A Study Of The Relationship Between Disciplinary Research Trends And Scholars’ Research Themes Based On A Dynamic Topic Model

In today’s rapidly evolving technological environment,the number of research topics has grown exponentially,which has created new opportunities and challenges for academic research.Throughout a researcher’s career,they may explore a variety of scientific questions,and research topics change over time and as technology advances.For finding a good research topic,there are many complex factors that researchers must consider,among which the direction of research trends in the topic can also have a significant impact on whether researchers choose a research topic.When conducting analysis of research trends in subject areas,existing studies focus on the disciplinary level,analyzing the hot spots or emerging topics at the forefront of research in the subject area.At the scholar level,fewer studies have explored the relationship between scholars’ research topics and disciplinary research trends.At present,assessing scholars’ tracking ability mainly relies on scholars’ selfevaluation or others’ evaluation,and the results are often too subjective.To this end,this paper proposes a method to quantify the relationship between scholars’ research topics and research trends by expanding the dynamic topic model and designing indicators related to research trend directions and scholars’ tracking trend degrees.Based on the data of scholars’ papers in three subfields of machine learning,computer networks and theoretical computer science in computing disciplines from2006 to 2017 on the AMiner platform,this study conducted an empirical research analysis to quantify the relationship between academic research trends and scholars’ research themes,explored the distribution and correlation of research trend indicators and tracking degree indicators,used clustering algorithms to identify five theme types and six types of scholars,and analyzed the distribution of themes and types to understand the behavioral characteristics of scholars such as topic selection.As well as the main findings include the following:(1)In terms of research trends: the three research trend directions in the field of machine learning are highly correlated,with high overlap between long-term research directions and current hotspot directions,while the other two fields are highly correlated and lower,and the research focus has changed in different time periods.According to the three trend directions of topics,all topics are clustered into five types: mainstream topics,rising topics,falling topics,non-mainstream topics and outbreak topics.Nonmainstream topics,declining topics and growing topics are the main types of research topics.(2)In terms of paper tracking degree: the short-term hotness tracking degree of papers in the three fields fluctuates greatly,while the correlation between mainstream direction and long-term direction is more obvious than other indicators.The percentage of non-mainstream topics is higher in all three fields,but the non-mainstream topics contribute fewer papers,i.e.,the non-mainstream topics are in the margins of most paper studies.In contrast,mainstream topics have a lower share and contribute papers at a higher rate than other types of topics.(3)In terms of scholar tracking degree: the trend of indicators shows that scholars in the three fields prefer research topics that have become mainstream or classic.Scholars were clustered in five dimensions: mainstream tracking degree,short-term hotspot tracking degree,long-term hotspot tracking degree,persistence degree,and number of papers,and six categories were categorized: general scholars,perseverers,pioneers,followers,producers,and sailors.General scholars,sailors,and producers are the common types of researchers.The follower category has the highest percentage of low-cited scholars,who are more inclined to conduct conventional and basic research,while the two groups of producers and perseverers have the highest percentage of highcited scholars,who are more inclined to pursue innovative and pioneering research directions.Long-term hotspots and short-term hotspots show a strong positive correlation with citation volume,while mainstream tracking degree and persistence have a low correlation with citation volume.