Research On Hot Spot Discovery And Sentiment Analysis Based On Social Media

With the deepening development of computer and network technology,the Internet has entered the era of social media.Users can publish objective or subjective opinions such as news,social realities,consumer products,etc.and they can also publish real-time information and personal life experiences on social media.In-depth mining of potentially useful information in social media has important applications in the discovery of hot events,public opinion monitoring,product reputation feedback and so on.This article mainly deals with hot spot discovery and sentiment classification of hot spot related texts in social media.In the first two chapters of this paper,the related and basic technology about this problem are introduced in detail.Then for the shortcomings of existing research,in the third to the fifth chapters,we put solutions about the hot spot discovery task,the message-level sentiment classification task and target-level sentiment classification task:(1)For hot spot discovery tasks,an online hot spot discovery algorithm based on fusion features and absolute distance is proposed.In the representation of hot spot related texts,this paper uses ruled methods and Kleinberg algorithms to find and select hot-topic words,and then uses the fusion features to represent the texts.Then this paper proposes an online hot spot discovery algorithm based on absolute distance,which can discover the hot spot in social media timely and effectively.(2)For message-level textual sentiment classification task,this paper proposes a dual model based on reversed algorithm and improves the accuracy of message-level textual sentiment analysis,aiming at the problem of sentiment shift in textual sentiment expression.Firstly,a sentiment reversed algorithm of sample is proposed to construct sentiment reversed samples for the original samples.Then an independent dual model and a joint dual model are proposed to model the original and reversed samples.Finally,by using the integrated algorithm,we consider the independent dual model and joint dual model to get better results.The dual model proposed in this paper has achieved significant improvement on the related standard datasets.(3)For target-level sentiment classification task,this paper proposes a target-independent representation neural network with bilateral attention model.For the phenomenon that target often contains multiple words,the independent representation module is used to learn the new representation of target.In order to better obtain the semantic and sentiment expression of the context of target,this paper proposes a mechanism of bilateral attention,which makes the sentiment indicative words play a more important role in the representation.The model proposed in this paper has achieved the best results on the standard datasets.