Design And Building Of The Domain-specific Knowledge Base System For Internet Videos

In recent years,the term big data in recent years has a high frequency of occurrence.With technologies related,new breakthrough progress has been made in fields.In the era of big data,vertical search and personalized recommendation derived from the traditional information retrieval help people from an ocean of data find the information more accurately that they are interested in.Technologies develop rapidly in data mining,machine learning distributed computing,providing more possibilities for evolution in search and recommend areas.Under the background of the above,knowledge base(KB)has received significant and growing attention,in both industry and academia.Using the knowledge provided by KB,the vertical search engine can better understand of the intention of user queries,and improves the user experience with more comprehensive and accuracy search results.Meanwhile,the KB system can better analyze user characteristics combining with the domain knowledge and better describe the the entities recommendation system involves,providing more space for personalized recommendation system optimization.This paper accomplish the task of design and building the domain-specific knowlege base system for internet videos on the basis of data from most domestic video website and wiki-pedia website.First,we research in the theory,the key technology and the using of KB and analyze the design of the domain-specific knowlege base including data source research,and the design of taxonomy tree,the process of data collection and recordlinkage,and the application of KB in vertical search and personalized recommendation.Then,through the analysis and research dynamic web page technology used in the different websites,we developed a topical web crawler system with the ability to obtain the specified topic data which is the essential basic data for the KB building.Last but certainly not least,we detail the implementation of video record linkage and solution to the problem occur in practice;we also proposes a new method for classification in record linkage which combines two-step approach to training Support Vector Machine classification with controllable manual review;The experiment result of video record linkage based on a large number of real data achieve 99%F-score which proves that most record linkage methods used in people records can expend to other domain and obtain satisfied effects.