Research On Social Media Text Processing Techniques Based On Fuzzy And Rough Set Theory

With the rise and development of a large number of social media platforms in recent years,as a tool for online user interaction,it is deeply changing people's working,living and communication styles.At the same time,vast amounts of user information have also been generated.And the large number of text files generated by users in social media has become one of the most representative data resources in the field of big data research.Research on these user-generated text data has great value on academic and commercial area.As a kind of information processing technology,Social Media Text Analysis processes,induces,analyzes and reason large-scale of data in many different web applications.Traditional social media text analysis and research models focus on modeling from the data themselves,lose sight of human language logic,which has effected the quality of text analysis.To fix the existing problems in social media analysis,in this dissertation,the advantages of fuzzy sets and rough sets theory on uncertainty analysis are applied for the research and modeling four different applications,such as multi-label emotional intensity analysis,multi-labeled and fine-grained emotion retrieval,knowledge graph and scholarly papers recommendation.Specifically,the main work and innovations of this dissertation are as follows:(1)Most research about emotion analysis of texts nowadays only focuses on sentiment orientation or single emotional label,which did not take multi-labeled sentiment especially emotional intensity into consideration.To solve that problem,this dissertation propose a model based on fuzzy theory and rough set theory to deal with multi-label and emotional intensity analysis which has no former research in the area.We first introduce fuzzy relational equation algorithm into the model to get the intensity range of sentiment words,which are taken into a text,then we use the improved fuzzy rough set model to get the tags and emotional value of the text.The results of the experiment which is based on the Chinese blog data set shows that the multi-label and emotional intensity analysis model we built has advantages in the text levels of sentences,paragraphs and documents,and the model can better predict the multi-label emotional tags and intensities.(2)Most current opinion retrieval models are based on the matching of keywords or sentiment orientation tags,which did not take multi-labeled and fine-granted emotion into consideration.To solve the problem,this dissertation proposes a model based on fuzzy relational equation and fuzzy lattice degree to deal with multi-labeled and fine-granted emotional retrieval.We first introduce fuzzy relational equation algorithm into the model to get the intensity range of sentiment words,then we use fuzzy lattice degree to get the emotional similarity distance of texts,and we can get the retrieval results based on it.From the experimental results on the Chinese blog dataset,we can see that the proposed model has obvious advantages in the accuracy and practicality.(3)Nowadays,in order to achieve better classification results,many translation improvement models based on TransE are frequently introduced into other complex mathematical models,resulting in higher algorithm complexity,more training sets and training time.To solve the problem,the method of fuzzy relation matrix algorithm is proposed in this dissertation to improve the existing model of knowledge-based knowledge map.We combine the fuzzy relation matrix algorithm with deep learning method to build a TransF model based on fuzzy set theory.Experimental results which are based on the dataset of WordNet and Freebase show that the fuzzy theory-based knowledge graph model constructed in this paper not only simplifies the process of reducing the required parameters,training process and shortens the training time,but also shows more advantages when facing smaller training sets.(4)Nowadays almost all the academic papers recommendation systems are only based on keywords matching,which have obvious shortage under some situations,such as the same concept with different names or something like that.To solve the problem,this dissertation proposes an academic papers recommendation method based on rough-fuzzy set theory.First,we use the TF-IDF algorithm to find out the key words of the papers,then use WordNet to calculate similarity distance of the key words between the query and papers in the recommended database.Finally,we use the rough fuzzy set model to simulate the similarity of these keywords,then calculate the similarity of the two texts and finally recommend the papers to the users based on the similarity ranking.The experimental results on the UCI dataset show that the proposed model has better performance in accuracy,time performance and comprehensive practicality.In summary,this dissertation is based on the advantages of fuzzy set and rough set theory on uncertainty data and human language logic,to research methods and techniques for analyzing and processing social media texts,which achieve research results including multi-label emotional intensity analysis,multi-labeled and fine-grained emotion retrieval,knowledge graph and scholarly papers recommendation.The experimental results show that fuzzy set theory and rough set theory can play a great role in the research of social media analysis.The research achievements lay theory foundations and provide technique supports for constructing and implementing social media text analysis.