AHM: A Novel Model for Mining Academic Hot Spots Based on a Scientific Knowledge Graph

Academic hot spots refer to a group of words that are widely considered in a specific period. Academic hot-spot mining is an essential task in the bibliometrics field, and the purpose is to mine academic hot-spots in a large number of studies. The common methods at present are bibliometrics, based analysis tools and machine learning methods, and these methods cannot fully apply deep semantic features. With the increase in academic papers, extracting hot spots is difficult through bibliometrics and machine learning methods. Combining deep learning technology to extract deep features and mine hot spots more efficiently and accurately is a challenge in the field of academic hot spot research. This paper proposed a novel model called AHM to compensate for the current shortcomings, which improves the feature representation of k-means++, obtains deep semantic and contextual features by applying deep learning technology, and fuses the two features as feature inputs for k-means++. The experimental results of a comparison of AHM with four baselines (i.e., k-means++, TF-IDF+k-means++, Word2vector+k-means++, and Node2vector+k-means++) on the literature datasets in the artificial intelligence field show that the AHM model has a better effect in academic hot-spot mining tasks. In addition, this paper presents an overall framework for evolution analysis and analyses the evolution path of academic hot spots during 2010−2020.