Research On User Perspective Discovery Method Based On Deep Learning In Cloud Service Community

The rapid development of cloud computing has promoted the integration of the industry resources coordination.More and more individuals and organizations move their local application resources in the form of Web services to the cloud computing resource pools(cloud service communities)for users to choose and invoke.Web service recommendation in the cloud service community based on SaaS platform presents new challenges for service computing due to ecosystem complexity.In recent years,more and more people are interested in the emotional tendency of consumer reviews to evalute product ratings and public preferences.Therefore,the research and technology of this kind of analysis become more and more common and mature.However,at present,few recommendation methods that take into account users' dynamic preferences and dynamic service quality,and the research on Chinese API service comments is less concerned.However,few people pay attention to the deep learning research on emotional analysis of Chinese API service usage evaluation.In order to better mine the effective information of user opinion text in cloud service community,this thesis proposes a method based on deep learning to analyze the user's opinion data.First,manually annotate and preprocess the original data in the cloud service community,and convert the text into the input model by using Gensim and Word2 Vec.Then,LSTM,BiLSTM and GRU deep learning models are designed and implemented to realize the task of emotion classification,and the experimental results are compared with the emotional tendency model established by k-nearest neighbor,SVM and naive bayes machine learning algorithms.The results prove the effectiveness of the deep learning method.Finally,the performance of the three established deep learning models is evaluated,and the optimal BiLSTM model is selected as the model used by the cloud service community in the future.The parameter comparison experiment of BiLSTM model is carried out,and the optimal tuning parameters of the model are obtained.Finally,the model achieves an accuracy of 89.68% and has better classification performance.