Topic-based Academic Network Construction

Academic related content, such as academic citations and author profiles, consistently draw researchers’attention. Deeply unearthing the latent structures or topics of this academic information has a beneficial effect on researchers’study. Yet nowadays popular search engines and large professional scientific literature databases are not making full use of the literature information. There is a huge development potential for scientific literature retrieval.As it can help researchers understand the history and hotspots of their current fields of research, discovering the inner structures and the underlying trends of massive academic literature can be significant for researchers’work. Under the consistency of topic and the evolution of underlying theme, a topic-based academic network has been built in this thesis. The major contributions of this dissertation are as follows:1、A divide-and-conquer strategy is used to help construct the topic-based academic network. Firstly, we separate the entire academic network into many groups. We then construct a detailed network diagram for each academic group. And finally, we use the connection between every two groups to reconstruct the entire topic-based network.2、 A tensor decomposition method is introduced to complete the divide step of the construction of topic-based academic network. The introduced method can be used to tackle the decomposition of massive complex structure hyper-graph. The method has strong scalability and low time complexity, and is suitable for large-scale data community discovery. Meanwhile, a comparison for K-means, dynamic topic model and tensor decomposition in the effectiveness and performance of clustering is accomplished.3、 Authority articles and topic-based path generation method are combined to build the academic community network. The combined method is easy to implement, and achieves a good effect in the experiment. Topic-based path generation method can avoid a few weakness of the shortest path method, for example, the shortest path method may generate inconsistent theme along the path and converge to a local minimum.4、 A system is constructed to help build the topic-based academic network. The main modules which the system provided are as follows:automatically collecting and abstracting academic information from online network, processing and indexing raw data, implementation of core algorithms for topic-based academic network construction.