Web APIs Recommendation Based on Topic Modelling and Clustering
Nowadays, Recommender Systems are widely used in various web portals, while service discovery is still a great challenge for better integrating appropriate services into business scenarios. Gaining insight of the development of recommender systems is helpful for tackling the issues. This thesis proposes a recommender system framework to achieve Web APIs recommendation based on the collected data of Web API directory: P rogrammableW eb.com. We intend to build a comprehensive Recommender system by combining the method of collaborative filtering recommendation with topic modeling in natural language processing. Specifically, we find that the collaborative filtering algorithms as the mainstream recommendation method is affected by the cold start problem, and the different kinds of recommendation algorithms lack systematic comparison in existing works. Therefore, we integrated topic vector and then document clustering to solved the cold start problem, and then implement the representative collaborative filtering algorithms. The various evaluation methods are used to evaluate the ranking quality and diversity of Web APIs recommendations. We discover that the topic vector extraction of LDA algorithm with the K-means al- gorithm combines the Agnes algorithm is able to achieve satisfying document clustering results, and the collaborative filtering algorithm has good recommendation performance by considering invoke relationship, content similarity, and latent semantic mining.