Applications Of Markov Logic Networks In Trust Based Recommender Systems And Chinese Temporal Relation Identification

To combine probabilistic graphical model and first-order logic in a single simpleform of expression has been a longstanding goal of artificial intelligence research. Wecan use probabilistic graphical model to handle uncertainty efficiently and usefirst-order logic to represent the knowledge compactly. In many practical applications,we need to combine them. Markov Logic Network(MLN) is one of those concise andpowerful languages, and it combines probabilistic graphical model and first-order logicwell. In recent years, Markov Logic Network has been a hot issue in the field ofartificial intelligence research. It has a wide application in the field of social relations,biology, computer science and so on.This thesis focuses on the research of the theory system of Markov Logic Networkand its applications in trust based recommender systems and Chinese temporal relationidentification. The main works and research results summarized as follows:Firstly, this thesis studies the related theory of Markov Logic Network.This thesis introduces the related background knowledge of statistical relationallearning, including its concept, method and application fields. Then it introducesMarkov Logic Network and its related concepts, including first-order logic and MarkovNetwork. It also discusses Markov logic network’s weight learning algorithms andreasoning algorithms.Secondly, the thesis studies the application of Markov Logic Network in trustbased recommender systems.The introduction of trust in recommender systems has solved problems which existin collaborative filtering recommender systems in some degree, such as cold start, datasparsity,"prop" attack problems and so on. However, trust is a complex concept,propagating and predicting trust determine the recommender system’s quality. Thisthesis uses Markov Logic Network to build the model of trust propagation, and uses itsinference algorithm to predict the relationship of trust. The experimental results showthat Markov Logic Network has a higher accuracy and better solution for cold-userproblem than MoleTrust approach.Thirdly, the thesis studies Chinese temporal relation identification based onMarkov Logic Networks. Chinese temporal relation identification is a fundamental task in Chinese semanticinformation processing. Using the traditional machine learning method to solvetemporal relation identification problem is considered as a local classification method.We use Markov Logic Network to solve Chinese temporal relation identificationproblems. According to the local features and global features, we construct thecorresponding Markov logic network models. The experimental results show that theglobal Markov Logic Network model is better than the maximum entropy method inChinese temporal relation identification.Based on the researches above, we could see that Markov Logic Network unifiedprobabilistic graphical model and first order logic. It can handle uncertainty well, andallow incompletion and contradictory. Moreover it can be used as a template ofconstructing Markov network, so it has a very wide range of applications.