Research On Emerging Topic Identification In App Reviews

User reviews generated by mobile application(app)can provide important information for app maintenance and update.Software developers can understand the software problems that users are concerned about by mining the effective information from app reviews,and then maintain apps to meet the needs of users.However,it is challenging to mine effective information from app reviews because of the short text features and noise data of app reviews.Topic mining is widely used in automatic mining of effective information in app reviews.The topic of app reviews refers to the central content mentioned by multiple reviews,and developers can quickly understand the main problems that users pay attention to through the topics of app reviews,such as software defects or software functions that affect user experience.Most of existing related researches mines topics in app reviews by topic models or unsupervised clustering.However,the topics of app reviews are faced with the problems of weak interpretability and inconspicuous differences between topics,which makes it unable to provide effective information for developers intuitively.Emerging topics in app reviews refer to app topics that users pay less attention to in the past but focus on at a specific time,such as emerging software bugs.Identifying emerging topics from app reviews timely and accurately can provide developers with more timely and intuitive information,and then maintain app to meet user needs and improve user software experience.This paper focuses on identifying emerging topics in app reviews,and the specific research contents are as follows:(1)App reviews are generally short in length and contain a lot of noisy data.Existing related research work have not fully considered the impact of error words,abbreviations and sparse characteristics of short text on emerging topic identification,which limits the accuracy of emerging topic identification.Therefore,this paper proposes an emerging topic identification method for app reviews,named IETI(Improved Emerging Topic Identification).Specifically,firstly,IETI corrects the error words in app reviews through pre-training model,and reduces noise data through a series of natural language preprocessing methods.Then,in the process of mining the topic distribution of app reviews under time series,IETI overcomes the sparse characteristics of short text through constructing biterm words.Finally,IETI identifies emerging topics from app review topic distribution through outlier detection method,and evaluates the effectiveness of emerging topics through official app changelogs.The reviews of several popular apps from different platforms are selected for experiments.Compared with the latest model,IETI has achieved higher precision,recall and F1 scores.(2)User sentiment tendencies usually affect the content of app reviews,which affect the accuracy of emerging topic identification.For example,app reviews with negative emotions tend to put forward specific problems that affect the user’s software experience.Therefore,this paper proposes an emerging topic identification method incorporating sentiment orientation prediction.In the process of identifying emerging topics,the emotional tendencies of user reviews are introduced into the contribution of related reviews to emerging topic identification through the joint modeling of sentiment tendency prediction model and IETI.Experiments show that the emotional tendencies of user reviews can help identify emerging topics in app reviews.(3)Different types of app reviews differ in text content,which affects the accuracy of emerging topic identification.Therefore,this paper proposes an emerging topic identification method that integrates text classification.Aiming at the characteristics of short texts of app reviews and the problem of overfitting of app reviews classification models,this paper proposes an app review classification method based on semantic expansion,named TLFS(Transfer Learning and Frame Semantics).The effectiveness and generalization of TLFS is verified by classifying the public manually labeled app review dataset.In the process of identifying emerging topics,TLFS and IETI are jointly modeled to identify emerging topics in app reviews under different labels and combinations of labels.Experiments show that app reviews labeled "Bug Reports" and "Feature Requests" are more helpful in identifying emerging topics in app reviews.