| Research methods based on social networks have been widely used in many fields of humanities and social sciences.The foundation of social network research is the construction of social networks.With the rise of big data research,automatic extraction of social networks from massive data has become an emerging hotspot in social network construction.Social network extraction refers to the technology of automatically extracting social members and their connections from data sources based on explicit or implicit information.Online encyclopedias contain a wealth of information about social members and their social relationships.How to extract large-scale social networks from such data source is a major issue worth exploring.In addition,the purpose of social network construction is to further analyze and utilize the constructed social network.In recent years,the semantic social network combining the semantic web technology and the social network has begun to attract the attention of the academic community.With the help of ontology and semantic query language,semantic social networks can have logical reasoning capabilities,which is helpful to mine a large amount of potential semantic information and social member connections from social networks,then further serve academic research.Based on the above background,the purpose of thesis is to use online encyclopedias as main information sources to explore a social network extraction mechanism for semistructured text,and to discover the semantization methods for large-scale social network as well as its applications in the digital humanities field.The main research contents of this thesis are as follows:(1)In order to solve the problem of social network extraction of online encyclopedia,a new method of large-scale social network extraction based on online encyclopedia is proposed.The main innovation of this method is to use Learning to Rank(L2R)method to synthesize a variety of features to calculate the weights of person relationships,and to identifier the spatiotemporal coupling relationship between persons by estimating his/her living time-space.Based on this method,this research extracted a cross time-space people network and a timespace coupling network from the Chinese encyclopedia.(2)Based on the investigation of the current social network ontology,a new social network ontology named MSTSN is proposed.Compared with similar social network ontologies,MSTSN ontology is mainly used for the construction of social networks in the field of digital humanities.In particular,it provides a fine-grained description of semantic information such as the space and time in regard to persons,the types of persons,as well as the types of different relationships between persons.By means of instantiation based on MSTSN ontology,the social network was transformed into a large-scale knowledge graph.(3)A new method of relation prediction in knowledge graphs is proposed in the process of constructing knowledge graph,aimed at solving the difficult problem of person relationship prediction tasks.The method matrixes relational paths and texts that reflect entity relationships,learns the structure and text pattern features related to the specific relation type through convolutional neural networks,then a model was trained for relationship prediction tasks.The results of comparative experiments show that the performance of our proposed method in evaluation data sets is superior to the state-of-the-arts.(4)Aiming at the triplet extractions of person relationships in unstructured texts,a new Chinese Named Entity Linking(NEL)method is proposed.This method integrates single and collective named entity disambiguation features,and adopts different combinations of features in accordance with the different text lengths.In addition,a two-stage disambiguation strategy,which can optimize the result of the first-round of disambiguation,was designed.A comparative experiment demonstrated that the performance of this NEL method is superior to that of a similar state-of-the-art system.(5)Based on the MSTSN ontology and the knowledge graph proposed in this study,four kinds of semantic query strategy of social network for digital humanities research are elaborated,and the implementation effect of each strategy is demonstrated by visualization method.Compared with the traditional social network construction strategy,the proposed methods in this study can extract social actors and their relationships from the semi-structured information in online encyclopedia,and pays more attention to the spatio-temporal and semantic information related to social actors.The algorithms of name disambiguation and relationship weight calculation are also improved in this study,which can provide a theoretical and methodological reference for the study of automatic construction of large-scale social networks.The large-scale knowledge graph constructed in this study has practical application value in social network analysis systems,question answering systems and knowledge discovery systems related to the humanities,which can provide a reference for exploring the application pattern of text mining and semantic web technology in digital humanities. |
PDF Full Text Request