Research On Web Service Long-tail Classification Based On Label Embedding And Attention Mechanism

Web services,as an adaptive,self-describing,modular,and well-interoperable application and functionally independent software component in the Internet era,have played a role in applications such as e-commerce,enterprise application integration,and business process management.It continues to influence the promotion of innovation models in the modern service industry.Web services have boomed over the past decade with the advent of Web 2.0 technology.Programmable Web,the largest online registry for Web services,has registered 23,038 services across 488 application categories.However,when a service provider registers a new service,it is often time consuming and difficult to select the most relevant labels from lots of service categories.Besides,it is difficult to select the functional category accurately that meets service needs.Therefore,it's important to classify service automatically by the function description provided by the publisher and recommend the most matching labels for them to select.Service classification has become a hot topic in the field of service computing in recent years.Most of the existing Web service classification methods focus on how to get the characteristic representation of services more accurately.But there are still the following deficiencies:(1)The number of categories considered in service classification is small,and it is difficult to achieve diversified service category recommendation.(2)The service dataset is uneven and has a long-tail distribution,and the accuracy of long-tail services is ignored in service classification.(3)Web service descriptions are usually short in text,and the features are sparse and difficult to extract.In this paper,we aim to increase the diversity of service classification and improve the accuracy of long-tail service classification.In this paper,we innovatively propose a long-tail web service classification model based on deep learning method.The main contributions of this paper are as follows:(1)Long-tail service dataset collection and visualization.To support the variety of service categories and the number of large-scale web services required for the long-tail service classification problem,we use a Python crawler to collect web services from the Programmable Web platform and build a long-tail service dataset containing multiple categories and web services.After analyzed the crawled data,we found that the web service dataset is significantly imbalanced,showing a long tail distribution.Also,the average length of the service description is about 40 words,which means the service descriptions are short texts.Finally,to observe the dataset,we have developed a system that can query and display web services.(2)Construct a long-tail service classification model based on a deep learning method.To solve the long tail service classification problem,we propose a model that combines label embedding and multi-head attention to extract the characteristics of short text services more accurately.We use the Focal Loss function to deal with the imbalanced datasets.To prove the effectiveness of the model for long-tail service classification,we take the top 80 categories as our dataset and set up four sub testing sets,which are named as Overall,Niche-20,Niche-30,and Niche-40.The experimental results show that the method we proposed outperforms state-of-the-art approaches for long-tail service classification in terms of multiple evaluation metrics.