Abstract
In recommender systems, the Internet has evolved over the years for recommending items such as music, movies, books and videos for users to boost the popularity or sales for a single item. One of the significant challenges is the mashup developers spend much time to refine their searches to find suitable APIs (application programming interface). A framework is needed for incorporating the matrix factorisation (MF) recommendation that recommends APIs for a mashup application. In this project, we intend to build the recommender systems prototype by implementing machine learning to learn from the previous data extracted from the API description list. The contribution involves many processes of data collection, preprocessing, document vectorisation, implicit learning and recommendation. It should achieve the likelihood for a mashup application to invoke APIs provided by the data from ProgrammableWeb.
There are many MF algorithms available for use in recommendation systems. However, we find that many algorithms can suffer from data sparsity and cold-start issues. The recommendation approaches and filtering methods introduced in the literature review may provide ideas for conducting this investigation. We have employed evaluation metrics to investigate if the data fits well with the testing set. Finally, we highlight the limitations and possible future improvements to this study.
