Page 70 - The-5th-MCAIT2021-eProceeding

which containsfinalized positive and negative adjective sentiment bearable words.

Fig. 1. Sentiment Lexicon Generation Model Based on Dictionary Approach

4. Results and Discussion

The result of the experiment is shown in Table 1.
Table 1. Total adjectives generated after five iterations.
Items Total Number
Number of POS 25541
Number of NEG 36064
Total Words
Generated 61605

The first iteration started with setting a pair of strong seed sets. The initial seed set was choose based on
our benchmark study, which is “baik” for the positive seed set and “buruk” for the negative seed set. After the
experiment was run for five iterations, the final iteration produceda total of 61605, with 36064 words that exist
in negative polarity, meanwhile 25541 words with positive polarity. The result also shows that the final lexicon
produced not only limited to a single word but also multiple terms such as "bersikapsabar”. It also can be seen
that some of the terms were generated more than once. For example, in the positive list, the word 'suci' which
means clean appears 45 times, and 'bangpak' which means cruel appear 18 times. Even though the same word
appears multiple times, but it was rooted in different synsets.This result also shows that the synonymy
generation process for positive sentiment words produced several same lemmas or words that duplicate with
antonymy words generation in negative lexicon list. This also happened in synonymy generation for negative
lexicon list, which also generated a few same lemmas in antonymy generation in the positive list.It can be seen
here that the first iteration collects the related words in the synset mostly, but the number of sentiment words
increases slowly in the remaining iterations. Due to this reason, the numbers of iterations were run at five (5)
iterations only as being done in previous research (Alexander & Omar, 2017).

5. Conclusion and Future Works

In this paper, it was demonstrated that the sentiment lexicon generation algorithm for the Malay
Languagebased ondictionary-based approach could lead to an expanded sentiment lexiconand promising
results. However, this proposed work only contains adjectives, even though it is known that other Part-of-speech
(POS) also carry the sentiment. Moreover, this work is only suitable to cater sentiment analysis task of formal

E- Proceedings of The 5th International Multi-Conference on Artificial Intelligence Technology (MCAIT 2021) [57]
Artificial Intelligence in the 4th Industrial Revolution