LLMs' performance in the task.

Muxin Pu, Mei Kuan Lim, Chun Yong Chong

Sign language recognition (SLR) refers to interpreting sign language glosses from given videos automatically. This research area presents a complex challenge in computer vision because of the rapid and intricate movements inherent in sign languages, which encompass hand gestures, body postures, and even facial expressions. Recently, skeleton-based action recognition has attracted increasing attention due to its ability to handle variations in subjects and backgrounds independently. However, current skeleton-based SLR methods exhibit three limitations: 1) they often neglect the importance of realistic hand poses, where most studies train SLR models on non-realistic skeletal representations; 2) they tend to assume complete data availability in both training or inference phases, and capture intricate relationships among different body parts collectively; 3) these methods treat all sign glosses uniformly, failing to account for differences in complexity levels regarding skeletal representations. To enhance the realism of hand skeletal representations, we present a kinematic hand pose rectification method for enforcing constraints. Mitigating the impact of missing data, we propose a feature-isolated mechanism to focus on capturing local spatial-temporal context. This method captures the context concurrently and independently from individual features, thus enhancing the robustness of the SLR model. Additionally, to adapt to varying complexity levels of sign glosses, we develop an input-adaptive inference approach to optimise computational efficiency and accuracy. Experimental results demonstrate the effectiveness of our approach, as evidenced by achieving a new state-of-the-art (SOTA) performance on WLASL100 and LSA64. For WLASL100, we achieve a top-1 accuracy of 86.50\%, marking a relative improvement of 2.39% over the previous SOTA. For LSA64, we achieve a top-1 accuracy of 99.84%.

Zihan Wu, Jie Xu, Yun Peng et al.

Large Language Models (LLMs) struggle to automate real-world vulnerability detection due to two key limitations: the heterogeneity of vulnerability patterns undermines the effectiveness of a single unified model, and manual prompt engineering for massive weakness categories is unscalable. To address these challenges, we propose \textbf{MulVul}, a retrieval-augmented multi-agent framework designed for precise and broad-coverage vulnerability detection. MulVul adopts a coarse-to-fine strategy: a \emph{Router} agent first predicts the top-$k$ coarse categories and then forwards the input to specialized \emph{Detector} agents, which identify the exact vulnerability types. Both agents are equipped with retrieval tools to actively source evidence from vulnerability knowledge bases to mitigate hallucinations. Crucially, to automate the generation of specialized prompts, we design \emph{Cross-Model Prompt Evolution}, a prompt optimization mechanism where a generator LLM iteratively refines candidate prompts while a distinct executor LLM validates their effectiveness. This decoupling mitigates the self-correction bias inherent in single-model optimization. Evaluated on 130 CWE types, MulVul achieves 34.79\% Macro-F1, outperforming the best baseline by 41.5\%. Ablation studies validate cross-model prompt evolution, which boosts performance by 51.6\% over manual prompts by effectively handling diverse vulnerability patterns.

Muxin Pu, Mei Kuan Lim, Chun Yong Chong et al.

Pre-training has proven effective for learning transferable features in sign language understanding (SLU) tasks. Recently, skeleton-based methods have gained increasing attention because they can robustly handle variations in subjects and backgrounds without being affected by appearance or environmental factors. Current SLU methods continue to face three key limitations: 1) weak semantic grounding, as models often capture low-level motion patterns from skeletal data but struggle to relate them to linguistic meaning; 2) imbalance between local details and global context, with models either focusing too narrowly on fine-grained cues or overlooking them for broader context; and 3) inefficient cross-modal learning, as constructing semantically aligned representations across modalities remains difficult. To address these, we propose Sigma, a unified skeleton-based SLU framework featuring: 1) a sign-aware early fusion mechanism that facilitates deep interaction between visual and textual modalities, enriching visual features with linguistic context; 2) a hierarchical alignment learning strategy that jointly maximises agreements across different levels of paired features from different modalities, effectively capturing both fine-grained details and high-level semantic relationships; and 3) a unified pre-training framework that combines contrastive learning, text matching and language modelling to promote semantic consistency and generalisation. Sigma achieves new state-of-the-art results on isolated sign language recognition, continuous sign language recognition, and gloss-free sign language translation on multiple benchmarks spanning different sign and spoken languages, demonstrating the impact of semantically informative pre-training and the effectiveness of skeletal data as a stand-alone solution for SLU.

Hyejin Park, Taaha Waseem, Wen Qi Teo et al.

Synthesising photo-realistic images from natural language is one of the challenging problems in computer vision. Over the past decade, a number of approaches have been proposed, of which the improved Stacked Generative Adversarial Network (StackGAN-v2) has proven capable of generating high resolution images that reflect the details specified in the input text descriptions. In this paper, we aim to assess the robustness and fault-tolerance capability of the StackGAN-v2 model by introducing variations in the training data. However, due to the working principle of Generative Adversarial Network (GAN), it is difficult to predict the output of the model when the training data are modified. Hence, in this work, we adopt Metamorphic Testing technique to evaluate the robustness of the model with a variety of unexpected training dataset. As such, we first implement StackGAN-v2 algorithm and test the pre-trained model provided by the original authors to establish a ground truth for our experiments. We then identify a metamorphic relation, from which test cases are generated. Further, metamorphic relations were derived successively based on the observations of prior test results. Finally, we synthesise the results from our experiment of all the metamorphic relations and found that StackGAN-v2 algorithm is susceptible to input images with obtrusive objects, even if it overlaps with the main object minimally, which was not reported by the authors and users of StackGAN-v2 model. The proposed metamorphic relations can be applied to other text-to-image synthesis models to not only verify the robustness but also to help researchers understand and interpret the results made by the machine learning models.

Chun Yong Chong, Patanamon Thongtanunam, Chakkrit Tantithamthavorn

Code review is a widely-used practice in software development companies to identify defects. Hence, code review has been included in many software engineering curricula at universities worldwide. However, teaching code review is still a challenging task because the code review effectiveness depends on the code reading and analytical skills of a reviewer. While several studies have investigated the code reading techniques that students should use to find defects during code review, little has focused on a learning activity that involves analytical skills. Indeed, developing a code review checklist should stimulate students to develop their analytical skills to anticipate potential issues (i.e., software defects). Yet, it is unclear whether students can anticipate potential issues given their limited experience in software development (programming, testing, etc.). We perform a qualitative analysis to investigate whether students are capable of creating code review checklists, and if the checklists can be used to guide reviewers to find defects. In addition, we identify common mistakes that students make when developing a code review checklist. Our results show that while there are some misconceptions among students about the purpose of code review, students are able to anticipate potential defects and create a relatively good code review checklist. Hence, our results lead us to conclude that developing a code review checklist can be a part of the learning activities for code review in order to scaffold students' skills.