Research On Dialogue Ethics Discrimination Method Based On Deep Learning

With the development of artificial intelligence,the corresponding ethical issues and potential risks have also attracted increasing attention.Among them,dialogue,as one of people’s daily communication methods,the ethical issues brought by its combination with artificial intelligence are worth considering.In order to better achieve this,the primary task is to address the issue of ethical discrimination in dialogue.The ethical discrimination of dialogue focuses on the ethical attributes of each round of discourse in a group of dialogues,which are closely related to the speaker and the contextual relationship of the discourse in the dialogue.Based on this,this thesis studies the ethical discrimination technology of dialogue based on deep learning,and studies and implements the ethical discrimination of Chinese multiple rounds of dialogue from the perspectives of changes in ethical attributes of discourse,as well as the combination of local and global contextual information in dialogue.The work of this thesis can be summarized as follows:Firstly,due to the fact that the Chinese dialogue dataset is still blank in the field of multi-turn dialogue ethics,in order to promote the development of dialogue ethics discrimination research and enrich dialogue content,this article constructs a Chinese multi-turn dialogue ethics dataset based on various sources such as M~3ED,CPED,and film subtitles,and names it Ethic Dialog.In addition,the dataset was applied to the discrimination task of achieving dialogue ethics.Through a series of experiments and combining the results of different datasets and models,the characteristics of the Ethic Dialog dataset were further analyzed,reflecting its role in the field.Secondly,this paper proposes a dialogue ethics discriminative model based on multi-task learning,which realizes the learning and understanding of discourse from the two perspectives of label deviation and semantics.The ethical deviation detection task of dialogue was used as an auxiliary task,and the ethical discrimination task of dialogue was used as the main task to achieve the final model.Finally,through analysis and experimental comparison,the results showed a certain improvement compared to the benchmark,verifying the effectiveness of the model.Thirdly,a dialogue ethics discriminative model based on dialogue unwinding is proposed.On the one hand,by using dialogue to unwind and obtain relevant information of the local context of the discourse,on the other hand,by adding a global graph and sequential context encoder,we focus on the global context information of the conversation.Finally,we analyzed the effectiveness of the model and achieved ethical discrimination through comparison with other models and ablation experiments.