Detecting User Significant Intention via Sentiment-Preference Correlation Analysis for Continuous App Improvement

Analysing app reviews for software engineering: a systematic literature review

“…Clustering is thus widely used as an exploratory analysis technique to infer topics commonly discussed by users (Pagano and Maalej 2013;Guzman and Maalej 2014;Liu et al 2018) and aggregate reviews containing semantically related information (Chen et al 2014;Palomba et al 2017;Zhou et al 2020). Clustering can be used for grouping reviews that request the same feature (Peng et al 2016;Di Sorbo et al 2016), report similar problems (Martin et al 2015;Villarroel et al 2016;Gao et al 2018b;Williams et al 2020), or discuss a similar characteristic of the app (Vu et al 2016;Chen et al 2019;Xiao et al 2020). The generated clusters might help software engineers synthesize information from a group of reviews referring to the same topics rather than examining each review individually (Fu et al 2013;Hadi and Fard 2020).…”

“…For example, Di proposed summarizing thousands of app reviews by an interactive report that suggest to software engineers what maintenance tasks need to be performed (e.g., bug fixing or feature enhancement) with respect to specific topics discussed in reviews (e.g., UI improvements). Other review summarization techniques give developers a quick overview about users' perception specific to core features of their apps (Iacob and Harrison 2013;Guzman and Maalej 2014;Xiao et al 2020), software qualities (Ciurumelea et al 2018), and/or main users' concerns (Iacob et al 2013a;Iacob et al 2016;Ciurumelea et al 2017;Tao et al 2020). With the addition of statistics e.g., the number of reviews discussing each topic or requesting specific changes, such a summary can help developers to prioritize their work by focusing on the most important modifications (Ciurumelea et al 2017).…”

“…A review can be modelled as a words sequence (Johann et al 2017), bag-of-words (BoW) (Maalej and Nabil 2015) or in vector space model (VSM) (Vu et al 2015a) to sereve as input for other mining techniques. In particular, primary studies refers to NLP techniques comparing text similarity (Vu et al 2015b;, pattern matching (Groen et al 2017;Johann et al 2017;Song et al 2020) and collocations finding (Guzman and Maalej 2014;Li et al 2018;Dalpiaz and Parente 2019;Xiao et al 2020).…”

“…A few studies motivated their approaches to capture potential reasons for design decisions (Groen et Validation by Users Evaluating a software system with users usually involves expensive usability testing in a laboratory (Iacob et al 2013a) or acceptance testing performed in a formal manner (IEEE 1990). In the case of mobile apps, software engineers can exploit user feedback to assess user satisfaction (Fu et al 2013 In line with the direction, several approaches have been proposed to mine user opinions (Guzman and Maalej 2014;Gu and Kim 2015;Bakiu and Guzman 2017;Shah et al 2019c;Da ¸browski et al 2020;Xiao et al 2020) and to generate bug reports (Iacob et al 2013a;Maalej and Nabil 2015;Man et al 2016;Ciurumelea et al 2018;Liu et al 2018;Shah et al 2019c). Opinion mining approaches help to discover the most problematic features and to quantify the number of negative opinions.…”

“…This was practiced for evaluating Classification (Li et al 2017), Clustering (Guzman and Maalej 2014;Vu et al 2015a;Palomba et al 2017), Information Extraction (Johann et al 2017;Li et al 2017), Searching and Information Retrieval (Wei et al 2017), and Recommendation (Shams et al 2020). -Seven studies used other software artefacts as an evaluation baseline rather than creating an annotated dataset Man et al 2016;Gao et al 2018b;Uddin et al 2020;Srisopha et al 2020a;Srisopha et al 2020c;Xiao et al 2020). To evaluate Recommendation (e.g., determining priorities for reported issues), the studies compared recommended priorities for issues with priorities for the issues reported in user forums or changelogs; to assess the quality of Clustering, the studies benchmarked the output of their approaches with topics from app changelogs; whereas to evaluate their approaches in Recommending reviews that need to be responded, the studies used information of already responded reviews that developers posted in app stores.…”