摘要 |
PROBLEM TO BE SOLVED: To provide a musical composition retrieval system which can develop a hierarchical neural network which conforms to a user's taste, and also provide a musical composition retrieval method thereof. SOLUTION: A neural network learning part 24 reads out the coupling weight value of each neuron from an impression degree conversion part 14, makes the coupling weight value of each neuron thus read an initial value, performs relearning of a hierarchical neural network by using relearning data stored in a correction data storage part 23, that is, makes the hierarchical neural network relearn the coupling weight value of each neuron, and updates the coupling weight value of each neuron in an impression degree data conversion part 14 to the coupling weight value of each neuron thus relearned. If the relearning data used in relearning are used again in the next relearning, the number of the relearning data to be used in relearning of the hierarchical neural network is increased, and consequently high-precision relearning is performed. COPYRIGHT: (C)2006,JPO&NCIPI
|