Abstract The topic addressed here is the role of emotions in music creation. The fact that musical fragments can be evaluated on a number of emotional scales allows one to create a semantic map of fragments and to use this map as a guidance in music creation. The work presents results of the study of a virtual composer assistant based on the semantic map and on the graph of similarity (computed using LibROSA) of selected 158 musical fragments (Apple Loops). The assistant was built based on the eBICA cognitive architecture framework (Samsonovich, 2013, 2018). Participants were 16 NRNU MEPhI Undergraduate Students. Fragments were combined into compositions by participants guided by the assistant, and then evaluated on a number of scales by other participants. Four experimental conditions were compared: with and without guidance by the map, with and without guidance by the similarity graph. Results indicate that the usage of similarity data by the assistant significantly improves the quality of resulting compositions. At the same time, the effect of usage of the semantic map was found significant only in the absence of hints based on the similarity data. Nevertheless, obtained results support the idea of the semantic map significance in composer assistant’s function. One of the next goals in this study will be to generate music automatically, following the dynamics of actual or expected human emotional states, e.g., during a videogame. In general, the advantage of this new technology is in its generality: it will have broad applications in many areas of artificial intelligence.