Despite the great advances in generative dialogue systems, existing dialogue generation models are still unsatisfactory in maintaining persona consistency. In order to make the dialogue generation model generate more ...Despite the great advances in generative dialogue systems, existing dialogue generation models are still unsatisfactory in maintaining persona consistency. In order to make the dialogue generation model generate more persona-consistent responses, this paper proposes a model named BERT-HCM (Personalized Dialogue Generation Model Based on BERT and Hierarchical Copy Mechanism). The model uses an encoder based on BERT initialization to encode persona information and dialogue queries and subsequently uses a Transformer decoder incorporating a hierarchical copy mechanism to dynamically copy the input-side content to guide the model in generating responses. The experimental results show that the proposed model improves on both automatic and human evaluation metrics compared to the baseline model and is able to generate more fluent, relevant and persona-consistent responses.展开更多
Nonverbal symbols are widely used in various social contexts in daily life.In the context of variety shows,nonverbal symbols play a role in constructing celebrity personas and driving plot development.In an era domina...Nonverbal symbols are widely used in various social contexts in daily life.In the context of variety shows,nonverbal symbols play a role in constructing celebrity personas and driving plot development.In an era dominated by short videos,the use of nonverbal symbols by some content creators has become more evident for segmented communication.Due to the inherent authenticity of nonverbal symbols,inconsistencies in their use within variety shows can lead to"persona collapse."This paper analyzes the role of nonverbal symbols in constructing celebrity personas within variety shows through case studies.Furthermore,it explores how the public's trust in information is influenced by various factors amid the widespread dissemination of information on short video platforms.The findings offer insights for the planning and performance of variety shows and the analysis of public opinion on the internet.展开更多
文摘Despite the great advances in generative dialogue systems, existing dialogue generation models are still unsatisfactory in maintaining persona consistency. In order to make the dialogue generation model generate more persona-consistent responses, this paper proposes a model named BERT-HCM (Personalized Dialogue Generation Model Based on BERT and Hierarchical Copy Mechanism). The model uses an encoder based on BERT initialization to encode persona information and dialogue queries and subsequently uses a Transformer decoder incorporating a hierarchical copy mechanism to dynamically copy the input-side content to guide the model in generating responses. The experimental results show that the proposed model improves on both automatic and human evaluation metrics compared to the baseline model and is able to generate more fluent, relevant and persona-consistent responses.
文摘Nonverbal symbols are widely used in various social contexts in daily life.In the context of variety shows,nonverbal symbols play a role in constructing celebrity personas and driving plot development.In an era dominated by short videos,the use of nonverbal symbols by some content creators has become more evident for segmented communication.Due to the inherent authenticity of nonverbal symbols,inconsistencies in their use within variety shows can lead to"persona collapse."This paper analyzes the role of nonverbal symbols in constructing celebrity personas within variety shows through case studies.Furthermore,it explores how the public's trust in information is influenced by various factors amid the widespread dissemination of information on short video platforms.The findings offer insights for the planning and performance of variety shows and the analysis of public opinion on the internet.
