Personalized dialogue generation based on social game data collection and two-stage response generation model

The development of artificial intelligence has promoted the widespread application of human-machine dialogue systems, and users urgently need personalized and emotionally resonant interactions. However, the scarcity of high-quality data and information hinders the generation of role consistent dialogues. To this end, the study first designed a social gaming environment to induce natural user interaction and constructed a Homeland sentence level fine-grained annotation dataset containing 1911 deep dialogue scenes. Then, a two-stage response generation (TSRG) model was proposed, including a selective person predictor (SPP) based on bidirectional encoder representation (BERT) and a multi-layer perceptron for filtering key attributes. The experimental results show that the Homeland dataset has an average dialogue turn that is 5.4 times that of the existing best dataset, and the dynamic distribution of attributes such as identity and personality conforms to interpersonal patterns. The TSRG model achieves a consistency score of 4.40 for generated responses, a 25% improvement over the best baseline. The average accuracy on the test set is 86.8%, with an average absolute error of only 0.0023. After edge computing optimization, the system’s login response time is 593 milliseconds, and the time to end a dialogue is 452 milliseconds. The above results validate the value of the Homeland dataset in alleviating data scarcity issues (with an average of 24.24 dialogue rounds), and demonstrate the superiority of the TSRG model in generating consistency and prediction accuracy. Clinical trial: Not applicable.