Jinchao Huang

Incorporating knowledge graphs as auxiliary information to enhance recommendation systems can improve the representations learning of users and items. Recommendation methods based on knowledge graphs can introduce user–item interaction learning into the item graph, focusing only on learning the node vector representations within a single graph; alternatively, they can treat user–item interactions and item graphs as two separate graphs and learn from each graph individually. Learning from two graphs has natural advantages in exploring original information and interaction information, but faces two main challenges: (1) in complex graph connection scenarios, how to adequately mine the self-information of each graph, and (2) how to merge interaction information from the two graphs while ensuring that user–item interaction information predominates. Existing methods do not thoroughly explore the simultaneous mining of self-information from both graphs and effective interaction information, leading to the loss of valuable insights. Considering the success of contrastive learning in mining self-information and auxiliary information, this paper proposes a dual-graph contrastive learning recommendation method based on knowledge graphs (KGDC) to explore a more accurate representations of users and items in recommendation systems based on external knowledge graphs. In the learning process within the self-graph, KGDC strengthens and represents the information of different connecting edges in both graphs, and extracts the existing information more fully. In interactive information learning, KGDC reinforces the interaction relationship between users and items in the external knowledge graph, realizing the leading role of the main task. We conducted a series of experiments on three standard datasets, and the results show that the proposed method can achieve better results.

Incorporating knowledge graphs as auxiliary information to enhance recommendation systems can improve the representations learning of users and items. Recommendation methods based on knowledge graph can introduce user-item interaction learning into the item graph, focusing only on learning the node vector representations within a single graph; alternatively, they can treat user-item interactions and item graphs as two separate graphs and learn from each graph individually. Learning from two graphs has natural advantages in exploring original information and interaction information, but faces two main challenges: 1) in complex graph connection scenarios, how to adequately mine the self-information of each graph, and 2) how to merge interaction information from the two graphs while ensuring that user-item interaction information predominates. Existing methods do not thoroughly explore the simultaneous mining of self-information from both graphs and effective interaction information, leading to the loss of valuable insights. Considering the success of contrastive learning in mining self-information and auxiliary information, this paper proposes a dual-graph contrastive learning recommendation method based on knowledge graphs (KGDC) to explore a more accurate representations of users and items in recommendation systems based on external knowledge graphs. In the learning process within the self-graph, KGDC has strengthened and represented the information of different connecting edges in both graphs, and extracted the existing information more fully. In interactive information learning, KGDC reinforces the interaction relationship between users and items in the external knowledge graph, realizing the leading role of the main task. We have conducted a series of experiments on three standard datasets, and the results show that the proposed method can achieve better results.

The next-generation network (6G) has more strict requirements for the online learning ability and high interpretability of the learned systems. Machine learning is expected to be essential to assist in making the networks efficient and adaptable, but most promising methods often are treated as “black boxes” due to the deep structures and high nonlinearity. Therefore, this article attempts to study the interpretations of machine learning algorithms to make them more applicable to the 6G-enabled Internet of things (IoT) networks. Typically, this article focuses on the new model GBDT2NN, which distills the knowledge learned by gradient boosting decision tree (GBDT) into neural network (NN) models to retain the learning ability of numerical data and rise the ability of online learning at the same time, but it loses the interpretability. This article conducts an empirical study on explaining individual prediction of GBDT2NN by taking use of the feature importance learned from GBDT, and then further explores whether the explanation can improve the approximation process. In addition, this article proposes two methods to obtain the interpretations: 1) the independent method and 2) the joint method. The experiments on several data sets of IoT networks show that the proposed methods can achieve better performance on both explanations and predictions.

The extra trust brought by the model interpretation has made it an indispensable part of machine learning systems. But to explain a distilled model's prediction, one may either work with the student model itself, or turn to its teacher model. This leads to a more fundamental question: if a distilled model should give a similar prediction for a similar reason as its teacher model on the same input? This question becomes even more crucial when the two models have dramatically different structure, taking GBDT2NN for example. This paper conducts an empirical study on the new approach to explaining each prediction of GBDT2NN, and how imitating the explanation can further improve the distillation process as an auxiliary learning task. Experiments on several benchmarks show that the proposed methods achieve better performance on both explanations and predictions.

As an important tool of data mining, classification is also one of the major components of the research of Internet of Things (IoT), which has been widely used in many cases, such as smart cities, information abstraction, wireless sensor networks, and so on. IoT could have broader characterization, where diverse data or information could come from ubiquitous and persistent sources. Influenced by various factors, there are a lot of scenes that the data collected from the IoT devices are in the distribution-based form. Therefore, the study of classification for the distribution-based data is very valuable in the field of IoT. To speed up the training process, this paper proposes a new general approach when the types and parameters of distributions are known. It transforms the original problem into a traditional point-valued classification problem with a sampling-based method. Then for the applications that the distribution parameters are not given in advance, this paper also gives an improved approach, which uses a new Bayesian-based method to estimate the distribution parameters. Empirical comparisons conducted on a series of standard benchmark datasets and a real-world dataset from a major Chinese online travel agent site demonstrate that both of our proposed approaches perform better than the existing methods.

With the rapid growing proliferation of online travel agent (OTA) services, personalized recommendations are highly valuable as they can improve customer experiences by preventing the information overload problem. The accurate prediction of users' expectations for price plays an important role in the personalized recommendation of hotels in OTA platforms. Considering that customers' preferences for hotel prices are actually acceptable ranges and traditional point estimations may neglect some informative aspects of the prediction, interval estimation is more suitable for the problem investigated in this paper. However, existing related methods are not applicable due to some specific issues. To provide a better personalized recommendation of hotels in OTA platforms, this paper proposes a novel interval forecasting solution to improve the accuracy of predicting users' price preferences. The novel interval forecasting solution first puts forward a customized objective function which could directly measure the quality of constructed intervals, while also allowing for adjustable tradeoffs between interval tightness and prediction reliability. Then, it combines alternating direction optimization and the gradient boosting framework to efficiently aggregate weak individual predictors to optimize the introduced learning objective. Empirical comparisons conducted on several benchmark standard datasets and a large-scale dataset shared with us by a major Chinese OTA site demonstrate the effectiveness of the proposed approach.