Mining domain knowledge from app descriptions

Abstract

Domain analysis aims at obtaining knowledge to a particular domain in the early stage of software development. A key challenge in domain analysis is to extract features automatically from related product artifacts. Compared with other kinds of artifacts, high volume of descriptions can be collected from app marketplaces (such as Google Play and Apple Store) easily when developing a new mobile application (App), so it is essential for the success of domain analysis to obtain features and relationship from them using data technologies. In this paper, we propose an approach to mine domain knowledge from App descriptions automatically. In our approach, the information of features in a single app description is firstly extracted and formally described by a Concern-based Description Model (CDM), this process is based on predefined rules of feature extraction and a modified topic modeling method; then the overall knowledge in the domain is identified by classifying, clustering and merging the knowledge in the set of CDMs and topics, and the results are formalized by a Data-based Raw Domain Model (DRDM). Furthermore, we propose a quantified evaluation method for prioritizing the knowledge in DRDM. The proposed approach is validated by a series of experiments.

Analyzing reviews guided by App descriptions for the software development and evolution

“…Thus, we also conducted a survey for Q3. The questionnaire in the survey was designed by referring to the one we used in Liu et al it includes 3 parts, and Table shows the main questions: The first and second parts aim at evaluating the recommendation and summarization of reviews in R‐TBDM separately; the third part evaluates the R‐TBDM by assessing the visualization of the information in it.…”

“…Compared with reviews, the descriptions given by developers always have higher quality and introduce the App features in a clear and systematical way. So we can extract features of Apps from them accurately: The precisions of most existing methods are above 80%, and this value is 86.15% for the method proposed in our previous work . Moreover, although the description of 1 App is short and always incomplete for giving all features commented by reviews, the ones collected from related (similar) products together can well cover the information of App features in a particular domain, and summarizing them can gain the information that concentrates on App features.…”

“…For facilitating users to read, App descriptions often conform to grammars, and this is beneficial for us to them. In our previous work, we proposed a method to extract App features from descriptions by analyzing the syntax of sentences in them . Here, we also use the method to analyze App descriptions and construct a TBDM to describe the results formally.…”

“…In this paper, we only focus on the functional App features. Many researches indicate that features are expressed by the words with particular part of speech (POS) and described by certain grammar structures, so we have defined a set of feature extraction rules based on the syntax in our pervious, and we used it here to mine App descriptions.…”

“…Based on the rules, each AppDes in the Dataset is analyzed and a set of functional features are gained, denoted by AppFeatureSet = {feature 1 , …, feature n }. We introduce the process briefly, and detailed information can be found in Liu et al For a data unit ∈ Dataset, we mine the AppDes by the following 4 steps.…”