The Modeling Of Object In Academic Social Network Based On Statistical Topic Model And Applications

With the rapid development of the Internet, Web information is becoming increasing rapidly. Academic network is constituted by many different types of object, such as paper, conference and author. These disparate objects are depending on each other, between which there exists a closely semantic association. The contents of the paper could reflect its semantics theme to some extent; Author write the paper based on their own research interests, and submit it for publication to the conference with the corresponding subject areas; Conference also has its main research areas, which is consistent with the semantic topic of the included papers and the authors’ research interest. Mining the semantic information in the academic networks is of great important significance for take full advantage of the academic network. Further the modeling for academic network object is a critical step.The object modeling methods for academic network include language model LM and several commonly used statistical topic model, such as LDA, Author-Topic (AT) Model and Author-Conference-Topic (ACT) Model. LM, LDA, and the AT can only be achieved on a single object modeling, which have ignored the semantic association between the academic network objects. ACT represents the network objects with the same topic space, and implements a unified semantic modeling of objects in a heterogeneous network. But the words topic and the conference topic are always not equivalent. The semantic topic characterized by words is fine-grained level, while the topic of conferences is relatively coarse-grained level. Therefore, the unified modeling effects and semantic mining for academic network objects is still to be improved.This paper conduct a further study on unified semantic modeling of heterogeneous academic network object based on the idea of statistical topic model, including the following three aspects:Firstly, this paper proposes a new statistical topic model named Author-Conference Topic-Connection (ACTC). With mining the semantic association information between the academic network objects, these objects could be mapped to the appropriate semantic topic layer, so that a unified semantic modeling for the academic network objects could be achieved. The experiments show that the ACTC model could achieve better semantic representation for the academic network objects than other methods.Secondly, we apply the ACTC model in academic search, including expert search, conference search and paper search. ACTC model could tap the semantic association between the academic network object more effectively, and obtain the semantically related expert, conference and papers with the latent semantic topics. In the experiment, the academic search results is compared between ACTC with several commonly used statistical topic model and statistical language model, including LM, LDA, AT and ACT, and the analysis for the lack of baseline method is obtained.Lastly, we establish academic network retrieval system ACLMiner with the ACTC model. With the unified semantic modeling for the academic network object in the field of computational linguistics, the system could mine the semantic information of all academic objects, so as to provide semantic academic retrieval services.