In online video platforms, accurate watch time prediction has become a fundamental and challenging problem in video recommendation. Previous research has revealed that the accuracy of watch time prediction highly depends on both the transformation of watch-time labels and the decomposition of the estimation process. TPM (Tree based Progressive Regression Model) achieves State-of-the-Art performance with a carefully designed and effective decomposition paradigm. TPM discretizes the watch time into several ordinal intervals and organizes them into a binary decision tree, where each node corresponds to a specific interval. At each non-leaf node, a binary classifier is used to determine the specific interval in which the watch time variable most likely falls, based on the prediction outcome at its parent node.

Online advertising serves as the primary revenue source for numerous Internet companies, which typically sell advertising slots through auctions. Conventional online ad auctions assume constant click-through rates (CTRs) and conversion rates (CVRs) for ads during the auction process. However, this paper studies a new scenario where advertisers can offer coupons to users, thereby influencing both CTRs and CVRs and consequently, the platform's revenue.

An accurate prediction of watch time has been of vital importance to enhance user engagement in video recommender systems. To achieve this, there are four properties that a watch time prediction framework should satisfy: first, despite its continuous value, watch time is also an ordinal variable and the relative ordering between its values reflects the differences in user preferences. Therefore the ordinal relations should be reflected in watch time predictions. Second, the conditional dependence between the video-watching behaviors should be captured in the model. For instance, one has to watch half of the video before he/she finishes watching the whole video. Third, modeling watch time with a point estimation ignores the fact that models might give results with high uncertainty and this could cause bad cases in recommender systems. Therefore the framework should be aware of prediction uncertainty. Forth, the real-life recommender systems suffer from severe bias amplifications thus an estimation without bias amplification is expected.

Modeling users' dynamic preferences from their historical behaviors is challenging and crucial for recommendation systems. Previous methods employ sequential neural networks to encode users' historical interactions from left to right into hidden representations for making recommendations. Despite their effectiveness, we argue that such left-to-right unidirectional models are sub-optimal due to the limitations including: \begin enumerate* [label=series\itshape\alph*\upshape)] \item unidirectional architectures restrict the power of hidden representation in users' behavior sequences; \item they often assume a rigidly ordered sequence which is not always practical. \end enumerate* To address these limitations, we proposed a sequential recommendation model called BERT4Rec, which employs the deep bidirectional self-attention to model user behavior sequences. To avoid the information leakage and efficiently train the bidirectional model, we adopt the Cloze objective to sequential recommendation, predicting the random masked items in the sequence by jointly conditioning on their left and right context. In this way, we learn a bidirectional representation model to make recommendations by allowing each item in user historical behaviors to fuse information from both left and right sides. Extensive experiments on four benchmark datasets show that our model outperforms various state-of-the-art sequential models consistently.

Recommendation with multiple objectives is an important but difficult problem, where the coherent difficulty lies in the possible conflicts between objectives. In this case, multi-objective optimization is expected to be Pareto efficient, where no single objective can be further improved without hurting the others. However existing approaches to Pareto efficient multi-objective recommendation still lack good theoretical guarantees.

Ranking is a core task in recommender systems, which aims at providing an ordered list of items to users. Typically, a ranking function is learned from the labeled dataset to optimize the global performance, which produces a ranking score for each individual item. However, it may be sub-optimal because the scoring function applies to each item individually and does not explicitly consider the mutual influence between items, as well as the differences of users' preferences or intents. Therefore, we propose a personalized re-ranking model for recommender systems. The proposed re-ranking model can be easily deployed as a follow-up modular after any ranking algorithm, by directly using the existing ranking feature vectors. It directly optimizes the whole recommendation list by employing a transformer structure to efficiently encode the information of all items in the list. Specifically, the Transformer applies a self-attention mechanism that directly models the global relationships between any pair of items in the whole list. We confirm that the performance can be further improved by introducing pre-trained embedding to learn personalized encoding functions for different users. Experimental results on both offline benchmarks and real-world online e-commerce systems demonstrate the significant improvements of the proposed re-ranking model.

Existing recommendation algorithms mostly focus on optimizing traditional recommendation measures, such as the accuracy of rating prediction in terms of RMSE or the quality of top-k recommendation lists in terms of precision, recall, MAP, etc. However, an important expectation for commercial recommendation systems is to improve the final revenue/profit of the system. Traditional recommendation targets such as rating prediction and top-k recommendation are not directly related to this goal.

Recommender systems aim to capture user preferences and provide accurate recommendations to users accordingly. For each user, there usually exist others with similar preferences, and a collection of users may also have similar preferences with each other, thus forming a community. However, such communities may not necessarily be explicitly given, and the users inside the same communities may not know each other; they are formally defined and named Implicit Preference Communities (IPCs) in this article. By enriching user preferences with the information of other users in the communities, the performance of recommender systems can also be enhanced. Historical explicit ratings are a good resource to construct the IPCs of users but is usually sparse. Meanwhile, user preferences are easily affected by their social connections, which can be jointly used for IPC modeling with the ratings. However, this imposes two challenges for model design. First, the rating and social domains are heterogeneous; thus, it is challenging to coordinate social information and rating behaviors for a same learning task. Therefore, transfer learning is a good strategy for IPC modeling. Second, the communities are not explicitly labeled, and existing supervised learning approaches do not fit the requirement of IPC modeling. As co-clustering is an effective unsupervised learning approach for discovering block structures in high-dimensional data, it is a cornerstone for discovering the structure of IPCs. In this article, we propose a recommendation model with Implicit Preference Communities from user ratings and social connections. To tackle the unsupervised learning limitation, we design a Bayesian probabilistic graphical model to capture the IPC structure for recommendation. Meanwhile, following the spirit of transfer learning, both rating behaviors and social connections are introduced into the model by parameter sharing. Moreover, Gibbs sampling-based algorithms are proposed for parameter inferences of the models. Furthermore, to meet the need for online scenarios when the data arrive sequentially as a stream, a novel online sampling-based parameter inference algorithm for recommendation is proposed. To the best of our knowledge, this is the first attempt to propose and formally define the concept of IPC.

Software defined networks (SDNs) have emerged as a promising paradigm by separating control plane from data plane. In this paper, we study the problem of achieving realtime quality of services in a SDN for both centralized control plane and distributed control plane. In the first place, we propose a system prototype that handles real-time end to end services. Then, we formulate the problem for a centralized control plane as a linear programming and apply prime-dual technique to derive an efficient algorithm. We also discuss about the method to achieve load balancing between link resources and node resources. Finally, we extend the algorithm to a distributed control plane and present efficient algorithms to achieve good competitive ratio.

Factorization Machine (FM) is a supervised learning approach with a powerful capability of feature engineering. It yields state-of-the-art performances in various batch learning tasks where all the training data is made available prior to the training. However, in real-world applications where the data arrives sequentially in a streaming manner, the high cost of re-training with batch learning algorithms has posed formidable challenges in the online learning scenario. The initial challenge is that no prior formulations of FM could directly fulfill the requirements in Online Convex Optimization (OCO) -- the paramount framework for online learning algorithm design. To address this aforementioned challenge, we invent a new convexification scheme leading to a Compact Convexified FM (CCFM) that seamlessly meets the requirements in OCO. However for learning Compact Convexified FM (CCFM) in the online learning settings, most existing algorithms suffer from expensive projection operations. To address this subsequent challenge, we follow the general projection-free algorithmic framework of Online Conditional Gradient and propose an Online Compact Convex Factorization Machine (OCCFM) algorithm that eschews the projection operation with efficient linear optimization steps. In support of the proposed OCCFM in terms of its theoretical foundation, we prove that the developed algorithm achieves a sub-linear regret bound. To evaluate the empirical performance of OCCFM, we conduct extensive experiments on 6 real-world datasets for online regression and online classification tasks. The experimental results show that OCCFM outperforms the state-of-art online learning methods for FM.

User feedback in the form of movie-watching history, item ratings, or product consumption is very helpful in training recommender systems. However, relatively few interactions between items and users can be observed. Instances of missing user--item entries are caused by the user not seeing the item (although the actual preference to the item could still be positive) or the user seeing the item but not liking it. Separating these two cases enables missing interactions to be modeled with finer granularity, and thus reflects user preferences more accurately. However, most previous studies on the modeling of missing instances have not fully considered the case where the user has not seen the item. Social connections are known to be helpful for modeling users' potential preferences more extensively, although a similar visibility problem exists in accurately identifying social relationships. That is, when two users are unaware of each other's existence, they have no opportunity to connect. In this paper, we propose a novel user preference model for recommender systems that considers the visibility of both items and social relationships. Furthermore, the two kinds of information are coordinated in a unified model inspired by the idea of transfer learning. Extensive experiments have been conducted on three real-world datasets in comparison with five state-of-the-art approaches. The encouraging performance of the proposed system verifies the effectiveness of social knowledge transfer and the modeling of both item and social visibilities.

In constructing the cognitive radio networks, a fundamental process is to establish a communication link on a same channel for every two users, which is referred to as rendezvous. Most studies focus on minimizing the time to rendezvous after they start the process synchronously or asynchronously. However, to the best of our knowledge, no work has considered the possibility that the users may achieve different utilities when they rendezvous on different channels for communication. The utility originates from the quality of the specific channel that two users rendezvous on, and it is influenced by the channel's bandwidth, transmission rate, stability, etc. In this paper, we formally formulate the problem of maximizing rendezvous utilities and propose a novel method by extending the channels to promote the users to achieve higher rendezvous utilities. We propose channel extension algorithms for both symmetric and asymmetric rendezvous scenarios, where the users may have the same or different sets of rendezvous channels respectively. These algorithms are built on the construction of Disjoint Relaxed Difference Set (DRDS) in [5], and we show the efficiency of achieving rendezvous on the channel with high utility theoretically. Moreover, we conduct thorough simulations to evaluate our algorithms and the results also corroborate our theoretical analyses.

Group recommendation has attracted significant research efforts for its importance in benefiting a group of users. This paper investigates the Group Recommendation problem from a novel aspect, which tries to maximize the satisfaction of each group member while minimizing the unfairness between them. In this work, we present several semantics of the individual utility and propose two concepts of social welfare and fairness for modeling the overall utilities and the balance between group members. We formulate the problem as a multiple objective optimization problem and show that it is NP-Hard in different semantics. Given the multiple-objective nature of fairness-aware group recommendation problem, we provide an optimization framework for fairness-aware group recommendation from the perspective of Pareto Efficiency. We conduct extensive experiments on real-world datasets and evaluate our algorithm in terms of standard accuracy metrics. The results indicate that our algorithm achieves superior performances and considering fairness in group recommendation can enhance the recommendation accuracy.

Rendezvous is a fundamental process in constructing cognitive radio networks (CRNs), in which two users find a common channel for communication. The licensed spectrum is assumed to be divided into n non-overlapping channels and the users can sense the spectrum by equipping with cognitive radios. Most of previous works assume that the user can find a set of available channels (the channels not occupied by the licensed users) after spectrum sensing stage and the status of all channels are stable all the time. However, this assumption may not be true in reality and we focus on designing efficient algorithms when the status of the channels varies dynamically. In this paper, we introduce two models to describe the dynamic rendezvous problem. Denote pij as the probability that channel j is available for user i. In the Independent model, assuming all pij variables are independently distributed and we propose efficient algorithms for both synchronous and asynchronous users, which guarantee rendezvous in O (log2 n) and O (log3 n) time slots with high probability respectively. In the Dependent model, two nearby users have relevant available probabilities and we introduce a sensing phase and an attempting phase to guarantee rendezvous in O (ε log3 n log log log n) time slots with high probability, where ε is a small constant. We also present an algorithm to increase rendezvous load in the long run, which guarantee rendezvous for at least 1/4 of all time slots.

Controlling a dynamic network is interesting and important in practical applications, which is to drive the network from any initial state to any desired state. Much research has been conducted in revealing the controllability and seeking the underlying correlations of the network. However, no existing works have considered the time needed to control the network, which we refer to as control latency. In this paper, we initiate the study of control latency of dynamic networks. First of all, we formulate the minimum control latency (MCL) problem for designing the controlling pattern with minimum number of controllers. We show that the MCL problem is NP-hard by reducing the multiprocessor scheduling problem to it. Then, we propose a greedy algorithm for designing a controlling pattern that can control the network within two times the minimum control latency. Moreover, when the control latency is bounded by a given value, we propose another constant approximation algorithm to design a controlling pattern which uses at most three times the minimum number of controllers. We conduct extensive simulations on both synthetic and real networks to corroborate our theoretic analysis.

Rendezvous is a fundamental process in constructing Cognitive Radio Networks (CRNs), through which the user can communicate with its neighbors by establishing a link on some licensed frequency band (channel). Most of the existing elegant rendezvous algorithms assume each user is equipped with a single radio. Nowadays the multi-radio cognitive radio architecture, where each user can access k ≥ 2 channels at the same time, has become a reality. In this paper, we study the rendezvous problem in multi-radio CRN to see whether and to what extent the multi-radio capability can improve the rendezvous performance. To begin with, we propose a family of deterministic distributed algorithms for two special situations when k=2 and k=O(√n), where n is the number of all channels. These algorithms show that the maximum time to rendezvous (MTTR) can be reduced (largely) in multi-radio CRN. Then we derive a lower bound of MTTR as Ω({|Vi||Vj|}/k2) for arbitrary k (Vi, Vj represents two users' available channel sets) and present a distributed algorithm to guarantee rendezvous in O({|Vi||Vj|}/k2) time slots, which meets the lower bound. Extensive simulations are conducted to corroborate our theoretical analyses.