Font Size: a A A

The Research Of Fine Grained Text Sentiment Analysis Based On Deep Learning

Posted on:2022-04-15Degree:MasterType:Thesis
Country:ChinaCandidate:L W HouFull Text:PDF
GTID:2518306764993159Subject:Automation Technology
Abstract/Summary:PDF Full Text Request
In today's era of big data,massive amounts of information are generated every day.The Internet has become an important way for people to express their opinions and obtain information.Internet users have changed from pure information obtainers in the past to major producers of online content.When users use Weibo,e-commerce and other platforms,they will publish their own opinions and opinions on a certain event or a certain product,and generate a large amount of personal emotional text information.If the potential user sentiment can be analyzed from these data Information,then will have huge scientific research value and practical value.Traditional text sentiment analysis is mainly based on the text and sentence level.It is difficult to deal with single sentences containing multiple emotions and cannot meet the current trend of information diversification.In response to the need for more granular sentiment analysis,fine-grained sentiment analysis based on goals and aspects has gradually become a research hotspot.This technology can analyze text data from multiple dimensions and fully mine the sentiment tendencies of various goals and aspects.This research conducted a detailed analysis of the characteristics of fine-grained emotions in texts,and conducted research from the two aspects of text's semantic feature enhancement representation and information interaction mode,and proposed a fine-grained emotion analysis model based on self-attention position information fusion.The contribution consists of three parts:(1)In order to enhance the fine-grained semantic representation ability of the sentiment analysis model,a self-attention based location information fusion mechanism(Location Fusion,LF)is proposed.First,the aspect word position information is embedded in the Encoder structure of the Transformer framework,and then the position-enhanced attention mechanism is used to enhance the model's text semantic representation ability.(2)In order to improve the information interaction between aspect words and context,combined with the deep memory network,a deep memory network model(LWATT-MN)based on the fusion of self-attention position information is constructed.The contextual semantic enhancement representation based on a specific aspect is used as an external memory unit that interacts with the aspect,and the aspect is embedded as the initial input of the deep memory network to adaptively capture the emotional features related to the aspect in the context.(3)Using the location information fusion mechanism and the deep memory network joint model,feature extraction and information interaction of multi-faceted emotional texts are used to build a fine-grained emotional analysis model.Finally,the data set is compared and analyzed,and the effectiveness of the key modules is verified through ablation experiments.
Keywords/Search Tags:sentiment analysis, deep learning, attention, location embedding
PDF Full Text Request
Related items