An Analysis of the Multimodal Representation and Linguistic Mechanisms of the Principle of (Im) politeness Reciprocity in Social Media Interactions

Abstract

Impoliteness is a negative attitude towards a specific behavior in a specific context, and an impolite strategy is a reflection of the type of impolite output. In this paper, we propose a model for impoliteness detection in social media interaction based on sentiment-dependency graph convolutional neural network with modality fusion from the perspective of pragmatics. The model enhances the emotional and syntactic information of text modality through emotion graph and syntactic dependency graph, uses graph convolutional neural network to obtain text information with rich emotional semantics, and then fuses multimodal features by modal fusion, and filters the redundant information by using the self-attention mechanism to detect impolite reciprocity based on the fused information. Based on the real corpus of Chinese and English blog discourse collected in real time, we study the similarities and differences of impolite language in online discourse between the two languages. The experimental results show that the accuracy of the model reaches 84.72, which is 0.64 percentage points higher compared to the better of the comparison models. Under similar online communication contexts, the formation of overall discourse features and local distinctive features of the two corpora stems from the combined effects of dynamic and diverse online communication contexts and communication resources as well as different linguistic and cultural contexts. The study provides valuable new discoveries for the study of social media online discourse features and enlightens people's comprehensive understanding of online language and functions.