Parallel Methods for Evidence and Trust based Selection and Recommendation of Software Apps from Online Marketplaces

Abstract

With the popularity of various online software marketplaces, third-party vendors are creating many instances of software applications ('apps') for mobile and desktop devices targeting the same set of requirements. This abundance makes the task of selecting and recommending (S&R) apps, with a high degree of assurance, for a specific scenario a significant challenge. The S&R process is a precursor for composing any trusted system made out of such individually selected apps. In addition to feature-based information, about these apps, these marketplaces contain large volumes of user reviews. These reviews contain unstructured user sentiments about app features and the onus of using these reviews in the S&R process is put on the user. This approach is ad-hoc, laborious and typically leads to a superficial incorporation of the reviews in the S&R process by the users. However, due to the large volumes of such reviews and associated computing, these two techniques are not able to provide expected results in real-time or near real-time. Therefore, in this paper, we present two parallel versions (i.e., batch processing and stream processing) of these algorithms and empirically validate their performance using publically available datasets from the Amazon and Android marketplaces. The results of our study show that these parallel versions achieve near real-time performance, when measured as the end-to-end response time, while selecting and recommending apps for specific queries.