3-7 基于图表征学习的跨领域情感分析方法.pdf

1、Graph Adaptive Semantic Transfer for Cross-domain Sentiment Classification（基于图表征学习的跨域情感分析方法）Kai ZhangUniversity of Science and Technology of China2022年6月25日 Saturdayhttp:/ WorkThe GAST ModelExperiment5Conclusion|3Sentiment Classification Sentiment LabelsSentiment classification aims to mine the user

2、s emotional tendency from text or pictures.l Opinion miningl Sentiment AnalysisBackground TextPicture?AttitudeEmotion TendencySpecial task of text classificationLabel:l favorable or unfavorablel Positive,negative,neutral4Transfer LearningAdvantagesRefers to the use of similarities between tasks to a

3、pply the knowledge learned in the old field to a new field.Background l Unlabeled datal Cold startl Model versatilityl Weak computing SourceTargetSource:with labeled data Target:less or no labeled data5Cross-domain Sentiment Classification(CDSC)ValuesRefers to the use the similarity knowledge learne

4、d in the source domain to the target domain.l Source Targetl Single InvariantThe use of common features between tasks to apply sentiment semantics learned from a old field to a new field.Backgroundl Unsupervised Sentiment Classificationl Focus on short-text mining and transfer learning methods61.Mak

5、ing full use of domain shared features is pivotalfor cross-domain sentiment classification.what information do we need to focus on in the whole sentence?How to pay attention?2.How to mine and utilize the information that is critical for cross-domain classification tasks?BackgroundKey ChallengesFindU

6、seOutline1234BackgroundRelated WorkThe GAST ModelExperiment5Conclusion|8Related WorklUnlabeled problemSample re-weightingSubspace matchingDeep methods9Related WorkPast researches can be categorized as:1.Traditional(Theoretical)Blitzer 2006;Pan et al.2010;Chen et al.2012)Analyzing the shared features