On personalized cloud service provisioning for mobile users using adaptive and context-aware service composition

The past few years have witnessed a growing trend in using mobile services for practical purposes in various contexts. People of all ages use their mobile devices to access various services such as shopping on-the-go, gaming and making travel arrangements, as well as retrieving educational resources, reading newspapers, making payments, monitoring health conditions, invoking corporate or partner services, and many more. As users increasingly rely on their mobile devices for carrying out these activities, new challenges and opportunities for mobile service providers emerged as they strive to improve and enrich further the proliferation of mobile experiences for consumers.

Smartphones have become primary personal information-processing tools. Mobile users aspire further at executing a wider range of applications on smartphones. Modern smartphones’ built-in sensors, such as GPS, accelerometer, proximity detector, ambient light sensor, and gyroscope are paving the way to advanced mobile applications. To facilitate these developments, mobile devices need further resources, in terms of storage and computing power, as well as enhanced software-provisioning services.

Mobile cloud computing (MCC) [1, 2] is poised to overcome substantially these limitations. MCC provides these resources from a remote computing platform such as Amazon AWS, Microsoft Azure, and Google AppEngine. As mobile devices could not process large-size applications locally, an alternative solution is to employ reusable services remotely from a cloud-hosting platform. A cloud service is typically the result of composing basic (in-house and third party) services on the fly to satisfy a user request. A personalized enhancement of this process includes an adaptation of the service presentation to the user interface and an adjustment of the service to the user context.

Personalization in cloud computing and related service provisioning processes are the emerging trends. The goal is to enable a dynamic customization of content and services that match mobile user preferences and current context. For example, a location-based service may recommend a specific store in the downtown of a city or in the duty free zone of an airport based on mobile users’ location, preferences, and previous purchases. Several definitions of personalization are available in the literature. Nevertheless, we adopt Riecken’s [3] and Hagen’s [4] definitions respectively:

Personalization is about building customer loyalty by building a meaningful one-to-one relationship; by understanding the needs of each individual and helping satisfy a goal that efficiently and knowledgeably addresses each individual’s need in a given context [3].

Personalization is the ability to provide content and services tailored to individuals based on knowledge about their preferences and behavior [4].

Personalization aims to improve mobile user experiences by adapting applications and services to meet individual preferences and contexts, which involves:

Current personalization and recommendation techniques in cloud environments are based on rule-based systems, content-based and collaborative filtering [7‐10]. However, solutions to mobile cloud service personalization expect services to adapt to the user’s functional needs, the device in-use, and the user’s context. Personalization in the context of MCC environments needs to answer the following questions: (a) How to adjust provisioned services to meet mobile user preferences and context? and (b) How to compose services to meet mobile users’ personalized needs?

In this work, we address service personalization in the context of a cloud service composition involving in-house and third-party services, while considering user preferences and context. In doing so, we propose a framework for adaptive personalization of mobile cloud service provisioning, that makes the following original contributions:

The novelty of this work lies in the fact that our proposed framework deals with the personalization of an adaptive dynamic composition of atomic services (in-house services and third-party services). The composition relies on the selection of appropriate basic services based on their QoS offerings. Also, the adaptation of the composed service to the current context of the user or her environment relies on the selection of suitable context services based on their QoC offerings. Furthermore, both the service composition and the adaptation process use a common description model for service offerings and context offerings respectively. These common description models allow to deal with the heterogeneity in the description of third-party services, and the heterogeneity in the description of context providers’ offerings.

We have reported a preliminary version of this work in a conference on cloud computing [11]. In this paper, we enhance the conference version by emphasizing the context dimension, which was overlooked in the conference paper. A sub-section describing the concepts of context services and quality of context (QoC) extends further the background section. We also extended the related work section by providing a critical literature review of personalization in cloud-service provisioning. We updated and completely redesigned the proposed framework to take into consideration the acquisition of context from provisioned context services. We extended the adaptation engine to wrap various context representations from third-party context service providers within a common semantic overlay. Finally, we added an algorithm for context service selection that is driven by QoC requirements of MCC providers.

The rest of this paper is organized as follows. The next section presents some background information and elaborates further on service personalization, context services’ platforms, and the concept of quality of context (QoC). Section 3 describes some related work, highlights the similarities and limitations of relevant works on cloud-based service personalization, and discusses the motivation of this work. Section 4 presents an overview of the proposed framework. Section 5 describes the adaptation structures for personalized service delivery. Section 6 describes our proposed algorithm for the selection of context services based on QoC requirements of MCC service provider. Section 7 presents the process of integrating and adapting third party service offerings based on mobile user profiles and device profiles as well as context adaptation. Section 8 reveals a scenario used to evaluate our framework where the results of selecting context providers based on QoC requirements of the MCC service provider are analyzed and discussed. Finally, Sect. 9 concludes the paper with a summary of the work presented in this paper, and highlights some future work.

There have been extensive works on personalizing Web experiences and cloud services. These works span a wide range of application environments and domain contexts. Next, we review some of the salient literature related to the investigation and evaluation of these recent developments.

In this section, we describe the process of selecting potential context services that can provide up-to-date contextual information about the user and her/his environment based on the cloud service provider’s QoC requirements. Service selection in general has been largely investigated in the area of service oriented architecture and Web services using different approaches and algorithms. These approaches include, but are not limited to:

In the following, we describe an algorithm capable of evaluating potential context service offers by taking into account QoC requirements of the cloud service provider and the various QoC service offerings.

Let \( {\text{Q}} = \left\{ {{\text{Q}}_{1} , {\text{Q}}_{2} , \ldots ,{\text{Q}}_{{\rm n}} } \right\} \) be the list of QoC attributes, such as freshness of context information, and its probability of correctness. These QoC attributes may be classified into two categories. In the first category, we find QoC benefit attributes, such as probability of correctness, which the cloud service provider wants to maximize. In the second category, we find QoC cost attributes, such as freshness of context information that the cloud service provider wants to minimize.

Let \( {\text{CS}} = \left\{ {{\text{CS}}_{1} ,{\text{CS}}_{2} , \ldots ,{\text{CS}}_{{\rm K}} } \right\} \) be the list of potential Context Services, which are capable of providing the type of context information requested by the cloud service provider. Let \( {\text{R}} = \left\{ {{\text{R}}_{1} , {\text{R}}_{2} , \ldots , {\text{R}}_{{\rm n}} } \right\} \), be the vector of quality requirements of the cloud service provider for each QoC attribute in Q.

The following vector expresses the QoC offering of the Context service \( {\text{CS}}_{{\rm m}} \):

The above QoC attributes are expressed in different units and cover a range of values. To be able to sort and rank the context services’ offerings with regard to their QoC offerings, we need to normalize values of QoC offers and define utility functions that will allow mapping the vector of QoC values into a single real value. We define the maximum and minimum values of QoC attribute Q_i considering all the offers of the potential context services in CS, namely:

The requirement of the cloud service provider R_i for Q_i may fall in the range [\( Q_{i}^{min} \), \( Q_{i}^{max} ] \) or outside of this range. To take into account the requirements of the cloud service provider during the normalization, we define the two following values:

We define \( q_{i}^{r} \), the normalized value for the context service \( CS_{m} \) offer with regard to the QoC attribute Q_i in Eq. (1) as:

Let \( RN = \left\{ {r_{1} , r_{2} , \ldots , r_{n} } \right\} \), with \( 0 \le r_{i} \le 1 \), be the vector of normalized values of the quality requirements of the cloud service provider.

We define the utility \( u_{i,m} \) as the level of satisfaction of the offering of context service \( CS_{m} \) with respect to the cloud service provider requirement for the quality attribute \( Qi \).

When \( u_{i,m} \) is greater than 1, it means that the offer of context service \( CS_{m} \) exceeds the expectation of the cloud service provider with regard to quality attribute \( Q_{i} \). When it is smaller than 1, it means that the offer does not meet its expectation for \( Q_{i} \).

The cloud service provider may assign an importance weight to each quality attribute. Let \( W = \left\{ {w_{1} , w_{2} , \ldots , w_{n} } \right\} \) be the weight vector associated with the quality attributes respectively. We define a Utility function (weighted combined level of satisfaction) associated with the offered context service \( CS_{m} \) as:

Equation (3) allows ranking QoC offers of potential context services that are capable of offering the context information required by the cloud service provider. The best context service offer corresponds to the offer that corresponds to the highest value of the utility function \( U_{m} \). Algorithm 1 summarizes the steps of the algorithm.

A similar algorithm, often referenced in literature, for comparing services based on the user QoS requirements is the TOPSIS algorithm developed by Hwang and Yoon [58]. The TOPSIS method aims to determine the best alternative from the concepts of the compromise solution. The solution that has farthest Euclidean distance from the negative ideal solution and the the shortest Euclidean distance from the ideal solution corresponds to the best compromise solution. The algorithm has six main steps [59]:

Section 8 describes a case scenario in which we apply our proposed algorithm and the TOPSIS algorithm to rank the offers of a numbers of context services.

In modern enterprise systems, applications are built from smaller services, described by service offering (SO) descriptions. SOs allow customizing services for each individual user or even for a whole organization. A single service offering may be used in diverse contexts for multiple purposes. In SO repositories, each SO has descriptive metadata allowing it to be easily found in a search process and integrated with other SOs to build much larger services. SOs metadata can be written in XML or in any other proprietary format. The OASIS WSDL Standard for Service Offering Metadata is an internationally recognized standard for describing services. Also, the Open Archives Initiative Protocol for Metadata Harvesting (OAI-PMH) [60] provides an application-independent interoperability framework based on metadata harvesting. A harvester is a client application that issues OAI-PMH requests. It is mainly operated by a service provider as a means of collecting metadata from data repositories.

As the goal of the cloud service provider is to offer an adaptive service, it should distinguish important contextual information for its business. The adapted content to the service consumer’s context may be for instance presenting information on new items, articles, promotions, and so on. As we mentioned earlier, context information may be obtained directly from the user’s mobile device, and/or from third party services that we referred to as context information services (CIOs).

With the heterogeneity in context offering descriptions and context information modeling, it is necessary to translate these offerings into a common model used by the adaptation engine [61]. ContextML [62], a light weight XML-based context representation schema, is a potential candidate for that common model. In ContextML, context information is categorized into scopes and may be related to different types of subjects, such as user and device. It may be incremented with attributes to describe various QoC indicators.

The proposed framework is open to third-party providers that can expose their service offerings as Web services. Their services can be plugged into the proposed framework. With the increasing support to the concept of open service offerings’ repositories (SORs), it will be possible to search for appropriate services that can be integrated with local services based on a mobile user’s profile and a composition plan.

The heterogeneity of the metadata used to describe SOs requires a process for adapting the service composition, built from various in-house and imported SOs, to the user’s profile, device profile, and context. The framework uses a common metadata model to describe SOs. This process, depicted in Fig. 5, involves the following steps:

We assume that the QoC requirement of the cloud service are: Freshness = 6 ms, Precision = 99.96%, and Probability of Correctness = 95.96%. Table 1 shows the QoC values offered by 24 hypothetical context service providers. Table 2 depicts the normalized values of each of these QoC attributes. Table 3 lists the weights that the MCC service provider assigns to these QoC parameters according to the mobile user requirements. By using our selection algorithm, normalized values are calculated for each QoC parameter while differentiating between benefit-driven QoC attributes and cost-driven QoC attributes. Then, using the weight table, the aggregate utility function is calculated for each context service offering.

Table 4 depicts the ranking table generated by MCC service using our proposed algorithm. The offer of the Cxt provider with ID 3 is the best offer as it maximizes the aggregated utility. In parallel with that, we also use the TOPSIS method [58] for ranking the Cxt offerings. First, we identify the positive ideal solution (PIS) and the negative ideal solution (NIS). Then, we calculate for each Cxt offering the separation from PIS and from NIS. Afterwards, we calculate the closeness to the ideal solution (CIS). Table 5 depicts the ranking using the TOPSIS method.

To our surprise, the above results show that both algorithms have obtained the same five best Cxt providers (Cxt ID: 3 is the best one). Subsequent Cxt offerings in the two rankings are very close, especially for the first ten best offerings. This process results in identifying a suitable Cxt provider for temperature provisioning by considering the three QoC attributes: freshness, precision, and probability of correctness.

The phenomenal proliferation of Internet-enabled mobile devices, such as tablets and smartphones, having multimedia and sensing capabilities, has triggered over the past few years the development of new applications and services of all kinds. These services include shopping-on-the-go, online hotel booking, accessing educational resources, and many more. However, these devices are still suffering from what is known as resource (computing and storage) poverty. Mobile cloud computing is a new paradigm that aims at addressing this shortcoming. Several research efforts studied the opportunities and challenges of this paradigm. Personalization of services, in the context of mobile cloud computing, remains a challenging issue due to the variety of mobile devices in the market and the heterogeneity across users’ profiles and preferences. The sensing capabilities of these devices add a new dimension, which is the user context, to the personalization challenge.

In this work, we proposed a conceptual framework that could serve as the basis for developing cloud services that can deliver personalized services. We described the different components to be deployed on the service provider side and the components to deploy on the mobile user’s device. The mobile cloud service provider typically composes its service from a set of in-house basic services and from third party services using a composition plan. We described how to achieve the adaptation of services by taking into account the user’s profile, the device profile, and the user and her/his environment context. The information structures used to support personalized service delivery include a service registry, the composition plan, the user model, and a service provisioning strategy. We also proposed an algorithm for the effective selection of suitable context services that meet the cloud service provider requirements in terms of context information and quality of context. Finally, we evaluated our framework in the context of an application scenario to illustrate the performance of the proposed algorithm. Our future work will further investigate a wide range of application contexts which integrate Internet of Things (IoT) services that potentially unleash further context patterns.