Current browse context:
cs.LG
Change to browse by:
References & Citations
Computer Science > Computation and Language
Title: Towards Label-Agnostic Emotion Embeddings
(Submitted on 1 Dec 2020 (v1), last revised 6 Nov 2021 (this version, v2))
Abstract: Research in emotion analysis is scattered across different label formats (e.g., polarity types, basic emotion categories, and affective dimensions), linguistic levels (word vs. sentence vs. discourse), and, of course, (few well-resourced but much more under-resourced) natural languages and text genres (e.g., product reviews, tweets, news). The resulting heterogeneity makes data and software developed under these conflicting constraints hard to compare and challenging to integrate. To resolve this unsatisfactory state of affairs we here propose a training scheme that learns a shared latent representation of emotion independent from different label formats, natural languages, and even disparate model architectures. Experiments on a wide range of datasets indicate that this approach yields the desired interoperability without penalizing prediction quality. Code and data are archived under DOI 10.5281/zenodo.5466068.
Submission history
From: Sven Buechel [view email][v1] Tue, 1 Dec 2020 00:54:13 GMT (438kb,D)
[v2] Sat, 6 Nov 2021 20:48:40 GMT (615kb,D)
Link back to: arXiv, form interface, contact.