Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 53671 (19)
- University: Sharif University of Technology
- Department: Computer Engineering
- Advisor(s): Heydarnoori, Abbas
- Abstract:
- In recent years, the number of applications in app stores such as Google Play has increased dramatically. In Google Play, people can write a review and give a rate to each application. Each review contains a lot of notable information such as ‘feature request’ which helps with software maintenance. Also, the user rate is a sign of user satisfaction or dissatisfaction. Users can change their reviews and rate anytime they want. The previous studies show that total rating affects the success of apps and the total number of downloads. They also found that a lot of users take the rating as one of the important factors while downloading the app and they usually don’t download the apps which have less than 3 star rating. Users who post reviews, expect the answer, as a result, Google Play has let developers respond to their users’ reviews since 2013 and now they can fix the user’s problems. Answering these reviews can lead to a total rating increase and consequently an increase in the app download rate. But answering reviews require a lot of time and effort. Previous studies show that only 13% to 18% of developers respond to reviews. Therefore, developers need An approach to classify the reviews. As a result, we classify the user’s reviews to respond in this study. In the first and second steps, we collected the dataset and preprocessed the reviews. In the third step, we extracted a lot of textual and semantic related features to represent each review. In the fifth step, we measured the impact of each feature, and consequently, we trained our machine learning models. Then, We used precision, recall, f1-score, and accuracy to measure the performance of our approach. We find out that the support vector machine with the precision of 0.76 for the first perspective and logistic regression with the precision of 0.86 for the second perspective work better than the other models. In the last step, for the first and second perspectives, some popular apps from different categories were selected and the performance of the best models was evaluated on them
- Keywords:
- Machine Learning ; Sentiment Analysis ; Mobile Application ; Application Stores ; User Reviews ; Classification
Digital Object List-
محتواي کتاب - view
Bookmark