A way to construct evolution model of scientific papers based on the seed document and OLDA models

Tracking the topic evolution in scientific papers is an important issue. This paper presents a topic evolution framework which is based on the seed document and Online Latent Dirichlet Allocation (OLDA) models. We define seed document and link the seed documents of current time slice into the documents of next slice to keep the continuity of topics in content. And then, we run the OLDA in each time slice and take the word-topic posterior probability in previous time slice as the word-topic prior probability in current time slice to keep the continuity of topics in timeline. The cosine measure is used to compute the topic similarity in the adjacent time slices to find the topic evolution. The experimental results show that the way can help find the topic evolution in continuous time.