A novel recommendation system for next feature in software

Abstract

Software that needs to fulfill many tasks requires a large number of components. Users of these software need a lot of time to find the desired functionality or follow a particular workflow. Recommendation systems can optimize a user’s working time by recommending the next features he/she needs. Given that, we evaluate the use of three algorithms (Markov Chain, IndRNN, and LSTM) commonly applied in sequence recommendation/classification in a dataset that reflects the use of the accounting software from Fortes Tecnologia. We analyze the results under two aspects: accuracy for top-5 recommendations and training time. The results show that the IndRNN achieved the highest accuracy, while the Markov Chain reached the lowest training time.