App Usage Prediction Research Articles

App usage on-the-move: Context- and commute-aware next app prediction

The proliferation of digital devices and connectivity enables people to work anywhere, anytime, even while they are on the move. While mobile applications have become pervasive, an excessive amount of mobile applications have been installed on mobile devices. Nowadays, commuting takes a large proportion of daily human life, but studies show that searching for the desired apps while commuting can decrease productivity significantly and sometimes even cause safety issues. Although app usage behaviour has been studied for general situations, little to no study considers the commuting context as vital information. Existing models for app usage prediction cannot be easily generalised across all commuting contexts due to: (1) continuous change in user locations; and (2) limitation of necessary contextual information (i.e., lack of knowledge to identify which contextual information is necessary for different commuting situations. We aim to address these challenges by extracting essential contextual information for on-commute app usage prediction. Using the extracted features, we propose AppUsageOTM, a practical statistical machine learning framework to predict both destination amenity and utilise the inferred destination to contextualise the app usage prediction with travelling purposes as crucial information. We evaluate our framework in terms of accuracy, which shows the feasibility of our work. Using a real-world mobile and app usage behaviour dataset with more than 12,495 trajectory records and more than 1046 mobile applications logged, AppUsageOTM significantly outperformed all baseline models, achieving Accuracy@k 46.4%@1, 66.4%@5, and 75.9%@10.

App2Vec: Context-Aware Application Usage Prediction

Both app developers and service providers have strong motivations to understand when and where certain apps are used by users. However, it has been a challenging problem due to the highly skewed and noisy app usage data. Moreover, apps are regarded as independent items in existing studies, which fail to capture the hidden semantics in app usage traces. In this article, we propose App2Vec, a powerful representation learning model to learn the semantic embedding of apps with the consideration of spatio-temporal context. Based on the obtained semantic embeddings, we develop a probabilistic model based on the Bayesian mixture model and Dirichlet process to capture when , where , and what semantics of apps are used to predict the future usage. We evaluate our model using two different app usage datasets, which involve over 1.7 million users and 2,000+ apps. Evaluation results show that our proposed App2Vec algorithm outperforms the state-of-the-art algorithms in app usage prediction with a performance gap of over 17.0%.

DeepApp

Smartphone mobile application (App) usage prediction, i.e., which Apps will be used next, is beneficial for user experience improvement. Through an in-depth analysis on a real-world dataset, we find that App usage is highly spatio-temporally correlated and personalized. Given the ability to model complex spatio-temporal contexts, we aim to apply deep learning to achieve high prediction accuracy. However, the personalization yields a problem: training one network for each individual suffers from data scarcity, yet training one deep neural network for all users often fails to uncover user preference. In this article, we propose a novel App usage prediction framework, named DeepApp , to achieve context-aware prediction via multi-task learning. To tackle the challenge of data scarcity, we train one general network for multiple users to share common patterns. To better utilize the spatio-temporal contexts, we supplement a location prediction task in the multi-task learning framework to learn spatio-temporal relations. As for the personalization, we add a user identification task to capture user preference. We evaluate DeepApp on the large-scale dataset by extensive experiments. Results demonstrate that DeepApp outperforms the start-of-the-art baseline by 6.44%.

Predicting and Recommending the next Smartphone Apps based on Recurrent Neural Network

The popularity of smartphones has witnessed the rapid growth of the number of mobile applications. Nowadays, there are millions of applications available, and at the same time, many applications are already installed on people’s smartphones. Installing numerous apps will cause some troubles in finding the specific apps promptly. Hence it is necessary to predict the next app(s) to be used in a short term and launching them as shortcuts, which will make the smartphone system more efficient and user-friendly. In this paper, we pay attention to two subproblems that are related to the app usage prediction. One is the $$\varDelta T$$ app prediction problem that focuses on predicting a set of apps that will be used in a time interval. The other is the Top-K app recommendation problem that focuses on recommending the K most probable APPs to be used next. In order to solve these problems, we propose a generic prediction model based on Long Short-term Memory (LSTM), which is an enhancement of the recurrent neural network (RNN) model. The proposed model converts the temporal-sequence dependency and contextual information into a unified feature representation for next app prediction. We implement the model in the Android platform. Extensive experiments based on real collected dataset demonstrate that the proposed LSTM model outperforms the baselines for app usage prediction, and achieves high accuracy for app recommendation.

Serendipity adjustable application recommendation via joint disentangled recurrent variational auto-encoder

• Mobile app recommender able to adjust the level of serendipity is proposed. • It generates next app usage prediction based on the disentangled representation. • We vary the attributes of the disentanglement for serendipitous recommender. • It shows better results than other benchmark methods in the experiments. In this study, we propose an advanced mobile application recommendation method that can systematically adjust the level of serendipity. Previous studies focused on generating recommendations that are relevant to users’ interests. However, this leads to a lack of serendipity and novelty for the user and is therefore not sufficient to achieve high levels of user satisfaction. Our model is trained to predict the next application to be recommended based on a disentangled representation of the sequence in which applications were previously used. We introduce variety into our model by varying the attributes of the disentangled representation. The results of our experiments indicate that the proposed method can systematically and effectively adjust the level of serendipity by varying the values and dimensions of the attribute variables in the disentangled representation. Additionally, our method produces recommendations that are superior to those of other benchmark methods in terms of serendipity and relevance.

Semantic-aware Spatio-temporal App Usage Representation via Graph Convolutional Network

Recent years have witnessed a rapid proliferation of personalized mobile Apps, which poses a pressing need for user experience improvement. A promising solution is to model App usage by learning semantic-aware App usage representations which can capture the relation among time, locations and Apps. However, it is non-trivial due to the complexity, dynamics, and heterogeneity characteristics of App usage. To smooth over these obstacles and achieve the goal, we propose SA-GCN, a novel representation learning model to map Apps, location, and time units into dense embedding vectors considering spatio-temporal characteristics and unit properties simultaneously. To handle complexity and dynamics, we build an App usage graph by regarding App, time, and location units as nodes and their co-occurrence relations as edges. For heterogeneity, we develop a Graph Convolutional Network with meta path-based objective function to combine the structure of the graph and the attribute of units into the semantic-aware representations. We evaluate the performance of SA-GCN via a large-scale real-world dataset. In-depth analysis shows that SA-GCN characterizes the complex relationships among different units and recover meaningful spatio-temporal patterns. Moreover, we make use of the learned representations in App usage prediction task without post-training and achieve 8.3% of the performance gain compared with state-of-the-art baselines.

Graph-based Method for App Usage Prediction with Attributed Heterogeneous Network Embedding

Smartphones and applications have become widespread more and more. Thus, using the hardware and software of users’ mobile phones, we can get a large amount of personal data, in which a large part is about the user’s application usage patterns. By transforming and extracting these data, we can get user preferences, and provide personalized services and improve the experience for users. In a detailed way, studying application usage pattern benefits a variety of advantages such as precise bandwidth allocation, App launch acceleration, etc. However, the first thing to achieve the above advantages is to predict the next application accurately. In this paper, we propose AHNEAP, a novel network embedding based framework for predicting the next App to be used by characterizing the context information before one specific App being launched. AHNEAP transforms the historical App usage records in physical spaces to a large attributed heterogeneous network which contains three node types, three edges, and several attributes like App type, the day of the week. Then, the representation learning process is conducted. Finally, the App usage prediction problem was defined as a link prediction problem, realized by a simple neural network. Experiments on the LiveLab project dataset demonstrate the effectiveness of our framework which outperforms the three baseline methods for each tested user.

App usage prediction for dual display device via two-phase sequence modeling

There have been many recent studies dealing with issues regarding app usage prediction. A number of smartphone manufacturers have recently released flexible dual display devices for multitasking purposes. In spite of many studies achieving higher levels of prediction accuracy, owing to certain practical limitations, they have not been effectively applied to real dual display devices. First, they trained their prediction model using only existing data collected from a single display device. Thus, they are unable to reflect the multitasking usage of dual display devices in their model prediction. In addition, previous studies have mostly been based on a Markov assumption, which needs to calculate the transition probability for all pairs of apps, which is difficult to calculate for newly installed apps. To address these limitations, we initially trained our prediction model using data collected from a single display during the first phase, and re-trained the prediction model using data collected from the prototype of a dual display device, which we developed along with LG Electronics, during the second phase. In addition, we applied an app prediction user interface to the prototype. Next, we applied a stacked long–short term memory architecture, a type of sequential deep learning architecture, to train the prediction model without calculating the transition probability. The experiment results show that our proposed method achieves an effective app usage prediction for dual display devices in a qualitative way, and outperforms all other approaches in quantitative experiments with top-3 and top-5 level accuracies. The reason for the higher performance of the proposed method is that it captures the real usage of a dual display and reflects multitasking use in its app usage prediction. In addition, we also verified the effects of a few variables such as the location, time, and two-phase training.

CAP

Context-aware mobile application (App) usage prediction benefits a variety of applications such as precise bandwidth allocation, App launch acceleration, etc. Prior works have explored this topic through individual data profiles and contextual information. However, it is still a challenging problem because of the following three aspects: i. App usage behavior is usually influenced by multiple factors, especially temporal and spatial factors. ii. It is difficult to describe individuals' preferences, which are usually time-variant. iii. A single user's data is sparse on the spatial domain and only covers a limited number of locations. Prediction becomes more difficult when the user appears at a new location. This paper presents CAP, a context-aware App usage prediction algorithm that takes both contextual information (location & time) and attribution (App with type information) into consideration. We find that the relationships between App-location, App-time, and App-App type are essential to prediction and propose a heterogeneous graph embedding algorithm to map them into the common comparable latent space. In addition, we create a user profile for each user with App usage and trajectory history to describe the individual dynamic preference for personalized prediction. We evaluate the performance of our proposed CAP with two large-scale real-world datasets. Extensive evaluations demonstrate that CAP achieves 30% higher accuracy than a state-of-the-art method Personalized Ranking Metric Embedding (PRME) in terms of Accuracy@5. In terms of mean reciprocal rank (MRR), CAP achieves 1.5× higher than the straightforward baseline Sta and 2× higher than PRME. Our investigation enables a range of applications to benefit from such timely predictions, including network operators, service providers, and etc.

The application of optimization algorithm based on incremental learning in app usage prediction

The application of optimization algorithm based on incremental learning in app usage prediction

A recommender system to address the Cold Start problem for App usage prediction

The Cold Start Recommender System (RS) for App usage prediction on mobile phones is important for improving new user experience on mobile operating systems. At present, the existing Cold Start RS computes the probability of App launching mainly by mining the potential information of new users and similar users (i.e., collaborative filtering algorithm CF). But for newly installed Apps, the default CF does not have any useful information for App Cold Start recommendation, resulting in the data sparseness. To tackle the problems, we think that App usage periodicity installed by the new user is regularly followed under different conditions. It not only compensates for the data sparseness of similar users but also increases the predictive flexibility when the user’s environment changes. Therefore, we designed Predictor, an efficient dynamic CF fusion algorithm that provides App Cold Start prediction for new users on mobile devices. It dynamically combines both App preferences of similar users (user-based CF) and App usage periodicity (item-based CF) through the conditional combination. Compared to other traditional methods, Predictor proposes more appropriate App launching recommendation and matches the launching expectations of most users.

Cluster-Based Smartphone Predictive Analytics for Application Usage and Next Location Prediction

Prediction of app usage and location of smartphone users is an interesting problem and active area of research. Several smartphone sensors such as GPS, accelerometer, gyroscope, microphone, camera and Bluetooth make it easier to capture user behavior data and use it for appropriate analysis. However, differences in user behavior and increasing number of apps have made such prediction a challenging problem. In this article, a prediction approach that takes smartphone user behavior into consideration is proposed. The proposed approach is illustrated using data from over 30000 users from a leading IT company in China by first converting data in to recency, frequency, and monetary variables and then performing cluster analysis to capture user behavior. Prediction models are then developed for each cluster using a training dataset and their performance is assessed using a test dataset. The study involves ten different categories of apps and four different regions in Beijing. The proposed app usage prediction and next location prediction approach has provided interesting results.

Positive and Unlabeled Learning for User Behavior Analysis Based on Mobile Internet Traffic Data

With the rapid development of wireless communication and mobile Internet, mobile phone becomes ubiquitous and functions as a versatile and a smart system, on which people frequently interact with various mobile applications (Apps). Understanding human behaviors using mobile phone is significant for mobile system developers, for human-centered system optimization and better service provisioning. In this paper, we focus on mobile user behavior analysis and prediction based on mobile Internet traffic data. Real traffic flow data is collected from the public network of Internet service providers, by high-performance network traffic monitors. We construct a User-App bipartite network to represent the traffic interaction pattern between users and App servers. After mining the explicit and implicit features from the User-App bipartite network, we propose two positive and unlabeled (PU) learning methods, including Spy-based PU learning and K-means-based PU learning, for App usage prediction and mobile video traffic identification. We first use the traffic flow data of QQ, a very famous messaging and social media application possessing high market share in China, as the experimental data set for App usage prediction task. Then, we use the traffic flow data from six popular Apps, including video intensive Apps (Youku, Baofeng, LeTV, and Tudou) and other Apps (Meituan and Apple), as the experimental data set for mobile video traffic identification task. Experimental results show that our proposed PU learning methods perform well in both tasks.

Matching Mobile Applications for Cross Promotion

The mobile app market is one of the most successful software markets. As the platform grows rapidly, with millions of apps and billions of users, search costs are increasing tremendously. The challenge is how app developers can target the right users with their apps and how consumers can find apps that fit their needs. Cross promotion, advertising a mobile app (target app) in another app (source app), is introduced as a new app promotion framework to alleviate the issue of search costs. In this paper, we model source app user behaviors (downloads and post-download usage) with respect to different target apps in cross-promotion campaigns. We construct a novel app similarity measure using LDA topic modeling on apps’ production descriptions, and then analyze how the similarity between the source and target apps influences users’ app download and usage decisions. To estimate the model, we use a unique data set from a large-scale random matching experiment conducted by a major mobile advertising company in Korea. The empirical results show that consumers prefer more diversified apps when they are making download decisions compared with their usage decisions, which is supported by the psychology literature on people’s variety-seeking behavior. Lastly, we propose an app-matching system based on machine learning models (on app download and usage prediction) and generalized deferred acceptance algorithms. The simulation results show that app analytics capability is essential in building accurate prediction models and in increasing ad effectiveness of cross promotion campaigns, and that, at the expense of privacy, individual user data can further improve the matching performance. The paper has implications on the tradeoff between utility and privacy in the growing mobile economy.