Text classification is a method for automatically assigning the best possible set of labels to a piece of text. Texts can be a highly rich source of information, but because they are unstructured, it can be challenging and time-consuming to find what we need. Texts are now classified using machine learning as multiple labels in order to solve this problem. One of this text is a toxic comment, which is a comment made during online contact that is unpleasant, insulting, or unfair and that prompts the user to stop using the page they are on and leave. The thesis provides an optionated text classification model based on deep learning techniques for classifying optionated texts and performing suitable measures for managing the negative effects caused on social media sites. The majority of previous research has been focused on single-level issues that only take into account mutually exclusive solutions. So, the major goal of our research is to use deep learning techniques to build a multi-label classification model for optionated Amharic texts. We build a multi-label text classification model for optionated Amharic text using deep learning methods and which are trained using 8k optionated Amharic texts. The model accepts optionated texts as input and classifies them into specified labels or categories based on the label of toxicity based on their content. This research uses a number of text processing tasks: - tokenization, normalization, stop word removal and stemming are the stages that we have applied for our dataset for text preparation. Finally, we have built word2vec and the experimental result of multi-label optionated Amharic texts classification models shows that, CNN, LSTM, Bi-LSTM and hybrid CNN-BiLSTM scores an accuracy of 91%, 85%, 90% and 94% respectively, is achieved by the proposed model. Based on the experimental result we propose the hybrid CNNBiLSTM model which has better classification accuracy, because of its better ability at handling context by combining the CNN and Bi-LSTM models. So, to improve the performance of the classification model for optionated Amharic texts we recommend as future research works the need to prepare standard dataset for experimentation. Key Word: Deep learning method, optionated text, multi-Label classification