Int J Performability Eng ›› 2020, Vol. 16 ›› Issue (1): 118-129.

• Orginal Article •

Hybrid Recommender System based on Deep Learning Model

Chang Suand  Deling Huangab*()

1. aInformation Center, Chongqing University of Posts and Telecommunications, Chongqing, 400065, China
bSchool of Engineering, Computer and Mathematical Sciences, Auckland University of Technology, Auckland, 1010, New Zealand
• Submitted on  ;  Revised on  ; Accepted on
• Contact: Deling Huang E-mail:huangdl@cqupt.edu.cn
• Supported by:
This research is supported by the National Natural Science Foundation (No. 61772099, 61772098), Artificial Intelligence Technology Innovation Important Subject Projects of Chongqing, and Technology Innovation and Application Development Projects of Chongqing.

Abstract:

Mashups, which can produce enriched results using fast integrating application programming interfaces (API) and data sources, play a pivot role in building web-based and mobile applications. However, the rapidly increasing number of APIs makes it difficult to choose suitable APIs for a mashup, especially when the historical relationships between APIs and mashups are very sparse. Many researchers have sought to develop algorithms to recommend APIs to mashups. Some hybrid approaches integrate model-based collaborative filtering and auxiliary information and have achieved good prediction performance. However, there are few models that analyze the contextual information of services' descriptions. In addition, many approaches explore the context information of APIs, but few of them consider the context information of mashups. This paper presents a new model named CDHMF for recommender systems, which uses the Word2Vec technique to integrate the contextual information into a probabilistic factorization matrix (PMF). We analyze descriptions not only for APIs but also for mashups. Our experiments are performed with datasets from ProgrammableWeb. The results show that CDHMF significantly outperforms some state-of-the-art recommender systems in mashup service applications.