In this paper we introduce an algorithm that is capable of recognizing emotions of userfs statements in order to achieve more effective and smoother human-machine conversation. Many studies of the emotion recognition have been actively conducted in order to quantify affect, but it is rather difficult to recognize it from more complicated sentences, often having double negatives. We describe our enhancements of emotion recognizer by combining emotive expressions lexicon, web-mining techniques, processing compound sentences, and adverb weighting for emotiveness degree modification. The effectiveness of the proposed algorithm for recognizing more complicated sentences was confirmed through evaluation experiments which results are also introduced.