Historical Behavior Data Research Articles

Time-Aware CF and Temporal Association Rule-Based Personalized Hybrid Recommender System

Most recommender systems usually combine several recommendation methods to enhance the recommendation accuracy. Collaborative filtering (CF) is a best-known personalized recommendation technique. While temporal association rule-based recommendation algorithm can discover users' latent interests with time-specific leveraging historical behavior data without domain knowledge. The concept-drifting and user interest-drifting are two key problems affecting the recommendation performance. Aiming at the above problems, a time-aware CF and temporal association rule-based personalized hybrid recommender system, TP-HR, is proposed. The proposed time-aware CF algorithm considers evolving features of users' historical feedback. And time-aware users' similar neighbors selecting measure and time-aware item rating prediction function are proposed to keep track of the dynamics of users' preferences. The proposed temporal association rule-based recommendation algorithm considers the time context of users' historical behaviors when mining effective temporal association rules. Experimental results on real datasets show the feasibility and performance improvement of the proposed hybrid recommender system compared to other baseline approaches.

Mobile App Cross-Domain Recommendation with Multi-Graph Neural Network

With the rapid development of mobile app ecosystem, mobile apps have grown greatly popular. The explosive growth of apps makes it difficult for users to find apps that meet their interests. Therefore, it is necessary to recommend user with a personalized set of apps. However, one of the challenges is data sparsity, as users’ historical behavior data are usually insufficient. In fact, user’s behaviors from different domains in app store regarding the same apps are usually relevant. Therefore, we can alleviate the sparsity using complementary information from correlated domains. It is intuitive to model users’ behaviors using graph, and graph neural networks have shown the great power for representation learning. In this article, we propose a novel model, Deep Multi-Graph Embedding (DMGE), to learn cross-domain app embedding. Specifically, we first construct a multi-graph based on users’ behaviors from different domains, and then propose a multi-graph neural network to learn cross-domain app embedding. Particularly, we present an adaptive method to balance the weight of each domain and efficiently train the model. Finally, we achieve cross-domain app recommendation based on the learned app embedding. Extensive experiments on real-world datasets show that DMGE outperforms other state-of-art embedding methods.

ReMEMBeR: Ranking Metric Embedding-Based Multicontextual Behavior Profiling for Online Banking Fraud Detection

Anomaly detection relies on individuals' behavior profiling and works by detecting any deviation from the norm. When used for online banking fraud detection, however, it mainly suffers from three disadvantages. First, for an individual, the historical behavior data are often too limited to profile his/her behavior pattern. Second, due to the heterogeneous nature of transaction data, there lacks a uniform treatment of different kinds of attribute values, which becomes a potential barrier for model development and further usage. Third, the transaction data are highly skewed, and it becomes a challenge to utilize the label information effectively. The three disadvantages result in both poor generalization and high false positive rate of anomaly detection, and we propose a ranking metric embedding based multi-contextual behavior profiling (ReMEMBeR) model to battle them effectively. We solve the original fraud detection problem as a pseudo-recommender system problem, where an individual is treated as a pseudo-user, his/her behavior as a pseudo-item, and the label as the corresponding pseudo-rating. With the idea of collaborative filtering, for an individual, information from other similar individuals can be used to establish his/her behavior profile. In order to obtain a uniform treatment of heterogeneous attributes, we turn to an embedding based method to learn both attribute embedding and individuals' behavior profiles within a common latent space simultaneously. To utilize the label information better, our model is designed to fit pseudo-users' correct preference ranking for pseudo-items. By doing so, it explicitly learns to tell the fraudulent from the legitimate. Last but not least, we propose to identify and distinguish individuals under different contexts and further generalize the behavior profiling model to be a multi-contextual one. The proposed model can, thus, integrate the multi-contextual behavior patterns and allow transactions to be examined under the different contexts. Extensive experiments on a real-world online banking transaction dataset demonstrate that our model not only outperforms benchmarks on all metrics but also can be combined with them to achieve even better performance.

Development of Personalized Learning Resources Recommendation System Based on Knowledge Graph

This study focuses on how to efficiently and accurately recommend personalized learning resources for users when they are in the face of massive learning resources. A recommendation system is developed with software engineering methods. Knowledge graph technology is integrated in the system; curriculum knowledge graphs are constructed to solve the problems of semi-structured data storage and knowledge fragmentation. Two kinds of recommendation methods are adopted to achieve the goal of learners' personalized learning, one is Euclidean distance recommendation algorithm based on user behavior graph library, the other is learning mode recommendation algorithm and sequential mode recommendation algorithm based on user session library. The recommendation system maintains the interpret-ability and the accuracy of recommendation based on user historical behavior data; and realizes recommendation based on user session library in the context of lacking users’ historical data.

Dynamic Pricing Strategy of Electric Vehicle Aggregators Based on DDPG Reinforcement Learning Algorithm

The fixed service charge pricing model adopted by traditional electric vehicle aggregators (EVAs) is difficult to effectively guide the demand side resources to respond to the power market price signal. At the same time, real-time pricing strategy can flexibly reflect the situation of market supply and demand, shift the charging load of electric vehicles (EVs), reduce the negative impact of disorderly charging on the stable operation of power systems, and fully tap the economic potential of EVA participating in the power market. Based on the historical behavior data of EVs, this paper considers various market factors such as peak-valley time-of-use tariff, demand-side response mode and deviation balance of spot market to formulate the objective function of EVA comprehensive revenue maximization and establishes a quarter-hourly vehicle-to-grid (V2G) dynamic time-sharing pricing model based on deep deterministic policy gradient (DDPG) reinforcement learning algorithm. The EVA yield difference between peak-valley time-of-use tariff and hourly pricing strategy under the same algorithm is compared through the case studies. The results show that the scheme with higher pricing frequency can guide the charging behavior of users more effectively, tap the economic potential of power market to a greater extent, and calm the load fluctuation of power grid.

Application of sample balance-based multi-perspective feature ensemble learning for prediction of user purchasing behaviors on mobile wireless network platforms

With the rapid development of wireless communication network, M-Commerce has achieved great success. Users leave a lot of historical behavior data when shopping on the M-Commerce platform. Using these data to predict future purchasing behaviors of the users will be of great significance for improving user experience and realizing mutual benefit and win-win result between merchant and user. Therefore, a sample balance-based multi-perspective feature ensemble learning was proposed in this study as the solution to predicting user purchasing behaviors, so as to acquire user’s historical purchasing behavioral data with sample balance. Influence feature of user purchasing behaviors was extracted from three perspectives—user, commodity and interaction, in order to further enrich the feature dimensions. Meanwhile, feature selection was carried out using XGBSFS algorithm. Large-scale real datasets were experimented on Alibaba M-Commerce platform. The experimental results show that the proposed method has achieved better prediction effect in various evaluation indexes such as precision and recall rate.

On recommendation-aware content caching for 6G: An artificial intelligence and optimization empowered paradigm

Recommendation-aware Content Caching (RCC) at the edge enables a significant reduction of the network latency and the backhaul load, thereby invigorating ubiquitous latency-sensitive innovative services. However, the effectiveness of RCC strategies is highly dependent on explicit information as regards subscribers’ content request patterns, the sophisticated caching placement policy, and the personalized recommendation tactics. In this article, we investigate how the potentials of Artificial Intelligence (AI) and optimization techniques can be harnessed to address those core issues and facilitate the full implementation of RCC for the upcoming intelligent 6G era. Towards this end, we first elaborate on the hierarchical RCC network architecture. Then, the devised AI and optimization empowered paradigm is introduced, whereas AI and optimization techniques are leveraged to predict the users’ content preferences in real-time situations with the assistance of their historical behavior data and determine the cache pushing and recommendation decision, respectively. Through extensive case studies, we validate the effectiveness of AI-based predictors in estimating users’ content preference and the superiority of optimized RCC policies over the conventional benchmarks. At last, we shed light on the opportunities and challenges in the future.

Social Recommendation Combining Trust Relationship and Distance Metric Factorization

The recommender system predicts user preferences by mining user historical behavior data. This paper proposes a social recommendation combining trust relationship and distance metric factorization. On the one hand, the recommender system has a cold start problem, which can be effectively alleviated by adding social relations. Simultaneously, to improve the problem of sparse trust matrix, we use the Jaccard similarity coefficient and the Dijkstra algorithm to reconstruct the trust matrix and explore the potential user trust relationship. On the other hand, the traditional matrix factorization algorithm is modeled by the user item potential factor dot product, however, it does not satisfy the triangle inequality property and affects the final recommender effect. The primary motivator behind our approach is to combine the best of both worlds, mitigate the inherent weaknesses of each paradigm. Combining the advantages of the two ideas, it has been demonstrated that our algorithm can enhance recommender performance and improve cold start in recommender systems.

Automatic Recommendation Method of Network Data Based on Big Data Technology

With the advent of the Internet era in modern society, the surge in network data has greatly increased the difficulty for users to obtain demand information. In order to provide customers with information that may be of interest to them, this paper proposes a belief collaborative recommendation optimization algorithm, which introduces DP synthesis rules and Smets synthesis rules to improve traditional DS synthesis rules and establishes network data recommendation models. This method realizes automatic recommendation of network data according to the self-property and historical behavior data of network customers. The experimental results show that the automatic recommendation method for network data proposed in this paper has higher recommendation accuracy.

Virtual-Taobao: Virtualizing Real-World Online Retail Environment for Reinforcement Learning

Applying reinforcement learning in physical-world tasks is extremely challenging. It is commonly infeasible to sample a large number of trials, as required by current reinforcement learning methods, in a physical environment. This paper reports our project on using reinforcement learning for better commodity search in Taobao, one of the largest online retail platforms and meanwhile a physical environment with a high sampling cost. Instead of training reinforcement learning in Taobao directly, we present our environment-building approach: we build Virtual-Taobao, a simulator learned from historical customer behavior data, and then we train policies in Virtual-Taobao with no physical sampling costs. To improve the simulation precision, we propose GAN-SD (GAN for Simulating Distributions) for customer feature generation with better matched distribution; we propose MAIL (Multiagent Adversarial Imitation Learning) for generating better generalizable customer actions. To further avoid overfitting the imperfection of the simulator, we propose ANC (Action Norm Constraint) strategy to regularize the policy model. In experiments, Virtual-Taobao is trained from hundreds of millions of real Taobao customers’ records. Compared with the real Taobao, Virtual-Taobao faithfully recovers important properties of the real environment. We further show that the policies trained purely in Virtual-Taobao, which has zero physical sampling cost, can have significantly superior real-world performance to the traditional supervised approaches, through online A/B tests. We hope this work may shed some light on applying reinforcement learning in complex physical environments.

Viewport Prediction for Live 360-Degree Mobile Video Streaming Using User-Content Hybrid Motion Tracking

360-degree video streaming has been gaining popularities recently with the rapid growth of adopting mobile head mounted display (HMD) devices in the consumer video market, especially for live broadcasts. The 360-degree video streaming introduces brand new bandwidth and latency challenges in live streaming due to the significantly increased video data. However, most of the existing bandwidth saving approaches based on viewport prediction have only focused on the video-on-demand (VOD) use cases leveraging historical user behavior data, which is not available in live broadcasts. We develop a new viewport prediction scheme for live 360-degree video streaming using video content-based motion tracking and dynamic user interest modeling. To obtain real-time performance, we implement the Gaussian mixture model (GMM) and optical flow algorithms for motion detection and feature tracking. Then, the user's future viewport of interest is generated by leveraging a dynamic user interest model that weighs all the features and motion information abstracted from the live video frames. Furthermore, we develop two enhancement techniques that take into consideration of user feedback for fast error recovery and view updates. Consequently, our predicted viewports are irregular and dynamically adjusted to cover the maximum portions of the actual user viewports and thus ensure a high prediction accuracy. We evaluate our viewport prediction approach using a public user head movement dataset, which contains the data of 48 users watching 6 360-degree videos. The experimental results show that the proposed approach supports sophisticated user head movement patterns and outperforms the existing velocity-based approach in terms of prediction accuracy. In addition, the motion tracking scheme introduces minimum latency overhead to ensure the quality of live streaming experience.

Uncovering cooperative behaviors with sparse historical behavior data in the spatial games

For past decades, the main attention of the evolutionary games has been focused on cooperation mechanism with the assumption that the strategy information of all players are known. However, it is difficult for observers to obtain the global information of players’ strategies in the real world, and some players even hide their strategy information to confuse their opponents. Here we try to solve the problem to predicate the hidden strategies with sparse historical behavior data in the evolutionary games. To quantify the similarity of strategies among the players in our method, the Euclidean distance of players is defined from the strategies of the players in the few past rounds. Then, the hidden strategy of a player will be determined from the tendency that players with minimum Euclidean distance will adopt similar strategies. The method has good performance on determining hidden strategy of human beings in both the prisoner’s dilemma game and the public goods game where strategies of twenty five percent players are hidden, and the success rate to determine hidden strategy reaches up to 0.9. It is also found that the success rate to determine hidden strategy depends on both length of historical behavior data and tempting payoff b (the prisoner’s dilemma game) or multiple factor r (the public goods game).

Communication’s Role and Technology Preferences during Hurricane Evacuations

Emergency evacuations are dynamic, time-critical events involving complex coordination decisions. To develop robust evacuation policies, it is important to understand how people respond to evacuation alerts, including choices of time to leave and routes to take. The purpose of this study was to investigate the preferred communication sources and modalities for evacuees to use and learn about changes in hurricane scenarios. A hurricane evacuation behavior survey was administrated to residents in the greater Charleston, South Carolina, area. In this study, two hurricane scenarios are examined to understand prehurricane evacuation decisions and communication preferences of residents. Results from the survey indicate that while driving, respondents prefer using radios to gather traffic and weather information. When at home, television is the most preferred modality among respondents to determine weather status and to find out about hurricane evacuation notices. Concerning evacuation behavior, respondents prefer leaving in the daytime rather than at night. To account for this preference, evacuation announcements and notifications should be broadcast early enough to allow for more daytime evacuations. This may result in a longer lead time if the announcement would have otherwise occurred at night. The evacuation timing decisions from the survey are then compared with historical behavior data from a retrospective Hurricane Hugo survey. The respondents indicate a tendency to evacuate earlier than what was shown in the Hurricane Hugo survey. This may imply that it takes longer to actually prepare to leave than respondents to the current survey realize. Alternatively, respondents to the Hurricane Hugo survey may pay more attention to weather patterns as well as wait for orders to be provided by the state rather than make a proactive decision to leave.