Research On Domain Knowledge Learning And Update Techology Based On Markov Logic Networks

It has been the focus of academia and application field since KnowledgeDiscovery in Data was proposed, but facing the increasingly complex domainapplication environment, most of the existing research methods are based onstatistical learning, which often neglect the relationship in domain knowledge andthe change of domain knowledge over time, leading to that the effect of applicationis often just passable. How to grasp the statistical relationship in domain knowledgeto accurately make the correct domain knowledge learning and domain knowledgeupdate, have been a problem in domain knowledge research. Recently, with thedevelopment of research on statistical relational learning, Markov Logic Networks,which combines the probability graph and first-order logic theory, is proposed andapplied successfully in Natural Language Processing, Machine Learning, SocialNetwork Analysis and so on. In this context, this paper adopts Markov LogicNetworks, constructs the domain knowledge base with domain data sets, and carriesout exploratory study on domain knowledge learning and domain knowledgeupdate.In this paper, the specific research content and the research results are asfollows:(1) Firstly, this paper introduces a traditional method based on SVM to do thedomain knowledge classification learning, then proposes a new method based onMarkov Logic Networks to do it. On the basis of the traditional method, the newmethod gets a better learning effect by bringing in first-order logic to express therelationship in domain knowledge. Compared with two experimental groups, it isfound that on average, the domain knowledge classification learning based onMarkov Logic Networks gets5~6percentage points higher effect than another.(2) In terms of domain knowledge update, facing the domain task of this paper—text classification, and considering three traditional knowledge update strategies,we come up with the domain knowledge update technology based on featurethesaurus incremental learning, and make a comparative analysis of the threestrategies to verify its effectiveness and feasibility. (3) Considering that traditional methods on domain knowledge update havesome shortcomings—not taking the relationship in domain knowledge into account,we extend the research on Markov Logic Networks, combine with the incrementalknowledge learning strategy, and raise the domain knowledge update technologybased on Markov Logic Networks. Having gotten knowledge via domain knowledgelearning based on Markov Logic Networks, we join the new knowledge into theoriginal one and update the domain knowledge base. The experimental results showthat the knowledge base gained by this method raises2~3percentage points inaccuracy when making classification decision.Based on the above analysis, we know that Markov Logic Networks unifiedprobability graph model and first-order logic, handles the statistical relationship indomain information very well, and has a good effect in domain knowledge learningand domain knowledge update. It has extensive research prospects.