Topic Discovery And Trend Analysis In Scientific Literature Based On Topic Model

We live in an information explosion age. The development of science and technology are also changed rapidly. How to obtain the latest research developments quickly is an important problem for researchers. In order to understand the latest research work, the scientific and technical worker will pay attention to the key issues in the field, such as what kind of technology has been used, and in a large number of technologies, which is a research hotspot at present, which has gradually been forgotten. Therefore, in order to automatically analyze the technology trends, we take use of topic model and some features of electronic documents. Then we propose a new method for topic discovery and trend analysis in scientific literature based on topic model.In this paper, we use the LDA model to extract the semantic information and generate topic from scientific literature, then we find the support documents for each topic according to the generated topic space and calculate the document support rate as topic strength. Next, we put the topic strength in the time axis, obtain the strength trends. Finally, we calculate the topic word weight in different time slot, and reorder the topic words, get the topic evolution path.The experiments on NIPS anthology and ACL anthology have shown the trend of machine learning and computational linguistics. We get the hot research field of these conference, and also get the developing course of these subfield such as machine translation,kernel method. The results reflect the situation of computational linguistics and machine learning.In contrast to the baseline method, we verified our method is more effective and have the better topic evolution.