Nowadays, social media plays an important role in receiving all kinds of information, including customer reviews or feedback for products or services. Data generated from social media may give significant input to a company, including, customer satisfaction. This study is conducted to assist in selecting a suitable sentiment analysis model focusing on Malay language and social media data type in the Food and Beverages (F&B) industry. Data were retrieved from online review platforms, using Python as the web-scraping technique. A standard text-processing approach was adapted to clean the textual data for succeeding analysis. Eight types of the Transformer model, namely BERT, Tiny-BERT, ALBERT, Tiny-ALBERT, XLNet, ALXLNet, Fast former, and Tiny-Fast former that have been pre-trained in Malaya Documentation were used and the sentiment class is grouped into three, namely positive, neutral and negative. Based on standard classification performance metrics, XLNet outperforms other models with 75.96% accuracy and 78.91% AUC value. This shows that, although the Malaya Documentation claimed that the fast former model has the highest accuracy for the general media social dataset. Ultimately, XLNet is presented as the suitable model for the F&B dataset.