Open AccessFeature extraction from app reviews in google play store by considering infrequent feature and app description
Author(s) -
Qonita Luthfia Sutino,
Daniel Siahaan
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1230/1/012007
Subject(s) - wordnet , computer science , cosine similarity , information retrieval , feature (linguistics) , app store , similarity (geometry) , sentence , feature extraction , synonym (taxonomy) , focus (optics) , semantic feature , matching (statistics) , semantic similarity , artificial intelligence , data mining , pattern recognition (psychology) , world wide web , linguistics , philosophy , botany , physics , statistics , mathematics , biology , optics , image (mathematics) , genus
Google Play Store is one of the platforms used for distributing various kinds of mobile app from the developer to the users. Through this platform, users are allowed to give their comments about the app. These user reviews could be used to extract potential app’s requirements. They are important information for developer to further develop the app. There have been some previous researches about extracting mobile app features which are frequently mentioned in user reviews. There are less researches that focus on extracting infrequent features. Nevertheless, extracting infrequent features is also important. It is because there is a possibility that important needs contained in the review which are not extracted as frequent features. One of the challenges in infrequent feature extraction was the irrelevant features contained in extracted features. To overcome the problem, this study aims to extract app frequent feature in reviews by finding collocation and infrequent feature in reviews based on dependency as extraction rules. Afterward, it compares the similarity of all the extracted features from review with features from app description. The implied technique of similarity measure includes similarity of 1) single-term by matching each term of feature, 2) synonym referring to WordNet synsets, and 3) sentence based on calculation of lexical semantic vector and cosine similarity. The implementation result is evaluated using precision and recall calculations. The result shows that features extracted by proposed method are more relevant than previous method.