A Similarity-Based Software Recommendation Method Reflecting User Requirements

Abstract

Existing software recommendation methods consider only the usage frequencies of software as far as we know. In this study, we propose a software recommendation method reflecting user requirements based on both the Boolean model and vector space model. A function matrix and function vector are made from the functional specification of each software type and stored in the database. First, it creates a requirement vector from a user's functional requirements of the desired software. Second, it makes a list of software with the same functions wanted using the function matrix based on the Boolean model. Third, the cosine similarities are calculated between the requirement vector and function vectors of the software in the list based on the vector space model. Finally, a software recommendation list is generated in descending order of similarity. Based on the experiment results, appropriate software well suited for user requirements can be recommended. This is because we searched for software that satisfies each user's requirements by using the cosine similarity function of information retrieval and recommended it according to the ranking. In the future, performance can be improved by utilizing statistical search techniques.