I am comparing deep learning and lexicon/rule-based models for sentiment analysis. When I was doing some research into the limitations of lexicon based models, I came across a journal article that mentioned sentiment bias. However, this article did not explain what exactly sentiment bias is.
I am providing the link here: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0202523
What the authors call sentiment bias is the tendency for such systems to give strong results, either positive or negative, as opposed to neutral. Very often sentences or documents are more or less sentiment-neutral, but accumulating the positive or negative weights associated to individual words makes it more likely to result in a non-zero "sentiment value".
Very simple example:
I am comparing deep learning and lexicon/rule-based models for sentiment analysis.
+1 +1 -1 +1 +1
Individually words such as "deep", "learning, "model", "sentiment" can be considered positive; "rule" can be considered negative. As a result your sentence would receive a strong positive score of +3, even though it's actually neutral.
