In paper proposed approach to solving the problem of interpretive visual analytics for neural network detection of social media posts sentiments, which differs from analogues by using hybrid architecture of convolutional neural network and bidirectional long short-term memory. The hybrid combination of CNN and BiLSTM architectures helps to highlight local patterns in the text, such as keywords or phrases, which important for the classification of emotional tonality (CNN advantage), as well as to take into account long-term dependencies in the text from both directions (Bidirectional LSTM advantage). This improves the neural network’s understanding of context, provides more accurate predictions, and enables visual analytics of results, which is important for emotional tonality classification. For local interpretive visual analytics of neural network model, the machine learning model of interpreted model-agnostic explanations LIME is used, which provides classification results explanation. As the result of applied research, it was established that developed method using the CNN-BiLSTM hybrid model provides accuracy of sentiments detection 97.4%, compared to accuracy of 86–89% of analogues, and method also has superstructure for interpreting the received solutions. Local visual analysis of results in developed method increases the transparency and interpretability of sentiments tonality analysis model, which is important for practical application in social communications field and promotes trust in neural network approaches.