Do popular apps have issues regarding energy efficiency?

Keywords

There are several common techniques used in app review analysis such as sentiment analysis, text mining, and keyword analysis. In this study, I mainly focused on keyword based analysis that could provide valuable insights into what users are talking about. To identify and analyze application energy consumption related app reviews by leveraging keywords, it is required to determine the most relevant problem-specific keywords that indicate energy issues or problems that users may be encountering with the app. For this purpose, previously identified energy-related keywords were obtained from existing studies (Wilke et al., 2013; Phong et al., 2015).

Table 1 provides a list of the keywords that are anticipated to be used in energy-related reviews. The first column, Keywords, refers the energy-related keywords which are battery, power, and energy. However, searching these keywords alone in the reviews can lead to false positives since the review may contain off-topic information. For example, the following review includes the energy keyword although it is not related to the energy consumption of the specific application being discussed: “I only got this App to keep in touch with family in Australia. I’ve spent the past hour making requests for a verification code to set up my account. I’ve got nothing, either by text or email. I can’t help but feel that these people just don’t care. -Utter waste of time and energy.”. To address this issue, adjunct words, such as those in the second column of Table 1, can be used in combination with the keywords. Stars in the adjunct words indicate a wildcard character that can represent one or more characters in the words (e.g., drain battery, draining battery, drains battery). Although the use of adjunct words can provide additional context and help reduce the risk of false positives, there is still a chance of such events occurring. As shown in the previous example, the adjunct word “wast*” is used in conjunction with the keyword “energy”. Therefore, instead of solely relying on keywords with adjunct words to extract app energy-related reviews, a preferred approach involves combining keyword-based analysis with the manual analysis of reviews that contain keywords. This approach ensures the extraction of more accurate data by eliminating irrelevant information.

Application store

There are different app stores available for users to download and install applications, such as Google Play Store, Apple App Store, and Amazon App Store. As the third quarter of 2022, Android users are able to download approximately 3.55 million apps from the Google Play Store, which makes it the biggest app store in terms of the number of available apps (Statista, 2022). As the same period of time, the Apple Store is the second-largest app store with approximately 1.6 million available apps for IOS and The Amazon Appstore is the third-largest app store with around 480,000 available apps for Android users.

Besides the number of available apps, Android-based mobile devices have a larger market share compared to IOS-based mobile devices. In the fourth quarter of 2022, Android have a global market share of over 71.8%, while IOS have a market share of around 27.6% (Statista, 2023). This means that the majority of smartphones sold worldwide run on the Android operating system.

In this study, Google Play Store was selected to investigate energy-related app reviews by leveraging keywords since it is one of the most popular app stores globally and widely used by Android smartphone users. It also allows users to leave detailed reviews and provide ratings for the apps they have downloaded. Moreover, it supports a variety of features that can help in the analysis of app reviews, such as filtering by language and sorting by most recent reviews. Overall, Google Play Store provides a large and diverse pool of reviews that are suitable for extracting information that is relevant to the specific aspects.

Although the Google Play Store does provide information on the number of app reviews and downloads for an app worldwide, the app ratings can vary by region due to differences in user behavior, preferences, and needs, as well as regional competition. For example, while Zoom app had a rating of 4.2 in the US region when the data was gathered, it had a rating of less than 4.0 in the Turkey region. Note that, all the data gathered for an app in this study is from the Google Play Store in the US.

Applications

This study has been conducted by leveraging popular apps, which refer to widely used and well-known mobile applications. The reason for choosing popular apps as the subject is based on several factors. First, popular apps can be considered reliable because they are often supported by a team of dedicated developers from large tech companies. These companies may invest a lot of funding for developing robust apps. Second, popular apps are frequently updated to fix bugs and address any issues that may arise, as well as to maintain their popularity and user satisfaction. Although these apps may have been well-tested before public releases, it is still possible bugs or other issues arise after the release. Third, popular apps can receive a significant amount of app reviews from users due to their widespread usage and visibility. By monitoring the feedback, developers can quickly address the users’ concerns and complaints. Overall, it can be expected that popular apps might be less problematic than other apps.

Table 2 lists the specific applications that are being considered in this study. The first two columns, Application Name and Category, list the name of each application and the category to which it belongs, respectively. The third column, Ratings, shows the overall rating given by users for each application. The final columns, Reviews and Downloads, shows the number of times each app has been downloaded and reviewed at the time right before data collection for each app, respectively. While the review count indicates the total number of user reviews that have been submitted for the app, the download count represents the total number of times the app has been downloaded.

These apps have been chosen among popular apps based on the criteria of having ratings of at least 4.0, more than 3 million reviews, and more than 100 million downloads from the Google Play Store. These selection criteria can indicate that an app is well-received by users and is likely to be of high quality. For example, the TikTok app has a 4.4 rating, 54,7 million reviews, and over one billion downloads. A large number of reviews can also provide a diverse range of opinions and perspectives from different users.

Note that, the 32 selected apps come from 16 different categories on the Google Play Store, including Communication, Social, and more. Table 3 gives the examples of each category that are provided by the Google Play Store (2023). Depending on the category an app belongs to, users may prioritize different aspects of the app. For example, apps in Map & Navigation category typically use GPS, which can consume a significant amount of energy. Therefore, it might be expected for users to leave reviews related to the energy consumption or battery drain of apps in this category. In contrast, users reviews for apps in education category may be less likely to include comments related to energy consumption or battery drain since they are not typically known to be heavy consumers of energy, as they generally do not require intensive processing or constant use of device sensors like GPS. Analyzing reviews from apps in different categories can be useful in identifying user priorities and common themes related to energy consumption and efficiency.

Data retrieval

To obtain the necessary app review data, a tool called App Review Collector (ARC) have been developed for Android App Store. The ARC tool provides various features including the ability to retrieve the review data written in a specific language, collect a specific number of most recent reviews, and save data in different file format such as json and csv.

For each considered app that is being analyzed, a variety of data points such as the date and the content of the review, the user’s rating, as well as any replies made by the app developer along with their respective time and content were collected via the ARC tool and the language for review data is set to English. The computer used to collect review data was configured with Intel i5-2400 CPU 3.10 GHz, 8 GB of DDR3 memory, a 500 GB 7200 RPM SATA disk drive, and runs on Windows 10. During the data retrieval process, the upper limit of reviews collected per app was set to 500,000 due to the limitations in computer resources, the memory-intensive nature of the process, and the variability in app review data sizes. Additionally, app review data was considered within a range of 5 years (between January 2018 and January–February 2023) to ensure that the collected reviews are relevant and reflect the current state of the app since the age of the app can vary greatly.

Table 4 provides information about the collected review data for each app in January–February 2023. The first column, Application Name, lists the name of each application. The second column, Collected Reviews, shows the number of reviews limited to either the most recent 500,000 reviews or the last 5 years, whichever comes first. Note that, the collected data includes reviews exclusively in the English language. Third column, Beginning of Reviews, indicates the earliest user review date for the collected data. For example, the Zoom app has a total of 500,000 user reviews, which are collected starting from November 2022. On the other hand, the Twitch app has 244,240 user reviews, which are collected starting from January 2018. Also, there is an exceptional case for the CapCut app that sets it apart from other applications, once it is carefully examined. The CapCut app has less than 500,000 user reviews and its earliest user review date is in April 2020. The reason behind this is the first release date of the app is in April 2020. In total, 13,818,950 user reviews for 32 apps have been collected.

RQ1: energy issue

App reviews for mobile applications, especially concerning energy consumption, are diverse in nature and may mainly focus on issues or problems encountered by users. In addition to that, positive reviews may also encountered in the feedback in case of user satisfaction with the app’s energy efficiency. Therefore, to answer the first research question, app reviews in the dataset need to be categorized as either negative or positive.

My initial attempt was to rely on the review scores, which are on a scale of 1 to 5. However, this option has a major drawback. The review score and the review content may not be correlated for the investigated aspect. The app might have energy consumption-related issues, but it can still receive a high score, as seen in the review example, “Works well drains battery.” since users’ evaluations, considerations, and preferences can vary. The user complains about the energy consumption of the app, but the given score was 4. The other option was to perform sentiment analysis, which required providing explicit lists of positive and negative words to determine the sentiment of the text. However, even with such positive and negative word lists, its effectiveness is limited when applied to domain-specific language, such as app energy consumption reviews. For example, the reviews “Good, battery usage is very high.” and “Best app ever… other than battery consuming.” can be categorized as positive while “Seems to have fixed the severe battery drain issue.” can be categorized as negative using sentiment analysis. An alternative option was to utilize artificial intelligence APIs, which offer pre-trained models for sentiment analysis. However, a significant drawback of this approach is that most of these APIs come with a cost, making them less accessible for free usage.

As a result, I have decided to manually tag the energy consumption-related app reviews as positive and negative. This approach will not only provide more precise sentiment analysis results but also enable the creation of a labeled dataset for future research. Having a labeled dataset will be beneficial for training and evaluating sentiment analysis models specific to app energy consumption reviews and improving their accuracy. Note that while negative categorization includes user reviews that contain complaints, point out missing features, or make comparisons with other apps in a negative manner, positive categorization consists of reviews expressing appreciation and making comparisons with other apps in a positive manner in terms of app energy consumption. App comparisons in the reviews are often done with competitor apps, such as those that offer similar features or functionalities, for example, Firefox, Google Chrome, and Opera Browser.

Figure 1 shows the percentage distribution of negative and positive app reviews in terms of energy consumption for each app. While the x-axis represents the app names, the y-axis represents the percentage scale in the figure. The red (values above the y-axis 0) and green (values below the y-axis 0) colors are used for negative and positive reviews, respectively. Based on the Fig. 1, it is observed that all the popular apps analyzed are facing energy consumption-related issues from the user’s perspective. The high percentage of negative reviews indicates that users have expressed concerns or dissatisfaction, specifically regarding energy consumption. For 21 out of 32 apps, the negative review percentage is more than 94%, and Uber, Snapchat, and TikTok apps have the highest negative review percentage. Out of the remaining 11 apps, four apps have a negative review percentage between 89% and 94%, five apps have a negative review percentage between 82% and 87%. For the Opera and Spotify apps, there are notable exceptions regarding negative review percentages. Despite having more negative reviews than positive ones, they stand out from other apps due to their higher positive review percentages. This suggests that while they may have some issues or complaints in terms of app energy consumption, they also receive positive feedback for certain aspects or features that contribute to their energy efficiency. Briefly, even popular apps can face energy consumption issues that might negatively impact the user experience with a mobile application.

RQ2: correlation

To investigate whether there is a correlation between the given app ratings and the app reviews related to energy consumption, a quantitative-based analysis have been conducted. Table 5 shows the percentage distribution of given ratings for both positive and negative reviews across 32 different apps. The first column, Rating, represents the possible rating values range from 1 to 5, where one represents the lowest rating, and five represents the highest rating. The remaining columns, Negative reviews (%) and Positive reviews (%), indicate the percentage of negative and positive reviews corresponding to each rating, respectively. For example, 39.86% of reviews with negative comments received a rating score of 1.

Rating percentages for negative reviews, in which the content expresses dissatisfaction with app energy consumption, as presented in Table 1, show that 57.41% of the users give a rating score of 1 or 2. This suggests that the primary focus of these reviews is to voice complaints specifically about the app’s energy consumption. Conversely, the remaining 42.59% of user reviews have been assigned a rating score of 3 or more. This indicates that, despite concerns regarding energy consumption, they may have considered other aspects of the app, such as functionality, performance, and distinct features, to be more noteworthy.

Rating percentages for positive reviews presented in Table 5 imply a connection between user contentment with the app’s energy consumption and their propensity to assign a higher rating score, notably a rating of 5, as 82.25% of user reviews have been assigned a rating score of 5. This correlation suggests that users who view the app’s energy usage positively are more inclined to provide favorable feedback by assigning the highest ratings. Furthermore, it is evident that the emphasis is primarily on energy consumption, rather than on the other features of the application. This is also supported by the near absence of low or medium scores, such as 1, 2, or 3, being given. Another point to note is that sometimes, the given rating score and the content of the review may not match. A user might accidentally give a rating of 1 instead of 5 or misunderstand that the highest score is 1. Therefore, the fact that 0.93% of the positive user reviews are associated with a rating of 1 could be attributed to such instances of rating inconsistency. For example, the following user review, “Nice app it doesn’t uses many battery and data that’s why I like it” received a rating score of 1.

When assessing the collective energy-related app ratings, none of the apps received ratings of 4 or higher in the overall context of combined energy-related app ratings, as illustrated in Table 6. For the purpose of recall, ratings of 4 or higher were one of the selection criteria for these popular apps. The CapCut, Opera Browser, and Wattpad apps received the highest rating scores of 3.36, 3.24, and 3.14, respectively. Conversely, the Messenger, Google Chrome, and Uber apps obtained the lowest rating scores, registering 1.74, 1.93, and 1.93, respectively. Upon analyzing the energy-related app ratings for all of the considered applications, it is found that the combined average energy-related app rating stands at 2.52 out of 5. This rating score is significantly low compared to the average rating of the considered popular apps, which is 4.37, suggesting the need for enhancements in addressing concerns related to app energy consumption.

Recognizing that users value not only functionality but also responsiveness and resource optimization in an app, the implementation of energy-efficient practices holds the potential to reinforce positive user perceptions. Even in the case of already well-rated apps, incorporating energy-efficient strategies can result in incremental improvements in user review ratings. Users may respond positively to applications that not only meet their functional needs but also demonstrate a commitment to efficiency and resource optimization.

In practice, this incremental improvement could manifest as users upgrade their reviews from, for example, ratings of 1 and 2 to the highest scores, such as 4 and 5. Notably, the majority of app energy-related reviews are negative, as shown in Fig. 1, and 96.27% of positive app energy-related reviews receive ratings of 4 or 5, as shown in Table 5. Such positive feedback not only contributes to an enhanced app reputation but also fosters a more satisfied and loyal user base, ultimately benefiting the app’s overall success and longevity in the competitive app market. In the case of popular apps, with an average rating of more than 4 and the expectation of encountering fewer issues, the impact of energy efficiency enhancements on their already high ratings may be less pronounced. Nevertheless, significantly low energy efficiency-related app ratings may guide developers in their efforts to enhance energy efficiency and, consequently, their app ratings.

In addition to that, while a correlation can be observed between positive reviews and their corresponding score ratings, a precise correlation cannot be observed between negative reviews and the assigned score ratings.

RQ3: category and functionality

App store categories help users easily find and browse apps that align with their interests and needs; they also group competitor apps within the same category, enabling users to compare and choose from similar apps more conveniently. The categorization of apps generally reflects their distinctive functionalities and features. Exploring the relationship between user reviews, particularly those with high negative ratios regarding energy consumption, and expectations based on app categories and functionality may reveal valuable insights. Among the apps listed in Table 4, Waze Navigation & Live Traffic, Avast Antivirus & Security, Zoom, Firefox, and Microsoft Teams apps have the highest number of negative review ratios compared to the other apps. These apps are in the categories of Map & Navigation, Business, Communication, Business, and Tool, respectively.

Apps in the Map & Navigation category are often associated with high energy consumption due to their usage of GPS functionality. Since there is an app that has lower negative review ratios in the Map & Navigation category, it may not be solely possible to attribute negative reviews to the nature of this category. Furthermore, while the Uber app is categorized in the Map & Navigation category, the Google Maps app is categorized under Travel and Local category as provided in Table 2. It is indeed anticipated that Waze and Google Maps apps would be in the same category, given their more similar functionality, compared to Uber app. Even if Waze and Google Maps apps were in the same category due to their similar functionality, they do not align with expectations in a similar manner; one exhibits high negative ratios in user reviews regarding energy consumption, while the other does not. Additionally, considering that Firefox, Google Chrome, and Opera Browser share the same category and functionality, one might expect similar energy consumption patterns. However, Firefox differs from them in terms of high negative energy consumption-related user review ratios. In the Tool category, Avast Antivirus & Security is the only app, and given the absence of clear expectations for this category in terms of energy consumption, it may not be thoroughly evaluated. For the Business category, both the Zoom and Microsoft Teams apps exhibit a high number of negative review ratios. The high negative review ratios for both the Zoom and Microsoft Teams apps in the Business category are likely attributed to the functionality of the apps as expected, particularly features such as meetings, video conferencing, and online classes, rather than being linked to the app category. Similarly, a high negative ratio might be expected for the YouTube app in the Video Players & Editors category, along with other video platform-based apps in the Entertainment category, owing to their characteristic focus on video content. However, this expectation does not align with the user reviews.

Consequently, the alignment between user reviews with high negative ratios regarding energy consumption and expectations based on app categories and functionality is not consistently accurate.

RQ4: trend

To address whether there is a changing trend in app energy consumption issues according to the collected reviews, an approach is adopted that involves analyzing the percentage of user reviews with negative feedback regarding energy consumption issues, considering equally spaced time intervals. This approach allows for a systematic evaluation of how app energy consumption issues have been vary over time. The time intervals consist of 15 data points each, ensuring a minimum of 1 week between intervals for all the apps under consideration, given the variation in starting points of app reviews. As an example, the Zoom app’s review start date is November-22, while the AliExpress app’s review start date is January-18, as indicated in Table 4, for the collected data.

Figure 2 shows the presentation of a plotted line graph for each app, with app names arranged sequentially from left to right. In each plot, the x-axis represents the time intervals, while the y-axis represents the percentage of user reviews with negative feedback for the app. The y-axis values in the figure vary for each individual app represented in the visualization. Based on the data, it is possible to categorize the trends in energy consumption issues of apps as either decreasing or fluctuating. For most of the apps, specifically 25 out of 32, there is a fluctuating trend with variations, but there isn’t a clear overall direction. On the other hand, Avast Antivirus & Security, eBay, Google Photos, Microsoft Teams, Pinterest, Waze Navigation & Live Traffic, and Zoom apps exhibit a decreasing trend in energy consumption issues. This may indicate that their energy efficiency may have improved or their usage may have become more optimized. The significance of the result might be further highlighted by the fact that among all the analyzed applications, Avast Antivirus & Security, Microsoft Teams, Waze Navigation & Live Traffic, and Zoom apps have received the highest number of negative reviews, all of which are part of the top 5.