Text-based Recommendation Systems for Software Developers: A Systematic Literature Review

Abstract

Purpose: The aim of this SLR is to look at recommendation systems which receive textual information as an input. By analysing them it is possible to understand how the textual information is preprocessed and which algorithms are then used to generate recommendations. Methods: With the Search Query I first identified 487 papers, from which 65 were removed as duplicates. After the IC and EC application, 28 articles remained as relevant. Results: From these articles' analysis, it was found that the most commonly used pre-processing techniques are tokenization, TF-IDF, and stopwords removal. I also determined that all algorithms for suggestions generation in such systems can be divided into 4 categories: classification, ranking, clustering, and heuristic-based algorithms. In the last step I found that the most frequent output of such systems are API, code, and workers suggestions. Conclusion: With this work, I looked at which pre-processing techniques are used in the text-based recommender systems for software developers and which are the most common. I have also looked at the classification of algorithms for such recommendation systems. Finally, I considered what kind of objects are recommended by these text-based recommendation systems.