Visualizing Spatial Semantics of Dimensionally Reduced Text Embeddings

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liu, Wei, North, Chris, Faust, Rebecca
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910592740098048
author Liu, Wei
North, Chris
Faust, Rebecca
author_facet Liu, Wei
North, Chris
Faust, Rebecca
contents Dimension reduction (DR) can transform high-dimensional text embeddings into a 2D visual projection facilitating the exploration of document similarities. However, the projection often lacks connection to the text semantics, due to the opaque nature of text embeddings and non-linear dimension reductions. To address these problems, we propose a gradient-based method for visualizing the spatial semantics of dimensionally reduced text embeddings. This method employs gradients to assess the sensitivity of the projected documents with respect to the underlying words. The method can be applied to existing DR algorithms and text embedding models. Using these gradients, we designed a visualization system that incorporates spatial word clouds into the document projection space to illustrate the impactful text features. We further present three usage scenarios that demonstrate the practical applications of our system to facilitate the discovery and interpretation of underlying semantics in text projections.
format Preprint
id arxiv_https___arxiv_org_abs_2409_03949
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Visualizing Spatial Semantics of Dimensionally Reduced Text Embeddings
Liu, Wei
North, Chris
Faust, Rebecca
Human-Computer Interaction
Dimension reduction (DR) can transform high-dimensional text embeddings into a 2D visual projection facilitating the exploration of document similarities. However, the projection often lacks connection to the text semantics, due to the opaque nature of text embeddings and non-linear dimension reductions. To address these problems, we propose a gradient-based method for visualizing the spatial semantics of dimensionally reduced text embeddings. This method employs gradients to assess the sensitivity of the projected documents with respect to the underlying words. The method can be applied to existing DR algorithms and text embedding models. Using these gradients, we designed a visualization system that incorporates spatial word clouds into the document projection space to illustrate the impactful text features. We further present three usage scenarios that demonstrate the practical applications of our system to facilitate the discovery and interpretation of underlying semantics in text projections.
title Visualizing Spatial Semantics of Dimensionally Reduced Text Embeddings
topic Human-Computer Interaction
url https://arxiv.org/abs/2409.03949