Chinese dialogue system for cognitively impaired adults based on cognitive stimulation therapy principles

A dialogue system for cognitively impaired adults based on cognitive stimulation therapy (CST) principles, comprising:

• an encoder module configured to classify utterance and execute a progressive mask context training part and a text classification part, wherein the progressive mask context training part is configured to perform a sentiment analysis function for extracting entities and sentences whose values are greater than a threshold and to perform a keyword extraction function for extracting keywords in the utterances, and wherein the text classification part is configured to perform text classification tasks comprising emotion, strategy, and CST classifications;
• a decoder module in electrical communication with the encoder module for response generation according to predicted values derived from attention weights in response to outcomes of the encoder module from the utterance;
• and
• an external knowledge module in electrical communication with the encoder module and the decoder module such that the external knowledge module, the encoder module, and the decoder module interact with each other, so as to form a dialogue framework embedded with CST principles through multi-source knowledge fusion and to guide the dialogue framework to generate responses in line with the CST principles;
• wherein the external knowledge module comprises:
• an extractor comprising a first sub-model with a sentiment analysis function and a keyword extraction function and a second sub-model configured to perform emotion classification tasks;
• and
• a dictionary database comprising an emotion dictionary and a keyword dictionary constructed using the first sub-model and the second sub-model of the extractor, wherein the extractor and the dictionary database are further configured to collectively serve as an external knowledge database for the external knowledge module so as to provide external knowledge sources for the encoder module and the decoder module;
• wherein the encoder module comprises:
• a bidirectional encoder representations from transformers (BERT) module coupled with the dictionary database and configured to mask data from dictionaries of the dictionary database according to predefined thresholds set by the emotion and keyword dictionaries stored in the dictionary database, so as to output hidden states for text classification;
• and
• wherein the decoder module comprises:
• a GPT-2 model configured to generate flexible and emotionally supportive responses, wherein the GPT-2 model is trained by the multi-source knowledge fusion which involved with the external knowledge sources provided by the external knowledge database and the CST principles;
• and
• a Chinese CST dialogue dataset applied to fine-tune the GPT-2 model.