A Visualization of Word and Document Embeddings

This paper introduces an open-source visualization tool designed to enhance the comprehension of word and document embeddings. Word embeddings, which translate words into high-dimensional numerical vectors, and document embeddings, which encapsulate the meaning of entire documents, are fundamental to the recent advancements in Natural Language Processing (NLP), particularly with the rise of large language models (LLMs). Following the broader movement to understand new NLP models, our tool is tailored for individuals in the Humanities and requires no prior technical knowledge, offering an interactive and user-friendly interface to explore complex relationships between words and documents in a high-dimensional space. Implemented using a web interface, the tool supports multiple datasets for document embeddings and utilizes pre-trained models like GloVe for word embeddings and Sentence-Transformers for document embeddings. User feedback indicates that the tool is effective in improving the understanding of embeddings. Future work includes enhancing the interface, incorporating more embedding models, and translating the interface into additional languages. This tool represents a step forward in making advanced NLP concepts accessible to a wider audience.