Abstract
Topic modelling and trend analysis are increasingly important in today's digital world, especially for identifying promising business ideas and trends. With the increasing amount. of data being generated daily, a key challenge is to effectively identify emerging business ideas/topics and trends from this large volume of data. Towards this effort, we introduce a framework that allows us to identify promising business ideas from a large stream of academic papers. Academic papers are suitable for this purpose as they study emerging areas and problems in different domains. Our framework comprises three main components, namely: (i) a data collection component that retrieves academic papers and their meta-data; (ii) a topic modelling algorithm that combines traditional topic modelling techniques with recent advances in word embeddings; and (iii) a trend analysis component that allows us to visualize the popularity of different business trends/topics across time. Results on a corpus of 287k academic papers show that our proposed methods outperform the standard baselines based on topic coherence scores and also allows us to understand key temporal trends.