An Emotion Analysis Model based on Fine-grained Emoji Attention Mechanism for Multi-modal We-Media

EasyChair Preprint 4397

Abstract

As an isolated model, emoji plays a crucial role in emotion expression on We-Media platforms, which has always been the hot topic of NLP researches in the new environment. This paper analyses a relatively new method to such researches, namely, Emoji Attention Mechanism and analyses two of its main disadvantages: 1) it catches the literary meaning of the emojis’ names only and discards the information illustrated from the emoji images, disturbing the final results of emotion classification; 2) it ignores the position of emojis in the text and it does not fit for the text emotion analysis that contains emojis in various clauses. Aimed to reduce the effect of the disadvantages mentioned above, this paper proposes the Semantic Vector Extraction based on Image(SVEI); taking the position of emojis in the text as a factor into consideration, this paper proposes an Emotion Analysis Model based on Fine-grained Emoji Attention Mechanism for Multimodal We-media, which enables the emotion of emojis to take effect in clauses more fine-grained. This paper adopts micro blog emotion analysis text published by NLPCC2014 as the data collection and designs the correspondent contrast experiments. This model raises the average F value and promotes the emotion identification accuracy by 3.18%.

Keyphrases: Emoji Attention Mechanism, Emotion Analysis, Semantic Vector Extraction

@booklet{EasyChair:4397,
  author    = {Chunxiao Fan and Siteng Chang and Yuexin Wu and Zheng Chen},
  title     = {An Emotion Analysis Model based on Fine-grained Emoji Attention Mechanism for Multi-modal We-Media},
  howpublished = {EasyChair Preprint 4397},
  year      = {EasyChair, 2020}}