For sentiment analysis, we address the problem of supervised-learning being domain-dependent. Additionally, we try to solve the limitations faced during unsupervised-learning where the bag-of-words (lexicon) are not reliable enough, as they might not cover the broad spectrum of words that represent sentiments. We try to overcome these limitations using a novel approach where we combine multiple lexicons and filter out the corpus using the lexicons, before forwarding it to the classifiers. As discussed in the results section, this approach led to the overall improvement in the accuracy of the classifiers.
Paper
Full text
Feature-Set for Sentiment analysis
Semantic Scholar · Computer Science · 2019
Abstract
For sentiment analysis, we address the problem of supervised-learning being domain-dependent. Additionally, we try to solve the limitations faced during unsupervised-learning where the bag-of-words (lexicon) are not reliable enough, as they might not cover the broad spectrum of words that represent sentiments. We try to overcome these limitations using a novel approach where we combine multiple lexicons and filter out the corpus using the lexicons, before forwarding it to the classifiers. As discussed in the results section, this approach led to the overall improvement in the accuracy of the classifiers.