Hybrid Web API Recommendation Based On Near Neighbor Reinforcement And Multi-Tasking Learning

Mashup is a method of creating high-level feature-rich applications by reusing the Web Application Programming Interface(Web API).As the number of Web APIs available in the network environment continues to increase,manual selection of Web APIs suitable for Mashup creation becomes increasingly infeasible.In order to solve this problem,researchers put forward the concept of Web API recommendation,and proposed a variety of algorithms to match the requirements of Mashup creation.However,the existing methods dominated by collaborative filtering and content analysis are still weak in feature fusion and utilization,and it is difficult to meet the multi-dimensional recommendation requirements.To this end,this paper proposes two deep recommendation models from the perspective of the use of Mashup data and the enhancement of feature data,and the two models are merged from the structural level to achieve better recommendation results.Aiming at the problem of incomplete utilization of model features in Mashuporiented Web API recommendation,a hybrid multi-task Web API recommendation(Hybrid Multi-Task Recommendation,HMTR)model is proposed.HMTR is based on Mashup document description,uses the semantic coding module based on Text CNN to generate Mashup requirements and document feature representation,and combines the feature interaction module to model the interaction between Mashup feature representation and Web API embedding.Introducing Mashup category prediction as an auxiliary task,so that HMTR has the ability of multi-task learning.The model parameters are optimized through the backpropagation algorithm,and the recommendation sequence is output to realize the hybrid Web API recommendation.Aiming at the problem of lack of data features in Web API recommendation,a Near Neighbor Reinforcement Recommendation(NNRR)model is proposed.Based on document description and category tags,NNRR uses the neighbor selection module to filter neighbor mashups,and uses the feature reinforcement module to mine document features and embedded category features,and compound features from the dimension of feature representation to achieve enhanced effects.Based on the enhanced features,NNRR transforms the Web API recommendation problem into a multi-label learning problem,uses the binary cross-entropy loss function to optimize the model,and outputs the recommendation sequence to realize the Web API recommendation.In addition,the Reinforcement Hybrid Multi-Task Recommendation(RHMTR)model is proposed by fusing HMTR and NNRR in the model structure,and the model is adjusted and trained in the same way to achieve more efficient Web API recommendation.Based on the data collection from the real scene,the model in this paper is analyzed quantitatively and qualitatively in a variety of ways.The experimental results show that the method proposed in this paper can accurately recommend Web API,and can achieve the purpose of improving the performance of Web API recommendation through fusion.