Feature Space Augmentation for Negation Handling on Sentiment Analysis

Abstract

One crucial issue affecting the performance of sentiment analysis tasks is negation. Handling negation involves determining the negation scope and negation cue. Feature space augmentation is one approach used to address negation. Feature space augmentation has been carried out by some previous researchers using a negation flag with the rule that the negation scope includes all words from the explicit negation cue to the punctuation mark. This study aimed to analyze the classifier's performance when negation handling was applied by adding a new rule for the negation scope. The new rule for determining the negation scope no longer took all words from the negation cue to the punctuation mark, but only considered or ignored words with certain POS tags. The results of this study showed that using the new rule for negation scope contributed to improving the performance of the classifier in sentiment analysis tasks. The proposed approach for negation handling was better than the previous approach in terms of accuracy, precision, recall, and f1-score.

Negation handling for sentiment analysis task: approaches and performance analysis

“…However, in some cases, this approach is no better than the FSW and fixed window length approach. Another study [24] that used this approach obtained the same results as previous research, namely that the next non-adverb was better when compared to RoS.…”

“…The negation cue in a sentence can be detected using a predefined lexicon. In the rule-based approach, some rules assume that the meaning of each word in a sentence is reversed (rest of sentence approach) [19]- [24], while other rules assume that the negation signal is in the following words (first sentiment-carrying word approach) [18], [21], [22]. Rule-based approaches can also include syntactic information to determine the subject or object in a sentence [25]- [28].…”

“…Rest of sentence (RoS): assumes that all words in a sentence containing a negation cue are within the scope of negation. This approach has been used by several previous researchers [18]- [24]. In research [21], [22], [24], this approach has lower accuracy performance compared to other rule-based approaches in the sentiment analysis task.…”

“…This approach has been used by several previous researchers [18]- [24]. In research [21], [22], [24], this approach has lower accuracy performance compared to other rule-based approaches in the sentiment analysis task. However, this approach is still better compared to not using negation handling at all for the sentiment analysis task [19], [20], [29].…”

“…Other research using NT, carried out by [43], [44], obtained increased classifier accuracy results after using NT. Other research [24] also uses this approach for negation handling, but for the negation scope detection approach, it uses the NNA approach. Experimental results show that the NT+NNA approach provides better results than the NT+RoS approach.…”