Emotion and sentiment detection from text have been one of the first text analysis applications. Practical use includes human-computer interaction, media content discovery and applications for monitoring the quality of customer service calls. In this paper we perform a review of established and novel features for text analysis, combine them with the latest deep learning algorithms and evaluate the proposed models for the needs of sentiment detection for monitoring of the customer satisfaction from support calls. The issues we address are robustness to the low ASR recognition rate, the variable length of the text queries, and the case of highly imbalanced data sets. The proposed approaches are shown to significantly outperform the accuracy of the baseline algorithms.