Building a robust sentiment lexicon with (almost) no resource
This work addresses the resource-intensive nature of lexicon creation for sentiment analysis, offering a more accessible approach for multilingual applications.
The authors tackled the problem of creating sentiment polarity lexicons without extensive resources by proposing a method that transfers words using cross-lingual word embeddings instead of machine translation, achieving no degradation in sentiment classification accuracy on tweets across four languages.
Creating sentiment polarity lexicons is labor intensive. Automatically translating them from resourceful languages requires in-domain machine translation systems, which rely on large quantities of bi-texts. In this paper, we propose to replace machine translation by transferring words from the lexicon through word embeddings aligned across languages with a simple linear transform. The approach leads to no degradation, compared to machine translation, when tested on sentiment polarity classification on tweets from four languages.