**2. Knowledge Map Visualization Analysis of International Research Evolution of Artificial Intelligence in China**

A knowledge map is an image of a knowledge domain and shows the relationship between the development process and structure of scientific knowledge. Citespace visualization software is one of the representative tools used for the visualization of scientometrics and knowledge map [69–71]. Through an international analysis of Chinese scholars in the field of artificial intelligence and visual analysis of the knowledge maps of English articles, this study explored the network structure and evolution of related research hotspots and keywords of Chinese scholars in the field of AI [69–72]. By analyzing the hotspots and keywords of international research in a certain field, scholars can provide necessary reference and identify implications for the development direction, policy formulation, knowledge base, and frontier trends of the field [73–76]. In addition to scientometrics and knowledge mapping, many scholars and experts also use other algorithms and technologies to study AI, such as data extraction, features fusion, and classification and recognition technologies [77–84]. Based on these studies, from the perspective of evolution and cooperation, this study used the methods of scientometrics and knowledge map visualization to research, which is helpful for further enrichment of the field and gives this research a certain uniqueness and novelty.

We chose to download and obtain the corresponding literature data from the "Web of Science Core Collection" of ClarivateAnalytics's Web of Science database (http://www.isiknowledge.com/). Most of the research team members were members of the China AI Development Report 2018 report research group for policy analysis, and they were not very familiar with many specific AI technologies due to the wide range of artificial intelligence research. To ensure that the literature was pertinent and reflected the targeted situation, this study only selected "Artificial Intelligence" as "Topic", "China" as "Address", and "2009–2018" as "Timespan" in the research literature in the Web of Science Core Collection, so as to carry out a comparative analysis of China's national-level AI policies between 2009 and 2018. Therefore, the sample selection of this study had certain limitations, and the literature trend chart (Figure 1) had a certain one-sidedness and misleading nature; it only showed that the research on AI by Chinese scholars is increasing year by year, and that research literature presents a certain significant growth trend with policy encouragement and financial support.

**Figure 1.** The annual number and stage distribution of China's international artificial intelligence (AI) research (2009–2018).

As mentioned above, this paper selectd the relevant papers of Chinese scholars from the core collection of ClarivateAnalytics's Web of Science database from 2009 to 2018, and used "Artificial Intelligence" and its key keywords for searching and selecting. A total of 3746 related papers were collected, and the three types of papers (1762 articles, 1830 proceedings papers, and 112 review; 3577 in total) which best reflected the international research of artificial intelligence were selected and served as samples for data and visualization analysis in this study. After sorting and screening, 3577 articles were collected for further analysis.

#### *2.1. Distribution of AI International Research Results*

The growth regularity and trend of papers published in international high-level journals as well as conferences are important indicators of knowledge accumulation or change in research fields [26]. After screening the topic paper data of China's AI international research, classifying and sorting them by year, the number and change trend of Chinese research papers on AI in 2009–2018 was obtained, as shown in Figure 1. Figure 1 has two coordinate systems. The left coordinate system corresponds to the blue column (number of papers published in that year), and the right coordinate system corresponds to the red curve (cumulative number of papers in that year).

It can be seen from Figure 1 that although the publication of international AI research papers fluctuates, in general, the growth with time showed a trend of increasing year by year. The cumulative number of papers over the year showed a very stable year-on-year growth trend, and the growth process was divided into three stages: (1) Stage 1 is the initial exploration stage (2009–2011). The number of relevant papers in this stage was around 100–200, among which the number of papers in 2009 was relatively large, and the overall situation was relatively stable. (2) Stage 2 is the steady rising stage (2012–2015). The number of papers was growing at this stage, fluctuating slightly, and the number of papers gradually increased from more than 200 to more than 300. (3) Stage 3 is the rapid development stage (2016–2018). The number of papers in this stage grew relatively rapidly, from 300 or 400 in 2015–2016 to 777 in 2018.

Regarding the division of the literature into these three stages: on the one hand, the division was based on the sorting and growth trend of the literature in the studied years, and, on the other hand, it was based on the three-stage division of China's national-level policy in 2009–2018 in the China AI

Development Report 2018 [19], so as to use the keywords of the three-stage literature and the keywords of the three-stage policy documents for comparative analysis.

### *2.2. Distribution of High-Yield Countries*/*Regions and Institutions with AI International Research Cooperation*

Based on the relevant international research articles on AI, the refinement statistics of "Countries/Regions" represented the countries and regions that have cooperated with Chinese institutions or scholars to conduct AI research. The number and percentage of total publications from these countries or regions in cooperation with Chinese scholars is shown in Table 1. From Table 1, the top 15 aside from China were the United States (250 articles, accounting for 6.99%), the United Kingdom (82 articles, accounting for 2.99%), Australia (68, 1.90%), Canada (62, 1.73%), Singapore (52, 1.45%), Japan (42, 1.17%), France (24, 0.67%), Iran (23, 0.64%), India (19, 0.53%), Italy (18, 0.50%), Spain (16, 0.45%), Saudi Arabia (15, 0.42%), South Korea (15, 0.42%). Among them, the United States was the first group, with 250 papers in cooperation; the second group was concentrated in other developed countries and regions in the field of AI in other continents, such as the United Kingdom (traditional powers), Australia in Oceania, Canada in North America, Singapore in Southeast Asia, etc.

**Table 1.** The high-yield countries/regions distribution of China's AI international research cooperation (2009–2018).


Based on the relevant international research articles on AI, the refined statistics of "Organizations" were calculated for countries and regions that cooperate with China, as shown in Table 2. The first 15 universities in the first group began with the Chinese Academy of Sciences (331 articles), in which research benefits from the Chinese Academy of Sciences' long-standing research in this field, and the huge research institutes under the Chinese Academy of Sciences. The second group included China's Ministry of Education and China's high level universities, such as the Ministry of Education of China (140 articles), Tsinghua University (111 articles), the University of Chinese Academy of Sciences (101 articles), the Hong Kong Polytechnic University (99 articles), Beihang University (96 articles), Zhejiang University (81 articles), Wuhan University (76 articles), Huazhong University of Science and Technology (73 articles), and Shanghai Jiaotong University (72 articles), which were all over 70. Among them, traditional science and engineering and defense science colleges had an advantage, and many other comprehensive universities also had good performance.


