| Recent years have experienced rapid development of mobile devices and mobile internet services. To facilitate the usage of devices and the access of services, the development of mobile apps has been greatly boosted. Such a trend raises an increasing challenge for users, as it becomes much more difficult for them to discover apps of interest in the huge number of candidates. To meet this challenge, the research and industry communities have focused on an new research direction, i.e. the recommendation of mobile apps(RMA). In RMA, they use the theories, technologies and methods of recommendation to meet the challenge of mobile app overload. The RMA is a newborn direction, in which research achievements are not abundant while systematic work is missing. For this reason, this paper carries out systematic and innovative research efforts for RMA, which are listed as below.1. The proposing and analysis of key technologies of RMA research. As RMA is in its initial development, this paper proposes a new research framework for RMA. That framework covers the whole life circle of mobile app recommendation and indicates the key technologies of RMA, i.e. the design and research of mobile app recommending algorithms, the research and development of mobile app recommender systems, the research and design of mobile app recommendation distribution, the design and research of mobile app recommending services. Moreover, this paper analyzes the specific characteristics of research elements in the direction of RMA, which defines the particularity of that direction. Based on the key technologies and the specific characteristics of RMA, this paper defines its main research efforts.2. The design and research of mobile app recommending algorithms. Existing mobile app recommending algorithms rely on the user experience of mobile apps, which are not able to provide recommendations of global optimization. Meanwhile, most existing recommending algorithms generate single-objective recommendations, which meets only limited app needs. To solve such problems, this paper firstly proposed a new recommending algorithm, namely RA-LSA, based on the method of Latent Semantic Analysis.RA-LSA computes the similarity among apps by mining out the latent semantics of app descriptions. Based on the app similarity, RA-LSA generates the most similar apps as recommendations, which are globally optimized. To understand RA-LSA, this paper carries out diversity measurements between RA-LSA and the existing recommending algorithm.Such measurements uncovers the pros and cons of both algorithms and suggests the design of hybrid algorithms. This paper thus designs a new hybrid algorithm for RMA, which leverages the concept of multi-objective and the method of rank aggregation, i.e., RAMRA. By adopting the concept of multi-objective, RA-MRA meets various app needs from diverse parties. This paper further collects the real data of 103348 apps and carries out evaluations. Results show that RA-MRA, which integrates the power of RA-LSA and the existing algorithm, effectively promotes the qualities of mobile app recommendations,with the increase from 67.2% to 494.1%.3. The research and development of mobile app recommender systems. Due to the lack of effective agile development framework, it is difficult for third-party system developers to deploy mobile app recommender systems, in the absence of large-scale data of apps and users. This leads to the slow development of mobile app recommender systems,which are behind the rapid growth of mobile apps. To solve such a problem, this paper develops a system development framework, namely, DF-MSC, leveraging the method of multi-system collaboration. DF-MSC not only integrates the power of different algorithms, but also expands the support data for developers. It thus lowers the threshold of system development. To support the implementation of DF-MSC, this paper further proposes a new algorithm of multi-system collaboration, i.e., CF-SPSO, using the method of Set based Particle Swarm Optimization. Such a algorithm generates optimal recommendations based on original recommendations of source recommender systems. For the evaluation, this paper gathers a data set of 108329 apps and implements various evaluating systems using both single-objective and multi-objective concepts. Results show that DFMSC and CF-SPSO are able to generates better recommendations(the average increase> 35.6%) with promising probability of problem solving(the average probability> 80%)and efficiency(the average iterations < 30).4. The research and design of mobile app recommendation distribution. Mobile app recommendations are distributed to users in the form of recommending links. Such links directly navigates the app browsing of users. Therefore the distribution of mobile app recommendations(D-MAR) is vital important to the discovery of mobile apps. However,existing research on the D-MAR is still missing. For this reason, this paper defines the mobile app recommending networks(MARNet) and models the D-MAR with MARNet.Afterwards, this paper collects real data of D-MAR and constructs two kinds of MARNets from the real data, i.e., the ARNet(App Relationship Network) and the UNNet(User Navigating Network). Through measurements of such MARNets, this paper reveals the shortage of existing D-MAR. Therefore, this paper further defines the mobile app discovery efficiency(MADE) of D-MAR and models the problem of improving the MADE of D-MAR. To solve the problem, this paper designs two reconstruction schemes of UNNet and evaluates them based on the data set of 103348 real apps. Results show that both reconstruction schemes promote the MADE of existing D-MAR, with the increase of 49.4%and 268.6%, respectively.5. The design and research of mobile app recommending services. The rapid growth of mobile apps leads to the increase of installed apps in mobile devices. Such a trend raises a new problem of information overload. Namely, it is much more difficult for users to find apps in the devices due to the limited space of device screens. To solve this problem, this paper proposes and designs the UMARS(Usage oriented Mobile App Recommending Services). UMARS monitors the usage of apps and the context of devices,through which it mines the MAUP(mobile apps usage patterns). Based on the MAUP,UMARS predicts the apps to be used by users and update the predictions on the device screens dynamically. Thus users need not cross screens to find apps, as the apps are by their hands. To implement UMARS, this paper defines three kinds of MAUP and accordingly designs mining methods for them. This paper further designs the system structure of UMARS and implements it. To carry out the evaluation, this paper calls for volunteers to use the devices with UMARS and gathers the logging data of their usage.Based on the 122547 records collected, evaluating results show that UMARS exhibits a promising hit-rate(>87.3%) of predictions. |
PDF Full Text Request