Common daily tasks require us to estimate surface area. Yet, area judgments are substantially and consistently biased: For example, triangles appear larger than same-area squares and disks. Previous work has explored small subsets of shapes and related biases in area perception to one or two geometric features, such as height or compactness. However, a broader understanding of shape-related biases is lacking. Here, we quantify biases in area perception for a wide variety of shapes and explain them in terms of geometric features. In four online experiments (each N = 35), typical adult observers made two-alternative forced choice judgments ("which stimulus has larger area?") for pairs of stimuli of different shape, orientation, and/or area. We found clear shape-related biases that replicate known biases and extend them to novel shapes. We provide a multipredictor model (R² = .96) that quantitatively predicts biases in perceived area across 22 shape/orientation combinations. (PsycInfo Database Record (c) 2025 APA, all rights reserved).
Taxonomies provide essential hierarchical structures for classifying information, enabling effective retrieval and knowledge organization in diverse domains such as e-commerce, academic research, and web search. Traditional taxonomy construction, heavily reliant on manual curation by domain experts, faces significant challenges in scalability, cost, and consistency when dealing with the exponential growth of digital data. Recent advancements in Large Language Models (LLMs) and Natural Language Processing (NLP) present powerful opportunities for automating this complex process. This paper explores the potential of LLMs for automated taxonomy generation, focusing on methodologies incorporating semantic embedding generation, keyword extraction, and machine learning clustering algorithms. We specifically investigate and conduct a comparative analysis of two primary LLM-based approaches using a dataset of eBay product descriptions. The first approach involves fine-tuning a pre-trained LLM using structured hierarchical data derived from chain-of-layer clustering outputs. The second employs prompt-engineering techniques to guide LLMs in generating context-aware hierarchical taxonomies based on clustered keywords without explicit model retraining. Both methodologies are evaluated for their efficacy in constructing organized multi-level hierarchical taxonomies. Evaluation using semantic similarity metrics (BERTScore and Cosine Similarity) against a ground truth reveals that the fine-tuning approach yields higher overall accuracy and consistency (BERTScore F1: 70.91%; Cosine Similarity: 66.40%) compared to the prompt-engineering approach (BERTScore F1: 61.66%; Cosine Similarity: 60.34%). We delve into the inherent trade-offs between these methods concerning semantic fidelity, computational resource requirements, result stability, and scalability. Finally, we outline potential directions for future research aimed at refining LLM-based taxonomy construction systems to handle large dynamic datasets with enhanced accuracy, robustness, and granularity.
1 January 2025 in Journal of Advances in Information Technology
JAIT is a scientific open access journal which focuses on empirical research results and critical analysis of technology development, use, management and impacts in information technology.
When picking up objects, we prefer stable grips with minimal torque by seeking grasp points that straddle the object’s centre of mass (CoM). For homogeneous objects, the CoM is at the geometric centre (GC), computable from shape cues. However, everyday objects often include components of different materials and densities. In this case, the CoM depends on the object’s geometry and the components’ densities. We asked how participants estimate the CoM of novel, two-part objects. Across four experiments, participants used a precision grip to lift cylindrical objects comprised of steel and PVC in varying proportions (steel three times denser than PVC). In all experiments, initial grasps were close to objects’ GCs; neither every-day experience (metals are denser than PVC) nor pre-exposure to the stimulus materials in isolation moved first grasps away from the GC. Within a few trials, however, grasps shifted towards the CoM, reducing but not eliminating torque. Learning transferred across the stimulus set, that is, observers learnt the materials’ densities (or their ratio) rather than learning each object’s CoM. In addition, there was a stable ‘under-reaching’ bias towards the grasping hand. An ‘inverted density’ stimulus set (PVC 3 × denser than steel) induced similarly fast learning, confirming that prior knowledge of materials has little effect on grasp point selection. When stimulus sets were covertly switched during an experiment, the unexpected force feedback caused even faster grasp adaptation. Torque minimisation is a strong driver of grasp point adaptation, but there is a surprising lack of transfer following pre-exposure to relevant materials.