Ontology Alignment Approach Based On Swarm Intelligence Algorithm

The Internet is an indispensable information exchange tool for people today.The massive amount of information is generated,which urgently needs to be automated and intelligently managed,and to provide users with rich and effective services.Ontology provides a normative description of terms that can be shared in the domain and the relationship between terms.This task-independent semantic model occupies an important position in artificial intelligence,information systems,and the semantic web.In recent years,as the importance of ontology has gradually emerged,people have developed many ontologies from the same domain of interest.At the same time,there has been a phenomenon of using different terms to describe the same thing.This phenomenon is called the problem of ontology heterogeneity.In order to solve this problem,the ontology alignment or ontology matching process has become a key semantic technology.This process can find a set of entity correspondences between given different ontology entities.This corresponding set can be used in fields such as ontology fusion,data integration,and translation.Therefore,the ontology alignment approach has important research value.In order to obtain high-quality ontology alignment results,researchers have proposed a large number of automatic or semi-automatic ontology alignment methods.However,these methods still have some problems,such as poor scalability and poor alignment quality.The ontology alignment approach based on swarm intelligence algorithm is a promising method.This method has inherent parallelism and can be flexibly extended through a variety of basic matchers to achieve high-quality alignment.This dissertation addresses the problems and challenges of current ontology alignment approaches based on swarm intelligence and conducts research from three aspects: selection and combination of basic matchers,user intervention and effective large ontology alignment.The specific research achievements are as follows:(1)A scalable meta-matching approach is investigated.In the ontology alignment community,recent empirical analysis shows that a single matcher that does not have a dominant position can perform best in all application cases.Therefore,designing an ontology alignment system often requires combining multiple basic matchers.However,how to fully combine the matchers to make the system achieve first-class performance has become one of the important challenges in the research of ontology alignment methods.In recent years,ontology alignment methods based on metaheuristics have been shown to be an effective method.However,these methods have two significant shortcomings: the weights of the combined basic matcher are manually set based on experiments;a reference alignment or prior knowledge needs to be provided by domain experts under the optimization alignment.This dissertation studies the recently proposed grasshopper optimization algorithm(GOA),and combines its advantages to propose a meta-matching method with scalability for ontology alignment based on GOA.This approach first transforms ontology alignment into an optimization problem.A suboptimal alignment is found by optimizing the weights of the combined matchers.To evaluate candidate solutions,A comprehensive evaluation function is formulated,which includes the number of correspondences found by the current solution and the average similarity score.Finally,experiments on the benchmark track and conference track provided by the Ontology Alignment Evaluation Initiative(OAEI)show that the proposed method achieves high-quality alignment and is significantly better than other methods based on meta-heuristics.(2)Based on the meta-matching method,a novel periodic learning ontology matching model is proposed.In the ontology alignment community,although researchers have proposed a large number of effective and fully automatic ontology alignment methods,it is necessary to allow users to intervene in the ontology alignment process in some real application scenarios.In order to further improve the quality of ontology alignment,user intervention has been considered one of the effective methods,especially in some cases that require precise matching.However,in the complex alignment process,user intervention faces new challenges,such as how the user intervenes is unburdened for the user and ensures the effectiveness of the interaction.In order to solve these problems,this dissertation proposes a novel periodic learning ontology matching model based on interaction grasshopper optimization algorithm.The model allows users to periodically feedback knowledge,rather than every generation,and introduces a roulette method to select the candidate mappings with the lowest matching degree to present to the user,instead of presenting all of them to the user to reduce the burden on user.In order to ensure the effectiveness of each interaction,a reward and punishment mechanism is proposed to propagate the knowledge of user feedback to the evolutionary population to guide the search direction of the algorithm.Experiments on two data sets for interactive methods provided by OAEI show that the proposed model significantly improves the matching quality.Compared with other state-of-the-art matching systems,the model proposed outperforms other methods in almost all cases given different error rates.Finally,this work studies a typical data integration case to show how the proposed approach is able to help enterprises to harmonize product catalogs.(3)Based on the meta-matching method,an effective approach for large ontology alignment is proposed.Existing ontology alignment approaches based on stochastic population search technology have been proven to obtain excellent results on small-scale ontology alignment tasks.However,as the size of the ontology increases,the solution space increases exponentially.These methods cannot effectively solve the problem of large ontology alignment,and even memory overflow errors occur.In order to solve these problems,this dissertation proposes an effective large ontology alignment approach based on improved grasshopper optimization algorithm(Solitarious and Gregarious Behavior GOA-OM,SGBGOA-OM).This method attempts to solve the problem of large ontology alignment from two aspects.On the one hand,the problem of slow convergence caused by the small step movement of the grasshopper optimization algorithm.A novel grasshopper optimization algorithm based on swarm state difference is proposed to improve the convergence speed and accuracy of grasshopper optimization algorithm in large-scale search spaces.On the other hand,a large ontology pruning technique based on Jaccard measurement is proposed to try to trim while keeping the original structure of the ontology unchanged.In order to verify the performance of the improved grasshopper optimization algorithm,a systematic experiment was first carried out on unimodal and multimodal benchmark function.Experimental results show that the improved algorithm is better than the four representative GOA and other meta-heuristic algorithms in more cases.Experiments on a real ontology alignment task show that the proposed SGBGOA-OM method can find high-quality alignments with a faster convergence rate.Compared with state-of-the-art OAEI systems and other meta-heuristic-based methods,the proposed method achieves high-quality alignments on multiple tasks.In order to solve the key semantic technology of ontology alignment,this paper studies from three aspects: the selection and combination of basic matchers,the interactive ontology alignment that allows user intervention,and the large ontology alignment technology.Specifically,for the problems and challenges of existing methods,this work studies and solves the problems that require manual configuration of combining weights,relying on reference alignment and poor scalability;this work studies and solves the challenge of reducing user workload while increasing interaction effectiveness;this work studies and solves the problem that the ontology alignment method based on random population fails for matching large ontologies.