Automatic systems are increasingly used to assess the originality of responses in creative tasks. They offer a potential solution to key limitations of human assessment (cost, fatigue, and subjectivity), but there is preliminary evidence of a self-preference bias. Accordingly, automatic systems tend to prefer outcomes that are more closely related to their style, rather than to the human one. In this paper, we investigated how Large Language Models (LLMs) align with human raters in assessing the originality of responses in a divergent thinking task. We analysed 4,813 responses to the Alternate Uses Task produced by higher and lower creative humans and ChatGPT-4o. Human raters were two university students who underwent intensive training. Machine raters were two specialised systems fine-tuned on AUT responses and corresponding human ratings (OCSAI and CLAUS) and ChatGPT-4o, which was prompted with the same instructions as human raters. Results confirmed the presence of a self-preference bias in LLMs. Automatic systems tended to privilege artificial responses. However, this self-preference bias disappeared when the analyses controlled for the idea elaboration. We discuss theoretical and methodological implications of these findings by highlighting future directions for research on creativity assessment.

• Extends the prior NLBSE tool competition work with an additional model family (SetFit). • Transforms the prototype into a reusable tool ( Beyond Balance ), providing easy-to-use access to the implemented loss-weighting strategies. • Compares the performance impact of loss-weighting across various strategies and model families. • Evaluates the generalizability of loss-weighting across two benchmark datasets from the NLBSE’24 and NLBSE’25 tool competitions. Datasets often contain heavily underrepresented classes. Class imbalance biases models toward frequent classes, reducing performance on rare but important categories; in-process strategies such as loss-weighting remain under-explored for software engineering artefacts. We investigate loss-weighting functions for code comment classification and package our methods into Beyond Balance , a reusable implementation offering multiple weighting strategies for Transformer- and Sentence-Transformer–based models. Loss weighting consistently improves F1 performance across datasets, demonstrating an effective and easily adoptable imbalance-handling technique through Beyond Balance .

Multi-task learning is a paradigm that leverages information from related tasks to improve the performance of machine learning. Self-Admitted Technical Debt (SATD) are comments in the code that indicate not-quite-right code introduced for short-term needs, i.e., technical debt (TD). Previous research has provided evidence of a possible relationship between SATD and the existence of vulnerabilities in the code. In this work, we investigate if multi-task learning could leverage the information shared between SATD and vulnerabilities to improve the automatic detection of these issues. To this aim, we implemented VulSATD, a deep learner that detects vulnerable and SATD code based on CodeBERT, a pre-trained transformers model. We evaluated VulSATD on MADE-WIC, a fused dataset of functions annotated for TD (through SATD) and vulnerability. We compared the results using single and multi-task approaches, obtaining no significant differences even after employing a weighted loss. Our findings indicate the need for further investigation into the relationship between these two aspects of low-quality code. Specifically, it is possible that only a subset of technical debt is directly associated with security concerns. Therefore, the relationship between different types of technical debt and software vulnerabilities deserves future exploration and a deeper understanding.

In this paper, we present MADE-WIC, a large dataset of functions and their comments with multiple annotations for technical debt and code weaknesses leveraging different state-of-the-art approaches. It contains about 860K code functions and more than 2.7M related comments from 12 open-source projects. To the best of our knowledge, no such dataset is publicly available. MADE-WIC aims to provide researchers with a curated dataset on which to test and compare tools designed for the detection of code weaknesses and technical debt. As we have fused existing datasets, researchers have the possibility to evaluate the performance of their tools by also controlling the bias related to the annotation definition and dataset construction. The demonstration video can be retrieved at https://www.youtube.com/watch?v=GaQodPrcb6E.

Speeding up development may produce technical debt, i.e., not-quite-right code for which the effort to make it right increases with time as a sort of interest. Developers may be aware of the debt as they admit it in their code comments. Literature reports that such a self-admitted technical debt survives for a long time in a program, but it is not yet clear its impact on the quality of the code in the long term. We argue that self-admitted technical debt contains a number of different weaknesses that may affect the security of a program. Therefore, the longer a debt is not paid back the higher is the risk that the weaknesses can be exploited. To discuss our claim and rise the developers' awareness of the vulnerability of the self-admitted technical debt that is not paid back, we explore the self-admitted technical debt in the Chromium C-code to detect any known weaknesses. In this preliminary study, we first mine the Common Weakness Enumeration repository to define heuristics for the automatic detection and fix of weak code. Then, we parse the C-code to find self-admitted technical debt and the code block it refers to. Finally, we use the heuristics to find weak code snippets associated to self-admitted technical debt and recommend their potential mitigation to developers. Such knowledge can be used to prioritize self-admitted technical debt for repair. A prototype has been developed and applied to the Chromium code. Initial findings report that 55% of self-admitted technical debt code contains weak code of 14 different types.