<![CDATA[The uncertainty of health news content, which is spread on social media, raises the need for validation of the truth. One validation approach is to consider the opinion or attitudes of most people, which is called a stance on a topic, whether they support, oppose, or being neutral. This paper proposes a stance analysis model to classify the relationship between sentences so that it can recognize the correlation of the opinion of the writer in the headline of the problem claim. The proposed model uses several Long Short-Term Memory (LSTM), which represent the interrelationship of news for analysis of the relationship between a claim with other news. The formation of word representation vectors is carried out in conjunction with LSTM-based stance classification training. Sentence embedding is done to get the vector representation of sentences with LSTM. Each word in a sentence occupies one time-step in LSTM and the output of the last word is taken as a sentence representation. Based on the results of trials with the Indonesian health-related dataset that was built for this study, the proposed stance classification model was able to achieve an average F1-score value of 71%, with the supporting value 69%, opposing as much as 70%, and neutral 74%.]]>
Copyrights © 2020
