Researchers are paying a lot of attention to artificial intelligence (AI). The development of AI in research and its contributions to other fields remain largely unexplored, nonetheless. We have identified a total of 435994 papers across all disciplines that involve AI methods based on 26 millions articles in the Web of Science database. Using bibliometrics and visualization techniques, the evolution process of AI technology in scientific research is identified. The results of the study show that the growth patterns of AI technology are consistent with the fundamental properties of emerging technologies, characterized by rapid growth, new applications, and changes in size.
The impact of technology on society is profound. It has transformed not only the way people live, but also affected every branch of science. Significant publicity has been given to AI and its application in science, the renewed scholarly interest in AI, including the bibliometrics community.
The goal of this study is to reveal how does AI assist scientific research domains, and to provide guidance for future research efforts. We conducted a search of 26 millions AI articles from 2000-2019, limiting the study to those that utilized AI technologies. We reveal different perspectives on how AI assists the scientific research domains and the behavioral patterns exhibited by disciplines following AI trends.
Our data source is the Web of Science (WoS) database, since the subsequent search requires search term matching for abstracts, the abstracts were processed for missing values, and after removing the articles with missing abstracts, a total of 26,408,350 articles were obtained. We used AI methods provided by the Papers with Code platform as a search term list to match article abstracts. Papers with Code provides a total of 2060 AI methods. To be conservative, we removed abbreviations as well as one-word terms and ensured that only unambiguous terms were included to prevent errors in matching. In addition, we retained the 50 most common AI terms. In our study, the final list of search terms contains 1960 AI methods.
For our analysis, we used the collection of articles from WOS 2000-2019, and we conducted three search sessions on the dataset. In the initial search, we performed search term matching on abstracts using the findall method in Python, requiring at least one search term to be present in the article abstract. In the second round of search, we removed data that only appeared in methods such as logistic regression and linear regression. In the third round of search, we performed manual verification to ensure that AI methods were mentioned in the article abstracts, and we manually removed some articles with ambiguous terms. Finally, we identified a total of 435,994 articles.
We collected a total of 435,994 articles on AI research, accounting for 1.6% of all articles. AI has been a hot topic in recent years, but our data shows that the application of AI technology is not yet common in most fields.
A total of 449 AI methods appear in the AI research articles we collected, and Figure 1 shows the distribution of AI method vocabulary counts. Most of the AI method vocabulary occurrences are concentrated between 0 and 10. However, the percentage of terms decreases as the number of occurrences increases. Although a large number of AI methods have emerged, the AI methods used in the articles we collected are still few and concentrated in a few common methods.
250 243 200 s d o teh150 m I A f o r eb100 m u N 50 0 108 22 9
Research output has increased not only in absolute numbers but also in the total number of papers equivalent to a given scientific field, albeit at a lower level. AI articles account for 1.6% of all papers, meaning that AI articles still represent only a small fraction of the overall research volume. However, recent growth rates in these shares are remarkable.
Our data confirm the explosive period of research activity in all scientific fields (Figure 2), with a high growth rate of AI articles around 2005, around 15%, then experiencing some decline around 2010, then gradually recovering and growing steadily. The growth rates rapidly increased after 2017, reaching 30% by 2019.
We use bibliometric and visualization methods
to study the assistance of AI technologies in
different disciplines from 2000-2019. Although AI
research has been receiving much attention in recent
years, it is still not applied in most research areas.
Our study found the same trend in the growth rate of
the number of AI articles in all disciplines. AI is an
emerging technology and the growth trend is in line
with the 'double-boom' cycle proposed by Schmoch
[