Arguable App Boosting Strategies - Are They Working in a Small Social Community?

Yuhong Liu1 and Yan (Lindsay) Sun2
1 Department of Information Science and Technology, Penn State Altoona, Altoona, 16601
2 Department of Electrical and Computer Engineering University of Rhode Island, Kingston, RI 02881,

Abstract—With the big success of the mobile application (app) sales, diverse app sales boosting strategies have attracted wide attention. However, many of these strategies may raise security concerns since they attempt to allure users by providing dishonest information. In this paper, we survey current arguable app sale boosting strategies and discussed the effectiveness of these strategies. Based on an app installation data set collected from a university campus community, we quantitatively investigate the effectiveness of these arguable app boosting strategies and find out that for a closely connected social community, the manipulation of app ratings/reviews cannot significantly affect app downloads when other social factors are present.

With the wide spread of smartphones and tablet computers, the sale of mobile applications is experiencing an overwhelming growth. In June 2012, Apple’s app store hit the 30 billion downloads milestone. Accompanied with this, dozens of other companies, such as Google, Amazon, BlackBerry, have also opened their own app markets and are achieving big success. 

Businesses from all over the world are attracted by this opportunity. App developing companies/individuals aim to make profit by selling their apps in the app store. A success story is that Steve Demeter, an independent developer, has made $250,000 in just two months by his $5 app “Trism” [1]. Other businesses, e.g. Starbucks, Macy’s, Bank of America, etc., target on promoting their businesses through the official apps. All these companies share the same goal: attracting more users to download their apps.

To achieve such goal, diverse app boosting strategies are gaining popularity [2]–[5]. Some of these strategies may make profit by misleading app users’ opinions, which makes them arguable or even malicious. How to protect the app market from malicious app boosting strategies can be a very interesting research question. However, before developing defense schemes, a key question to answer is whether these arguable or malicious app boosting strategies can really increase app downloads effectively. 

In this paper, we aim to answer this question in a small social community. Specifically, we first conduct a survey on existing arguable app boosting strategies in Section II. Then, in Section III, we analyze the effectiveness of some specific strategies using real user data collected from a university campus community. Finally, we summarize our findings and discuss their implications in Section IV.


As more and more developers and entrepreneurs are attracted 
by the promising app market, how to make profit has 
become one of the most heated discussions. Diverse strategies have been considered as effective in boosting app sales. In this section, we summarize these strategies according to generally accepted opinions.

A. Improve App Rank in the General App Market
It is commonly believed that apps with higher ranks in the app charts are more likely to be installed. Therefore, promoting the rank of apps is the primary goal of different boosting strategies. In this section, we would like to answer the following questions.
  • What are the factors affecting app rank in general app market?
  • Why the download number is believed to be an influential factor?
  • Why the rating/review is believed to be an influential factor?
  • What are the arguable strategies used for manipulating (e.g. falsely boosting) app rank?
Let’s first look at how app markets rank their apps in their featured categories.

Before April, 2011, people believed that Apple determined app rankings purely based on the download number. Some studies, such as [6], [7], even figured out how Apple assigned weights to sales on different days. Since April, 2011, Apple has changed its ranking algorithm by considering more factors other than just download number. People believe that some qualitative information, such as ratings/reviews and frequency of usage, is taken into consideration. Different from Apple app store, Google Android market employs more complex ranking algorithms which consider app download number, retention rate, usage frequency, rating values, number of ratings, installing/ uninstalling rate and so on [8].

By comparing these two different app markets, we find out that in both markets, (1) download number is the most important factor in determining app rankings; and (2) ratings and reviews start to play more important roles in influencing app rankings. Therefore, we would like to analyze these two factors in details below. 

Download Number: An app’s download number and its rank facilitates each other. On one side, a larger download number helps to promote an app’s rank. On the other side, a higher rank will cause more exposure and more new downloads. The new downloads will further increase the total download number and lead to higher rank. Therefore, app developers try all kinds of methods to increase the download number, such as putting apps on sale, making app updates, conducting advertisements and even hacking into the app market.

Ratings and Reviews: Ratings and reviews may affect app developers’ revenue in two ways. First, current app ranking algorithms are taking more consideration about apps’ rating and review information [8]. And higher rating scores or positive reviews will lead to a higher app rank. Second, when comparing several similar apps, users tend to choose the app with higher rating scores or more positive reviews. 

Tricks for increasing the number of downloads: To manipulate the app download number, one well known way is the pay-per-install model, where app developers pay for each install to drive the download number [9]. There are usually two ways of pay-per-install. First, some companies, such as Tapjoy [10] and Flurry [11], provide pay-per-install networks composed by plenty of apps. Once your app joins this network, its download number will be dramatically promoted by other apps in the network. For example, some apps encourage their users through virtual currency or level upgrading to download your app. Once your app has been installed due to this promotion, you will pay the company (e.g. Tapjoy, Flurry) and the apps that generate this install. Second, some companies, such as App Lifter [12], provide services for app developers to directly pay users for installing their apps. Usually the users will be paid a little bit higher than the app price [13].

Tricks for obtaining positive ratings and reviews: Numerous advices and tricks on how to obtain positive ratings and reviews are prevalent. Here are some examples.
  • Getting negative feedback coming to the developers instead of the app store [2]. Some app developers implement a feedback interface with two buttons: “send love” and “send feedback”. When users click the first button, their positive ratings and reviews will be sent to the app store, while when users click the second button, their criticized opinions will be sent to the developers. It works as a filter and only leaves the positive ratings and reviews in the app store.
  • Asking for ratings after several usages [3]. Usually users who do not like the app will quit using it after very few uses. Therefore, developers are suggested to ask users’ feedback after users have used this app for several times. These users will provide positive feedback with higher probability.
  • Learning from feedback of competitors’ apps [2]. An app user usually has similar expectation for apps with similar functions, and his/her comments on your competitors’ apps could be a useful guide for improving your own app.
  • Conducting frequent app updates [3]. An update which improves previous shortcomings could drive more positive ratings and reviews.
Hack in App Stores: Besides the above methods, the developer can hire a company that generates positive ratings and reviews. The same companies that help the developers increase the number of downloads can surely help insert positive ratings and reviews. Besides the methods discussed above, there exist diverse ways to hack the app store. Although Apple keeps its ranking algorithm as a secret, there are some companies that have figured out at least part of it. Temple jump, a copycat of the popular game - Temple run, had taken up the 1st rank of paid apps for a while, whereas its average
rating value is only one and half stars. Currently, this app and 58 other apps from the same developer are removed from the app store [14]. Some companies, such as ComboApp [15], GTekna [16], provide services to promote app rankings through diverse ways. And in some cases, the companies even provide guaranteed services to push apps into the top lists of the app store. On the other side, Apple has posted a notice [17], and said that the manipulation of apps ranking or services that guarantee the placement of the top lists were prohibited, and violations may lead to revoking of the developer program
membership. In November 2010, an iTunes user posted on Apple forum that his store credit was stolen to purchase some iPhone apps, and his personal profile was modified, especially that the home address was changed to Towson, Maryland [18]. Unexpectedly, after this post, more and more users start to share their similar store credit stolen stories continuously, and their addresses were all changed to Towson, Maryland.

There definitely exist some malicious people who make profit through these transactions, which are so called “Towson Hack” [19]. The attackers could make money by (1) using these stolen accounts to purchase either their own bogus apps or the apps that they aim to promote, or (2) selling the access of the victim accounts to other users.

B.  Improve App Rank in the Search Results

The app which is on the top list of the search results usually will have a larger chance to be installed by the second type of buyers. Then how do these App stores rank apps in their search results? Generally speaking, key word match, apps’ downloading number, ratings/reviews are factors considered by app search algorithms.

According to [21], [22], both the key word match and number of downloads will greatly affect the ranking of the search results, whereas the impact of ratings/reviews is obscure. Specifically, the search algorithm is not identical among different App markets. Google Android market maintains the same search rules as Google website, whereas Apple App store keeps its own search rules. In Apple App store, apps containing the exact keywords in the titles, descriptions or developer names will be listed higher, whereas

apps with synonyms will not be searched out. Some developers take advantage of this and name their developer account as “best free apps”, a popular search key word, or “angry birds”, the name of a popular game. Surprisingly, they successfully get high ranks in the search results because of it [22].

With the explosion of apps, more and more third-party companies have joined in the app search engine competition. Some of them, such as chomp, Quixey and Appgravity, have already developed advanced algorithms for searching apps [23]–[25].

Compared to App markets’ own search algorithms, these app search engines consider more factors, such as app price and functionalities, and even support app searches on multiple app markets, and therefore have also attracted a large amount of customers. Since these companies are still at their initial stage,

there are not many discussions about how to improve app ranking in their search results yet.

C. Influence Buyers’ App Shopping Decisions through Social Factors

With the emergence and rapid development of social media, the competition of boosting app sales goes beyond the App store. The third type of app buyers, who tend to follow others’ choices, will be greatly influenced by the social factors. App developers are suggested to cultivate online communities who are interested in their apps. Besides this, online forums, message boards and blogs are also good places for app promotions. Companies [26], [27] provide services to popularize apps in social networks (e.g. FaceBook, Twitter, etc.), and submit apps to review websites.


In this section, we will describe a quantitative study on app downloading behaviors in a very special community: university campus. The results for this special community may not be generalized to a broader consumer base, but will yield some interesting insights. We aim to understand, for this university campus community, (1) whether manipulating the ratings and reviews can significantly affect app downloads when other social factors are present, and (2) the importance of offline social factors, such as telephone calls, face-to-face meetings, etc. Some interesting results have been obtained.

A. Data Set
We use a real user data set collected by MIT Media lab [20] as the testing data set. This data set, collected from March to July 2010, recorded the installations of 821 apps from 55 participants who were residents living in a graduate student residency of a major US university. In this data set, the following information was collected. 
  • App related information, such as app name, prices, ratings and global download number.
  • User related information.
    • Users’ app installation information (i.e. which user installed which app at what time).
    • Call log and bluetooth hits information. During the data collection period, each participant was given an Android-based cell phone with a built-in sensing software to capture all call logs and bluetooth hits among the given phones. Call logs were used to indicate participants’ interactions through phone calls. Bluetooth hits recorded participants’ face-to-face interactions, during which the phones were within each other’s vicinity. These two types of information described participants’ daily interactions. 
    • Users’ friendship, affiliation and race information was also collected through a survey. In the survey, each participant provided his/her affiliation and race, and rated his/her friendship relationship to other participants. Such information reflected more about participants’ long term relationship. 
    • Using this data set, we can verify whether some commonly acknowledged app boosting strategies can really boost app sales in the university campus community. Due to the difficulties to collect the app installations for each individual user, we can only conduct the following analysis for the 55 users. The results obtained may be applied to other closely connected social communities, but may not be applicable to everyone.
Finally, we would like to emphasize that the results presented in this section describe app download behaviors in a closely connected social community. These results are particularly helpful for app developers who want to promote the apps designed for special interest groups or special communities. Although these results may not be generalized to the global app market, they provide a new view point for further investigations.

B. Manipulations of Rating and Download Number
As presented in Section II, ratings and download number are two of the most important features to determine an app’s rank in app markets. It is widely believed that the installations of an app can be greatly boosted by an increase in its rating values or download number. Therefore, many companies provide diverse app promotion services by manipulating app ratings or download number. A well known method is the pay-per install model mentioned in Section II.

From May 2011, Apple started to reject apps joining the pay-per-install networks. However, it would be very difficult to detect the apps directly paying users money for installation. Tapjoy, Flurry, App Lifter, and other similar companies are still making money by providing app downloads promotion services. Such kind of manipulations turn the app market into an unfair competition environment and raise a new challenge to the security researchers.

To estimate the impact of app ratings and download number, we apply the prediction model proposed in [20], which predicts app installations by constructing a composite network containing multiple sources of information. When compared with other models, this prediction model yields a significantly higher prediction accuracy. To our best knowledge, this is currently the best model in terms of predicting app installations considering social factors. Therefore, we adopt this prediction model as the base to investigate the manipulation gain.
In [20], the prediction model considered users’ social information within the community, as discussed in Section III-A. In this paper, we introduce app rating and download number as additional input information to the original prediction model. Thus, all possible factors that can be used to predict app download numbers are shown in Table I.

As described in [20], the prediction model is developed as a discriminative model which combines different social factors, and is tuned by a mathematical optimization process. The output of the prediction model is the probability that user k installs a specific app a. This probability is denoted Pa(k). The details can be found in [20].

For a given app a, we use the prediction model to calculate the impact of rating or download number as follows. 
1. Optimizing the parameters of the prediction model. We use all available data (i.e. information I1~I7 for all participants, as well as the apps they have installed) to train the prediction model.
2. Calculating download probability before manipulation. For a given app a, use the optimized prediction model to predict the probability that user k installs the app. This probability is denoted by Pa(k)org. The input of the prediction model is the information of user k (i.e. I1 ~I5) and the information of app a (i.e. I6 and I7). This calculation is performed for all 55 users.
3. Adjusting the app information, as if manipulation has occurred. If we study the impact of app rating, we increase the app rating value (i.e. I7) by a certain amount. If we study the impact of the download number, we increase the download number (i.e I6) by a certain amount.
4 Calculating download probability after manipulation. Use the user information and the adjusted app information to predict the download number of app a. Let Pa(k)adj denote the probability that user k will download app a after the adjustment of app information. Obliviously, Pa(k)adj should be no less than Pa(k)org.
5 Computing the total download increase. The total download increase for app a due to the manipulation, denoted by Mainc is calculated as
where N = 55.

1) Impact of Rating Value Increase: Among the 821 apps collected in the data set, 5 apps are preset in the Android phones. Besides these 5 apps, only 273 apps have been installed by at least two users. In the discussions below, we just consider these 273 apps. To train the optimized prediction model, we construct the composite network by considering the
information I1; I2; I3; I4; I7 in Table I. Except rating values, the other four types of information were found to be influential factors in predicting app installations [20].

Figure 1 demonstrates for each specific app how many more installations will be triggered when we increase the app rating value by a certain amount. In Figure 1, the x axis represents the app index, and y axis represents the installation probability increment. The rating information used for prediction is the raw rating value ranging from 1 to 5. Figure 1 is obtained by increasing the rating value of each app by 1.

Figure 1. App Installation Probability Increment Vs Rating Increment.

From Figure 1, we can observe that for each app, when the rating value increases by 1, the installation probability also increases. However, the installation probability increase value
is very small, around 10􀀀-10. It indicates that for a closely connected community, the global rating information has trivial impact on influencing users’ app installations. The possible reason is that in such kind of community, instead of global rating information, users could refer to their friends, colleges or family members for app installation recommendations. In other words, the local app “rating” information, which is reflected by installations from a users’ local connections, overwrites the global rating information, and has significantly influenced users’ app installation decisions.

2) Impact of Download Number Increase: Similarly, we trained the prediction model to investigate the impact of current app download number on the future app installations. To train the optimized prediction model, we construct the composite network by considering the information I1; I2; I3; I4; I6 in Table I.

Through the experiment, we have observed that (1) the relationship between the download number and installation probability is also a linear curve, meaning that the installation probability monotonically increases with the download number increase; and (2) when the download number increase by 106, the installation probability increment is only about 0.85* 10-􀀀14, which is negligible. Similarly, for a closely connected community, the global download information does not have an obvious impact either. The possible reason is that when users have closely connected friends, colleges or family members to obtain app installation suggestions, the global app download information is not important any longer.

As a summary, we have studied the impact of apps’ rating information and download number on users’ app installation decisions. The value change of these two factors will not affect users’ decision too much. Recall that in Section II, we discuss that some companies provide promotion services by manipulating apps’ rating information or download number. Based on our experiment results, this type of manipulation is definitely not a good choice. Note that, our data is a university campus community data, where local contacts (e.g. call log, friendship, and etc) may dominate users’ installation decision. And these results may not be able to apply directly on other type of user communities. However, we do provide a way to estimate the cost and gain of the rating and download number manipulation strategies. The app developers who plan to promote their app sales through the rating and download number manipulations need to reconsider its effectiveness carefully.


In this work, we survey current popular strategies for app sales boosting and further evaluate their effectiveness based on real user data collected from a university campus community. The interesting experiment results show that some commonly acknowledged app sales boosting strategies may not work as expected. App developers could refer to this work for selecting their promotion strategies carefully. Furthermore, for research purpose, this work quantitatively evaluates the effectiveness of different app boosting strategies, and discussed all possible reasons, which helps to build up fundamental understandings of the app market.


[1] B. X. Chen, iPhone Developers Go From Rags to Riches, September, 2008,
[2] D. Wooldridge and M. Schneider, The Business of iPhone and iPad App Development Making and Marketing Apps that Succeed. Apress, 2011.
[3] K. Yarmosh, App Savvy- Turning Idears into iPad and iPhone Apps Customers Really Want. O’Reilly, 2011.
[4] H. Koekkoek, The Amazon Appstore: Show Me the Money, Feberary, 2012.
[5] J. Lowensohn, How apps stay on top in the App Store, May, 2011,
[6] Apple’s iphone app ranking algorithm figured out, September, 2009,
[7] What is the algorithm behind the App Store rankings?,
[8] R. Charles, Android Market Ranking Algorithm: The New Black Box, August, 2010,
[9] B. Lenox, Cost per Install App Marketing, February, 2011,
[10] Tapjoy Marketplace,
[11] Flurry,
[12] App Lifter,
[13], Pay Per Install A Legit Way To Send Your iOS App Up The Charts?,
[14] M. Dan Moren, Apple warns developers against gaming App Store rankings, February, 2012,
[16] GTekna,
[17] Apple, Adhering to Guidelines on Third-Party Marketing Services, February, 2012,
[18] iTunes store account hacked. [Online]. Available:
[19] M. Lex Friedman, The Towson Hack: The mystery of vanishing iTunes credit, September, 2011,
[20] W. Pan, N. Aharony, and A. Pentland, “Composite social network for predicting mobile apps installation,” in Proceedings of the 25th Conference on Artificial Intelligence (AAAI-11), August 2011.
[21] M. Nilsson, Want to Rank in iPhone App Store Search? Get your Keywords Right,

[22] S. Kittelson, ”What factors does the search algorithm for the iTunes App Store take into account?”,

[23] M. Bryant, Appgravity is like a better Google Search for Android apps,

[24] J. Evans, App search engine competition heating up,

[25] S. Martin, Quixey bags $20 million for App search,

[26] ComboApp, mobile app promotion agency,

[27] Marketing Your App. [Online]. Available:

Yuhong Liu is an assistant professor in the Department of Information Science Technology at Penn State Altoona. She received her Ph.D. degree from University of Rhode Island in 2012. She was the recipient of the 2013 University of Rhode Island Graduate School Excellence in Doctoral Research Award. With expertise in trustworthy computing and cyber security, her research interests include three major directions: online social network security, trust management in cyber-physical systems and trustworthy cloud computing. Her work on securing online reputation systems received the best paper award at the IEEE International Conference on Social Computing 2010 (acceptance rate = 13%). 

Yan (Lindsay) Sun received her B.S. degree with the highest honor from Peking University in 1998, and the Ph.D. degree in electrical and computer engineering from the University of Maryland in 2004. She joined the University of Rhode Island in 2004, where she is currently an associate professor in the department of Electrical, Computer and Biomedical Engineering.

Dr. Sun is an elected member of the Information Forensics and Security Technical Committee (IFS-TC), in IEEE Signal Processing Society. She is the Chief Editor of Sigport. She also serves on the editorial broad of IEEE Security and Privacy Magazine. She is an associate editor of Signal Processing Letter since 2013, and an associate editor of Inside Signal Processing eNewsletter (2010-2014). Dr. Sun’s research interests include power grid security, trustworthy social computing, wireless network security, and reliable biomedical systems. She applied signal processing techniques in modeling, detection, and estimation of abnormal behaviors in various computing and communication systems. Dr. Sun received the best paper awards at the IEEE International Conference on Communications (ICC’14) and the IEEE International Conference on Social Computing (SocialCom’10). She was the recipient of NSF CAREER Award. She is an IEEE senior member.