面向bilibili短视频评论数据的情感分析,旨在挖掘视频观看者对短视频的看法,使视频作者也可以快速得到自己想要的评价,进而对后续作品做出改进.针对短视频评论更新快、词汇新颖、评论过长、一词多义等因素造成的短视频评论情感分析准确率低的问题,文章构建了bilibili短视频评论数据集,并提出了ELMO(Embedding From Language Model)用以构建动态词向量解决一词多义及新词的问题,通过构建TextCNN和Reformer双通道神经网络结构来提取局部、全局特征.由于Reformer采用了局部敏感哈希的特殊注意力机制,更能联系全局特征,之后将两者得到的结果拼接送入分类器得出情感分析的结果,并将得出的结果与多个深度学习模型进行对比.
Sentiment Analysis of bilibili Comment Based on ELMO-TextCNN-Reformer
Sentiment analysis of bilibili short video comment data aims to explore viewers'opinions on short videos,enabling video creators to quickly obtain the feedback they desire and improve their subsequent works.In response to the low accuracy of sentiment analysis in short video comments caused by factors such as fast updates,novel vocabulary,long comments,and polysemy,this paper constructs the bilibili short video comment dataset and proposes ELMO(Embedding From Language Model)to construct dynamic word vectors to solve the problem of polysemy and new words.Local and global features are extracted by constructing a TextCNN and Transformer dual-channel neural network structure.Due to the use of a special attention mechanism with locally sensitive hashing in the Transformer,it can better connect global features.Then,the results obtained from both are concatenated and fed into the classifier to obtain the results of sentiment analysis,and the results obtained are compared with multiple Deep Learning models.
国家科技期刊平台
NSTL
万方数据
