Emotional well-being plays a vital role in mental health care, and accurately detecting negative emotions in patient-doctor text conversations can significantly enhance the quality of care provided. Locating such emotions in the text has proven to be a difficult endeavour, mainly when working with a large text data series. This has sparked the development of innovative strategies to treat mental illness and automated technologies to assist mental health practitioners. Because of all display comparable verbal patterns and emotive aberrations, it could be challenging to distinguish between various mental diseases. Using Bidirectional Long Short-Term Memory (BiLSTM) [1] and Multi-Head Attention (MHA) benchmarks, we present a novel technique in current research for identifying negative emotions in question-answer exchanges between mental health patients and specialists. The study focuses on a curated dataset of patient-doctor text conversations, specifically targeting negative emotions expressed by patients during their interactions. The MHA-BiLSTM representative can determine and categorize negative emotions, as it has been trained to capture the emotional nuance and contextual dependency within conversations.