Han Qin

Considering data size and privacy concerns in a distributed setting, it is neither desirable nor feasible to translate data from one resource to another in data mining. Rather, it makes more sense to first mine knowledge from one data resource and then translate the discovered knowledge (models) to another for knowledge reuse. Although there have been successful research efforts in knowledge transfer, the knowledge translation problem in the semantically heterogenous scenario has not been… Show more

Considering data size and privacy concerns in a distributed setting, it is neither desirable nor feasible to translate data from one resource to another in data mining. Rather, it makes more sense to first mine knowledge from one data resource and then translate the discovered knowledge (models) to another for knowledge reuse. Although there have been successful research efforts in knowledge transfer, the knowledge translation problem in the semantically heterogenous scenario has not been addressed adequately. In this paper, we first propose to use Semantic Web ontologies to represent rule-based knowledge to make the knowledge computer “translatable”. Instead of an inductive learning approach, we treat knowledge translation as a deductive inference. We elaborate a translation method with both the forward and backward chaining to address the asymmetry of translation. We show the effectiveness of our knowledge translation method in decision tree rules and association rules mined from sports and gene data respectively. In a more general context, this work illustrates the promise of a novel research which leverages ontologies and Semantic Web techniques to extend the knowledge transfer in data mining to the semantically heterogeneous scenario. Show less

Financial forecasting is the basis for budgeting activities and estimating future financing needs. Applying machine learning and data mining models to financial forecasting is both effective and efficient. Among different kinds of machine learning models, kernel methods are well accepted since they are more robust and accurate than traditional models, such as neural networks. However, learning from multiple data sources is still one of the main challenges in the financial forecasting area. In… Show more

Financial forecasting is the basis for budgeting activities and estimating future financing needs. Applying machine learning and data mining models to financial forecasting is both effective and efficient. Among different kinds of machine learning models, kernel methods are well accepted since they are more robust and accurate than traditional models, such as neural networks. However, learning from multiple data sources is still one of the main challenges in the financial forecasting area. In this paper, we focus on applying the multiple kernel learning models to the multiple major international stock indexes. Our experiment results indicate that applying multiple kernel learning to the financial forecasting problem suffers from both the short training period problem and non-stationary problem. Therefore we propose a novel multiple kernel learning model to address the challenge by introducing the Gompertz model and considering a non-linear combination of different kernel matrices. The experiment results show that our Gompertz multiple kernel learning model addresses the challenges and achieves better performance than the original multiple kernel learning model and single SVM models.
Show less

Integrating existing relational databases with ontology-based systems is among the important research problems for the Semantic Web. We have designed a comprehensive framework called OntoGrate which combines a highly automatic mapping system, a logic inference engine, and several syntax wrappers that inter-operate with consistent semantics to answer ontology-based queries using the data from heterogeneous databases. There are several major contributions of our OntoGrate research: (i) we… Show more

Integrating existing relational databases with ontology-based systems is among the important research problems for the Semantic Web. We have designed a comprehensive framework called OntoGrate which combines a highly automatic mapping system, a logic inference engine, and several syntax wrappers that inter-operate with consistent semantics to answer ontology-based queries using the data from heterogeneous databases. There are several major contributions of our OntoGrate research: (i) we designed an ontology-based framework that provides a unified semantics for mapping discovery and query translation by transforming database schemas to Semantic Web ontologies; (ii) we developed a highly automatic ontology mapping system which leverages object reconciliation and multi-relational data mining techniques; (iii) we developed an inference-based query translation algorithm and several syntax wrappers which can translate queries and answers between relational databases and the Semantic Web. The testing results of our implemented OntoGrate system in different domains show that the large amount of data in relational databases can be directly utilized for answering Semantic Web queries rather than first converting all relational data into RDF or OWL.
Show less

Lead authors: Jun Li and Dejing Dou
Short paper

Lead authors: Dejing Dou and Jun Li
Short paper

Creating executable semantic mappings is an important task for ontology-based information integration. Although it is argued that mapping tools may require interaction from humans (domain experts) for best accuracy, in general, automatic ontology mapping is an AI-Complete problem. Finding matchings (correspondences) between the concepts of two ontologies is the first step towards solving this problem but matchings are normally not directly executable for data exchange or query translation. This… Show more

Creating executable semantic mappings is an important task for ontology-based information integration. Although it is argued that mapping tools may require interaction from humans (domain experts) for best accuracy, in general, automatic ontology mapping is an AI-Complete problem. Finding matchings (correspondences) between the concepts of two ontologies is the first step towards solving this problem but matchings are normally not directly executable for data exchange or query translation. This paper presents an systematic approach to combining ontology matching, object reconciliation and multi-relational data mining to find the executable mapping rules in a highly automatic manner. Our approach starts from an iterative process to search the matchings and do object reconciliation for the ontologies with data instances. Then the result of this iterative process is used for mining frequent queries. Finally the semantic mapping rules can be generated from the frequent queries. The results show our approach is highly automatic without losing much accuracy compared with human-specified mappings. Show less