Research On Multi-Label Classification Of Web Services Based On Deep Learning

As a modular,self-adaptive and service-oriented architecture technology,Web services play an important role in many fields such as financial business and software development.The emergence of a large number of Web services is the best proof of their vigorous development.How to automatically classify a large number of services and recommend the most suitable series of tags for services has become a hot issue in the field of service discovery research.However,the existing service classification methods consider a small number of categories,ignoring the long-tail characteristics and multi-label characteristics of real-world service data.Therefore,driven by the diversity of service classification and aiming at improving the performance of service multi-label classification,this thesis conducts research on Web service multi-label classification based on the deep learning method.First,for the long-tail problem in multi-label web service data,this thesis designs a balanced method combining resampling and reweighting.The method enriches the tail category information based on multi-granularity text enhancement technology,and designs a reweighted loss function according to the cost-sensitive idea to promote the balanced learning of multiple categories.On this basis,this thesis constructs two models for multi-label classification of long-tail web services.Facing the problem of insufficient feature extraction by a single neural network,this thesis proposes a Web service multi-label classification model WSMLCM based on a hybrid neural network,which fully captures the contextual and local features in the service text.In order to further utilize the service label and text information,this thesis proposes an improved model WSMLCM-LE based on GCN and Caps Net.The model trains the classifier through the relationship between the service labels,and comprehensively extracts text features by combining the pre-trained model BERT and the capsule network,which effectively improves the multi-label classification performance of the model.Experimental results on real-world datasets show that the balancing method proposed in this thesis successfully reduces the negative impact of the long-tailed distribution of service data.At the same time,the two proposed models can effectively extract service text features and realize multi-label classification for Web services.Compared with the existing methods,all the three indicators are significantly improved.