TruBeRepec: a trust-behavior-based reputation and recommender system for mobile applications

Abstract

Mobile applications are software packages that can be installed and executed in a mobile device. Which mobile application is trustworthy for a user to purchase, download, install, execute or recommend becomes a crucial issue that impacts its final success. This paper proposes TruBeRepec, a trust-behavior-based reputation and recommender system for mobile applications. We explore a model of trust behavior for mobile applications based on the result of a large-scale user survey. We further develop a number of algorithms that are used to evaluate individual user’s trust in a mobile application through trust behavior observation, generate the application’s reputation by aggregating individual trust and provide application recommendations based on the correlation of trust behaviors. We show the practical significance of TruBeRepec through simulations and analysis with regard to effectiveness, robustness, and usability, as well as privacy.

Appendix

1.1 A. Evaluation on individual trust calculation

From Nokia SmartPhone 360 usage statistics [50], we can figure out one usage model that is periodically changed, e.g., mobile email usage. We use function $\left| {{ \sin }(\omega t)} \right|$, (ω = 1) to model it in our simulation with regard to usage frequency. The second usage model could be a logistic function, also known as Richards’ curve, which is widely used for growth modeling. We use a modified logistic function$\left( {1 - e^{ - \gamma t} } \right)/\left( {1 + e^{ - \gamma t} } \right)$, $(\gamma = 1/2)$in our simulation in order to make the growth start from 0 at t = 0. The third usage model is a growth curve at the beginning and then reducing to a stale level (including 0, which can be controlled by the function parameters). Herein, we use a $\Upgamma (\alpha ,\beta )$ distribution $t^{\alpha - 1} e^{ - \beta t} (\alpha = 2;\,\beta = 0.5)$ to model it. We also propose a linear increase model $\eta t(\eta = 0.1,\,\eta t < 1)$ to roughly model, for example, recommendation percentage, elapsed usage time, and the number of usages. The above usage models can be applied in usage time, the number of usage, frequency, or context index. The user-experienced feature ${\text{EF}}(i)/F(i)$ could be increased quickly and then gradually stay in a stable level. We use the logistic function to model it.

Figures 10a, b, and c show the simulation results of usage behavior formalization, reflection behavior formalization, and correlation behavior formalization, respectively. The usage models (or functions) applied in the simulations are listed in Table 3. For simplification, we apply function (2’) and (4) in our simulation. Figure 10d shows the aggregated trust value ($T_{i} (t)_{0} = 0. 5$) based on function (8) and the data of T(UB)_3; T(RB)_3; and T(CB)_1, T(CB)_2, and T(CB)_3 in Table 3, respectively.

Fig. 10

Individual trust value calculated based on usage behavior

Table 3 Usage models applied in simulations $(\eta = 0.1;\,\gamma = 1/2;\,\alpha = 2;\,\beta = 0.5)$

From the simulation, we can see that the individual trust value calculated based on the proposed formalization reflects usage change no mater it is periodically up and down or increased or decreased. It also implies the context’s influence on trust. The trust value contributed by the correlation trust behavior indicates the impact of application similarity and usage difference on trust. To uniform the result, we apply a sigmoid function to map final trust value into (0, 1). We can also use this function to map different part of trust contribution into (0, 1) and then aggregate them together. In this case, the general metric becomes:

$$ T_{i} (t) = f\left\{ {T_{i} (t)_{0} + \rho f\left\{ {T_{i} (t)_{\text{UB}} } \right\} + \vartheta f\left\{ {T_{i} (t)_{\text{RB}} } \right\} + \varsigma f\left\{ {T_{i} (t)_{\text{CB}} } \right\}} \right\}\quad (\rho + \vartheta + \varsigma = 1) $$

1.2 B. Random data generated for simulations in 6.5

The random data generated for simulations in Sect. 6.5 are provided in Tables 4 and 5.

Table 4 Simulated random data 1

Table 5 Simulated random data 2