3.2. Geographic Distribution
Publication performances of all the retrieved papers are from 122 countries/territories, and the top ten countries with the most publications are shown in
Figure 3. China has contributed with 3174 (38.75%) papers on AI, which makes China top ranking country in this list, followed by the USA (1509, 18.42%), the UK (830, 10.13%), Germany (563, 6.87%), India (431, 5.26%), Singapore (361, 4.41%), France (355, 4.34%), Australia (338, 4.13%), Spain (321, 3.92%) and South Korea (308, 3.76%).
The countries with most of the contributions are China, the USA and the UK, whereas Huazhong University of Science and Technology, Chinese Academy of Science and Nanyang Technological University were the three institutions with the most publications in AI research. Huazhong University of Science and Technology ranked first with 322 papers, followed by Chinese Academy of Sciences with 246 papers. Other leading institutions in the area are Nanyang Technological University (221), University of Chinese Academy of Sciences (126), Massachusetts Institute of Technology (111), Zhejiang University (110), Xi’an Jiaotong University (109), Tsinghua University (105), National University of Singapore (97) and Shanghai Jiaotong University (96).
Publication trends and cumulative numbers of publications for the main countries are shown in
Figure 4. The number of publications in the USA is larger than that in China and the UK in the period from 1979 to 1994. The records of the USA have grown rapidly recently and rank in the first place now. China also performed well in 1995. The amount of published literature once surpassed the USA and the UK. After 2015, the number of publications from China grew rapidly, far surpassing those in the USA and UK. Until 2021, the number of publications in China was far ahead in the world. The cumulative numbers of publications in all three countries have shown a trend of rapid growth since 2015. The cumulative number of publications in the three countries and the world shows the same trend.
The cooperation network of authors in AI with ≥70 papers are shown in
Figure 5. The size of labels and nodes show the number of papers that the authors published, and the link means a cooperation between two authors. The major groups are marked and show that organizations from China, the USA and the UK play an important role in the international cooperation in AI research. On the other side, the organizations from China collaborate actively with other countries in the network.
3.4. High Frequency Topic Terms
According to literature on AI published in Web of Science from 1979 to 2021, VOSviewer software was used to perform cluster visualization analysis of terms based on keywords in a co-occurrence relation. The network visualization of topic terms with a frequency ≥ 100 are shown in
Figure 7. The figure shows multiple circles of different sizes and colors. The size of the font and circles reflects the frequency (weight) of each term, and the color of the circles represents the cluster of the topic terms. The distance between two circles indicates the relatedness between two circles. A short distance means a strong connection, while a long distance means a weak connection.
The results of keyword analysis in the period 1979–2021 are shown in
Table 1. We set the threshold of frequency as 100. The keywords with a frequency ≥ 100 are listed in the table below. The total link strength (TLS) is also shown in the table. It indicates total co-occurrence times of keywords and other keywords The keywords with high occurrence in the database are classified by VOSviewer in three cluster. The table below shows the occurrences and TLS of keywords in each cluster.
The topic of publications and keywords with occurrences and total link strength (TLS) in three periods are shown in
Table 2,
Table 3 and
Table 4. We conducted a cluster analysis of keywords from 1979 to 1994. The threshold number of minimum keyword occurrence is set as 5, and 30 words meet the threshold of the 618 words. The results of keyword analysis are shown in
Table 2. The keywords occurring the most in the articles from 1979 to 1994 are ‘artificial intelligence’, ‘expert systems’, ‘simulation’, ‘expert system’ and ‘design’.
For the articles in the 1995–2007 period, the threshold number of minimum keyword occurrences is set as 20, and 33 words meet the threshold of the 4614 words. The results of keywords analysis are shown in
Table 3. The keywords occurring the most in the articles from 1994 to 2007 are ‘artificial intelligence’, ‘design’, ‘neural networks’, ‘simulation’ and ‘optimization’.
In the last period, the threshold number of minimum keyword occurrences is set as 100, and 29 articles meet the threshold of the 23,515 words. The results of keyword analysis from 2008 to 2021 are shown in
Table 4. The keywords occurring the most in the articles from 2008 to 2021 are ‘artificial intelligence’, ‘optimization’, ‘design’, ‘machine learning’ and ‘model’.
The number of topic terms increased over the years. There are 618 keywords in the articles published in the 1979–1995 period, and there are 4614 keywords in the 1995–2007 period. After that, the number of keywords increases rapidly. In the 2008–2021 period, the number of keywords was about four times that of the previous period. The increase of keywords indicates the focus on AI and manufacturing research greatly increased with time. After that, we extracted and classified the topic of keywords to clarify the development trend of AI.
The keywords of the literature from 1979 to 1994 were classified via VOSviewer. We set the threshold of keyword occurrences as 5, and the 30 words that meet the threshold were classified via VOSviewer in five clusters. Each cluster includes the similar keywords, and their occurrences and TLS are also listed in
Table 2.
The keywords with high frequency in the literature from 1995 to 2007 are displayed in the above table. We set the threshold of keyword occurrences as 20, and 33 words meet the threshold and were classified via VOSviewer in five clusters. Each cluster includes the similar keywords, and their occurrences and TLS are also listed in
Table 3.
Table 4 indicates the occurrences and TLS of the keywords with high frequency. We set the occurrences threshold of the keywords as 100, and 29 keywords meet the threshold and were classified via the VOSviewer.
The analysis consisted of the clustering of keywords with occurrences in
Table 5,
Table 6 and
Table 7. The corresponding topic maps are shown in
Figure 8. The results of the extracted topics showed that the research in the first period (1979–1994) mainly focused on ‘machine learning’, ‘data system’, ‘programming design’, ‘flexible manufacturing’ and ‘neural networks’. The topics extracted are shown in
Table 5. In
Table 6, the keywords in the publications mostly focused on ‘machine learning’, ‘genetic algorithms’, ‘expert systems’, ‘neural networks’ and ‘concurrent engineering’ in the second period. Between 2009 and 2021, the keywords in the publications mostly focused on the topics ‘internet of things’, ‘smart manufacturing’, ‘deep learning’, ‘digital twin’ and ‘neural networks’. The topics extracted are shown in
Table 7. Each topic corresponds to multiple keywords. The sums of their occurrences are listed in the adjacent column.
We grouped the similar keywords into the same topic in the period 1979–1994. These keywords with high frequency were divided into five topics as shown in
Table 5. We summed the occurrences of the keywords in the same topic and listed them in the above table.
As shown in
Table 6, the similar keywords are divided into the same topic from 1995 to 2007. We summed the occurrences of the keywords in the same topic and listed them in the above table. ‘Machine learning’ is the research hotspot in the period of 1995–2007.
As shown in
Table 7, the similar keywords are divided into the same topic in the period of 2008–2021. We summed the occurrences of the keywords in the same topic and listed them in the
Table 7. The topic with the highest occurrence is ‘internet of things’. It indicates the trend of artificial intelligence in recent years.
3.5. Highly Cited Articles
We used VOSviewer software to conduct bibliographic coupling analysis. The documents were selected as the unit of analysis. The counting method is full counting. We chose the minimum number of citations of a document and set the threshold as 300, and 28 articles meet the threshold. The overlap visualization is shown in
Figure 9. The retrieved articles cited more than 600 times and the corresponding total citations (TC) are listed in
Table 8. The most cited paper is “Machine learning: Trends, perspectives, and prospects” written by M. I. Jordan et al. in 2015, and it has been cited 2189 times since it was published in Science [
31]. On the other hand, “Deep neural networks: A promising tool for fault characteristic mining and intelligent diagnosis of rotating machinery with massive data” written by F. Jia et al. in 2017 ranks the second, which has been cited 987 times [
32]. The aforementioned articles are all in the field of deep learning. The next three highly cited papers mainly deal with big data processing techniques [
33,
34,
35]. It represents an interest shift in AI from simple logic operations to data mining and analysis, reasoning, judgment, conception and decision-making driven by big data.
We selected the first five articles of the highly cited articles in
Figure 9. We extracted the topics of the highly cited articles. The topics and TC of these articles are listed in
Table 8. The topics of these articles are focus on ‘Deep learning’ and ‘Big data’. This is in line with the trend of AI we discussed in the previous sections.