Abstract Document priors that encode our prior knowledge about the importance of different documents are essential to an expert finding system. This study proposed a TopicRank-based document priors model for expert finding. TopicRank algorithm is an extension of the DocRank algorithm. Latent dirichlet allocation was used to extract topics of the documents. We assumed there was a link between two documents that share common topics. Link analysis techniques were then used to obtain document priors. The proposed model was evaluated using the CSIRO Enterprise Research Collection and the results showed that the performance of the expert finding system was dramatically improved by introducing TopicRank-based document priors. In particular, Mean Average Precision increased 19.9% while Mean Reciprocal Rank rose as much as 23.4%.
A TopicRank Based Document Priors Model for Expert Finding
2017-01-01
8 pages
Article/Chapter (Book)
Electronic Resource
English
British Library Conference Proceedings | 2016
|Image-Based Rendering Using Image-Based Priors
British Library Online Contents | 2005
|