Customized dialogue system based on biographical knowledge retrieval for elderly

This work presents a customized dialogue system with a human assisted learning mechanism. The system feedback strategy is adapted according to the enrolled user information for customized requests. The dialogue system prototype is designed based on Finite-state-script model and the proposed POS Reference Column Based Row Based(PRCBRB) sentence similarity algorithm. The aforementioned customized mechanism is fundamentally realized by the following proposed techniques, including, the personal information knowledge representation format and two kinds of knowledge retrieval methods. In the experiments, the Mandarin Conversational Dialogue Corpus (MCDC) corpus is applied for testing and analyzing the efficiency of the PRCBRB algorithm. In the inside-test, the accuracy can achieve to 97.3% accuracy and the computation cost is 5 times faster than the other algorithms. In the outside-test, the general user and the senior users, TCR can be reached to 86.7% and 64%, respectively.