Microblogging sites have millions of people sharing their thoughts daily because of its characteristics, short and simple manner of expressions. From the tweets they send, what sentiments are going on in the public space are expressed. Sentiment analysis usually use an integral part of social listening. Sentiments analysis are often being used to analyze the user or customer opinions regarding brand images or products. This research used the case of existing sentiment analysis platform Telkom UData, which grab tweets from microblogging using certain keywords such as 'indihome'.
For some reasons, not all sentiment generated by this existing machine-based algorithms yields gave satisfied results. Words found in Twitter mostly used were unstructured and informal. While only some grammatical structures that formed the phrases, clauses, or sentences could be recognized by the machine, and many sentiments of sentences were still not well defined. This was mostly due to the uniformity of the informal language used in the social media sentences. This condition also occurred in UData on this preliminary study, where the machine-based provided less then optimal results in analysing the sentiment. Thus human interaction through crowdsourcing method was proposed.
This research offered concepts with human interaction using crowdsourcing where people were involved to analyze sentiments, while at the same time forming the new dataset training for sentiment that does not recognized by the machine.
The results of the experiments found that sarcastic and contradictory sentences were able to be recognized by humans, and to be utilized as new training datasets for further machine learning. From this experiments, that approach was likely to increase the accuracy of the sentiments in UData from neutral to become positive or negative polarized up to 39%.
Simulated trust concept was conducted as well through sociometric to ensure that the crowdsource workers were trusted and capable enough of analyzing the sentiments on social media.