The "Cyanide Coffee" case has attracted public attention because it involved the death of Mirna Salihin after consuming coffee containing cyanide. This study conducted a sentiment analysis public comments related to this case on social media Youtube using the LSTM and CNN algorithms with the Word2Vec extraction feature. Dataset used con-sisted 4,361 comments in Indonesian classified into two categories: positive and negative. The results showed that the LSTM model with Word2Vec produced an accuracy 81,40%, a precision 81,73%, a recall 90,44%, and an F1-score 85,86%, outperforming CNN model which achieved an accuracy 79,15% for the positive and negative categories. These findings prove the superiority of LSTM in handling sequential data contexts compared to CNN, especially in text-based sentiment analysis tasks. This study makes a significant contribution to under-standing public opinion on legal cases through a deep learning ap-proach, as well as demonstrating the effectiveness of the Word2Vec feature in improving the performance of sentiment analysis models.