2019/10/15 by Marius Cătălin Iordan, Tyler Giallanza, Iordan, Marius Cătălin +7 · 1 citation
Computer Science · Social Sciences · #Advanced Text Analysis Techniques #Computation and Language (cs.CL) #Computational and Text Analysis Methods #FOS: Computer and information sciences #I.2.7 #Information Retrieval (cs.IR) #J.4 #Machine Learning (cs.LG) #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1910.06954
openalex publication_date 2019/10/15 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Applying machine learning algorithms to large-scale, text-based corpora\n(embeddings) presents a unique opportunity to investigate at scale how human\nsemantic knowledge is organized and how people use it to judge fundamental\nrelationships, such as similarity between concepts. However, efforts to date\nhave shown a substantial discrepancy between algorithm predictions and\nempirical judgments. Here, we introduce a novel approach of generating\nembeddings motivated by the psychological theory that semantic context plays a\ncritical role in human judgments. Specifically, we train state-of-the-art\nmachine learning algorithms using contextually-constrained text corpora and\nshow that this greatly improves predictions of similarity judgments and feature\nratings. By improving the correspondence between representations derived using\nembeddings generated by machine learning methods and empirical measurements of\nhuman judgments, the approach we describe helps advance the use of large-scale\ntext corpora to understand the structure of human semantic representations.\n