Kun Yue

A Parallel and Incremental Approach for Data-Intensive Learning of Bayesian Networks

Bayesian network (BN) has been adopted as the underlying model for representing and inferring uncertain knowledge. As the basis of realistic applications centered on probabilistic inferences, learning a BN from data is a critical subject of machine learning, artificial intelligence, and big data paradigms. Currently, it is necessary to extend the classical methods for learning BNs with respect to data-intensive computing or in cloud environments. In this paper, we propose a parallel and incremental approach for data-intensive learning of BNs from massive, distributed, and dynamically changing data by extending the classical scoring and search algorithm and using MapReduce. First, we adopt the minimum description length as the scoring metric and give the two-pass MapReduce-based algorithms for computing the required marginal probabilities and scoring the candidate graphical model from sample data. Then, we give the corresponding strategy for extending the classical hill-climbing algorithm to obtain the optimal structure, as well as that for storing a BN by <;key, value> pairs. Further, in view of the dynamic characteristics of the changing data, we give the concept of influence degree to measure the coincidence of the current BN with new data, and then propose the corresponding two-pass MapReduce-based algorithms for BNs incremental learning. Experimental results show the efficiency, scalability, and effectiveness of our methods.

Discovering semantic associations among Web services based on the qualitative probabilistic network

In recent years, the intelligent management and decision of Web services have attracted more and more attention due to the wide applications in various aspects of the real world. With the increase of Web services in an organization, the desired on-line services should be located rapidly requiring not only the syntactic but also the semantic techniques. In addition, aiming at fulfilling complex applications by discovering and composing available services automatically and precisely, it is indispensable to develop an underlying model and the corresponding measure for semantic associations among given Web services. In this paper, by mining the historical invocations of component services, we first construct a semantic model to describe their behavior rules based on the qualitative probabilistic network. Further, we propose a distance measure and the approach to discovering semantic associations among Web services. Preliminary experiments and performance studies show that our methods are feasible. Moreover, high recall and precision can be achieved when our methods are applied to Web service search.

Constructing the Bayesian network structure from dependencies implied in multiple relational schemas

Research highlights? We propose the method for constructing Bayesian network structures from multiple 3NF relational schemas. ? We establish our method based on the acyclic database theory and its relationship with probabilistic networks. ? The Bayesian network structure is constructed from relational schemas instead of database instances. Relational models are the most common representation of structured data, and acyclic database theory is important in relational databases. In this paper, we propose the method for constructing the Bayesian network structure from dependencies implied in multiple relational schemas. Based on the acyclic database theory and its relationships with probabilistic networks, we are to construct the Bayesian network structure starting from implied independence information instead of mining database instances. We first give the method to find the maximum harmoniousness subset for the multi-valued dependencies on an acyclic schema, and thus the most information of conditional independencies can be retained. Further, aiming at multi-relational environments, we discuss the properties of join graphs of multiple 3NF database schemas, and thus the dependencies between separate relational schemas can be obtained. In addition, on the given cyclic join dependency, the transformation from cyclic to acyclic database schemas is proposed by virtue of finding a minimal acyclic augmentation. An applied example shows that our proposed methods are feasible.

An approach for adaptive associative classification

As a branch of classification, associative classification combines the basic ideas of association rule mining and general classification. Previous studies show that associative classification can achieve a higher classification accuracy comparing with traditional classification methods, such as C4.5. It is known that new frequent patterns may emerge from the classified resources during classification, and these newly emerging frequent patterns can be used to build new classification rules. However, this dynamic characteristics in associative classification has not been well reflected in traditional methods. In this paper, we propose an enhanced associative classification method by integrating the dynamic property in the process of associative classification. In the proposed method, we employ co-training to refine the discovered emerging frequent patterns for classification rule extension and utilize the maximum entropy model for class label prediction. The empirical study shows that our method can be used to classify increasing resources efficiently and effectively.

Qualitative probabilistic networks with reduced ambiguities

A Qualitative probabilistic network (QPN) is the qualitative abstraction of a Bayesian network that encodes variables and the qualitative influences between them. In order to make QPNs be practical for real-world representation and inference of uncertain knowledge, it is desirable to reduce ambiguities in general QPNs, including unknown qualitative influences and inference conflicts. In this paper, we first extend the traditional definition of qualitative influences by adopting the probabilistic threshold. In addition, we introduce probabilistic-rough-set-based weights to the qualitative influences. The enhanced network so obtained, called EQPN, is constructed from sample data. Finally, to achieve conflict-free EQPN inferences, we resolve the trade-offs by addressing the symmetry, transitivity and composition properties. Preliminary experiments verify the correctness and feasibility of our methods.

A MapReduce-Based Method for Learning Bayesian Network from Massive Data

Bayesian network (BN) is the popular and important probabilistic graphical model for representing and inferring uncertain knowledge. Learning BN from massive data is the basis for uncertain-knowledge-centered inferences, prediction and decision. The inherence of massive data makes BN learning be adjusted to the large data volume and executed in parallel. In this paper, we proposed a MapReduce-based approach for learning BN from massive data by extending the traditional scoring & search algorithm. First, in the scoring process, we developed map and reduce algorithms for obtaining the required parameters in parallel. Second, in the search process, for each node we developed map and reduce algorithms for scoring all the candidate local structures in parallel and selecting the local optimal structure with the highest score. Thus, the local optimal structures of each node are merged to the global optimal one. Experimental result indicates our proposed method is effective and efficient.

Containment of competitive influence spread in social networks

To contain the competitive influence spread in social networks is to maximize the influence of one participant and contain the influence of its opponent. It is desirable to develop effective strategies for influence spread of the participants themselves instead of blocking the influence spread of their opponents. In this paper, we extend the linear threshold model to establish the diffusion-containment model, abbreviated as D-C model, by incorporating the realistic specialties and characteristics of the containment of competitive influence spread. Then, we discuss the influence spread mechanism for the D-C model, and give the algorithm for the propagation of the diffusion influence (D-influence) and containment influence (C-influence). Further, we define the sub-modular set function of the C-influence in the D-C model and consequently give a greedy algorithm for solving the problem of maximizing the competitive influence containment approximately. Experimental results show the feasibility of our method.

Context-aware recommendation via graph-based contextual modeling and postfiltering

Context-aware recommender systems generate more relevant recommendations by adapting them to the specific contextual situation of the user and have become one of the most active research areas in the recommender systems. However, there remains a key issue as how contextual information can be used to create intelligent and useful recommender systems. To assist the development and use of context-aware recommendation capabilities, we propose a graph-based framework to model and incorporate contextual information into the recommendation process in an advantageous way. A contextual graph-based relevance measure (CGR) is specifically designed to assess the potential relevance between the target user and the items further used to make an item recommendation. We also propose a probabilistic-based postfiltering strategy to refine the recommendation results as contextual conditions are explicitly given in a query. Depending on the experimental results on the two datasets, the CGR-based method is much superior to the traditional collaborative filtering methods, and the proposed postfiltering method is much effective in context-aware recommendation scenario.

Probabilistic representation and approximate inference of type-2 fuzzy events in Bayesian networks with interval probability parameters

It is necessary and challenging to represent the probabilities of fuzzy events and make inferences between them based on a Bayesian network. Motivated by such real applications, in this paper, we first define the interval probabilities of type-2 fuzzy events. Then, we define weak interval conditional probabilities and the corresponding probabilistic description. The expanded multiplication rule supporting interval probability reasoning. Accordingly, we propose the approach for learning the interval conditional probability parameters of a Bayesian network and the algorithm for its approximate inference. Experimental results show the feasibility of our method.

A data-intensive approach for discovering user similarities in social behavioral interactions based on the bayesian network

Discovering user similarities from social media can establish the basis for user targeting, product recommendation, user relationship evolution and understanding. User similarities not only depend on the topological structure but also the dependence degrees between users. In this paper, we adopt Bayesian network (BN), an important and popular probabilistic graphical model, as the underling framework and propose a data-intensive approach for discovering user similarities. First, upon the massive social behavioral interactions, we give the method for measuring direct similarities between users and the MapReduce-based algorithm for constructing a BN to describe these similarities, called user Bayesian network and abbreviated as UBN. We also give the idea for storing large-scale UBNs in a distributed file system. Then, to measure indirect similarities between users, we give the method for measuring the closeness of user connections in terms of the properties of UBNs graphical structure. Further, we give the MapReduce-based algorithm for measuring the dependence degrees by means of UBNs probabilistic inferences. By combining the above two perspectives of measures, the indirect similarity degree between users can be achieved, while guaranteeing the applicability theoretically. Finally, we give experimental results and show the efficiency and effectiveness of our method.