This paper focuses on the review of the emotion identification in texts related to sexual harassment on social media using the LLM approach, especially in the Malay language context. The phenomenon of cyber sexual harassment is increasingly worrying in Malaysia, but traditional approaches such as the use of keywords and conventional machine learning such as Naïve Bayes and SVM have proven to be insufficient in identifying texts in the form of implicit, figurative or sarcastic language used by victims. Furthermore, existing models have not been successful in identifying emotions contained in the texts such as trauma, fear and sadness expressed by victims. This study proposes the use of LLM based on fine-tuned encoders using a local data corpus obtained from the Twitter platform. The propose model will assess eight basic emotions such as happy, sad, trust, fear, anger, surprise, disgust and anticipation. Multi-Task Learning (MTL) approach is used to perform tasks simultaneously with a shared encoder and separate output head. The main objectives of this study are: (1) to improve the ability of LLM in detecting emotion in sexual harassment texts, and (2) to propose a model using the MTL approach to perform simultaneous identification of emotions based on the Plutchik model and to evaluate the accuracy of the model. The results of this study are expected to contribute to the development of more contextual emotion monitoring technology for victims of sexual harassment.