*2.3. Research Hotspots of China's AI International Research*

In this study, the topics and keywords in the literature were used to explore the hot topics and hotspots of Chinese scholars in the field of AI in 2009–2018. It is generally believed that keywords are highly condensed and concise in the topic and research hotspots; with the assistance of the knowledge map of co-occurrence keywords drawn by CiteSpace software, this study quickly elucidated the structure, network distribution, and frequency of co-occurrence of related articles keywords, so as to clarify the research hotspots in the field [73–76]. Through the visualization analysis of 3577 related papers collected in this paper, a knowledge map of the international research hotspots of China's AI by node type and time-zone type was obtained. The two maps are shown in Figures 2 and 3.

**Figure 2.** The knowledge map of China's international AI research hotspots (node type) (2009–2018).

**Figure 3.** The knowledge map of China's international AI research hotspots (time-zone type) (2009–2018).

The keywords in the knowledge maps shown in Figures 2 and 3 represent the nodes in the network, and the connections between the network nodes represent their co-occurrence relationships. The larger the nodes, the more rings, indicating that the co-occurrence frequency of the keywords is higher. Figure 2 shows that besides the node of "artificial intelligence" (673 times), the nodes with high frequency were "neural network" (307 times), "model" (261 times), "system" (244 times), "optimization" (202 times), "genetic algorithm" (172 times), and "cloud computing" (97 times), etc. Furthermore, new keywords and research hotspots of "China" (32 times), "extreme learning machine" (29 times), "big data" (29 times), "face recognition" (25 times), "electronic skin" (28 times) and other nodes emerged 2015. By reviewing the keywords and subject words that have appeared over the years, the hotspots and interests of Chinese scholars' research on AI can be understood. In the initial exploration stage of AI, Chinese scholars relied on "artificial neural networks" and "genetic algorithms" to conduct research and gradually formed models and systems, and then transferred focus to "cloud computing" and "data sharing" in the steady rising stage. In recent years, hot topics such as "China", "big data", "face recognition", and "electronic skin" have been closely related to the frontier areas of AI, or closely related to current policies and social concerns. In addition, some experts mentioned new functions of AI applications in different fields, such as pattern tracking, data classicization, neural network, deep mining, etc. [85–91], which are worthy of further attention in this study. The organic connection and quantitative comparative analysis between the hot spots of AI international research in the past ten years and policy keywords at the national level is discussed in the following sections.

#### **3. Quantification Analysis of China's AI National-Level Policy Documents**

#### *3.1. Stage Division of China's AI National-Level Policy*

The quantification analysis of policy documents is based on the analysis of policy documents; on the basis of collecting policy documents and constructing subject words, it further integrates the information contained in the documents, and carries out co-occurrence analysis, clustering analysis, and trend evolution analysis of relevant information, keywords and subject words from multiple dimensions and multiple angles. Starting from the quantitative law of inductive document attributes, it puts forward deeper policy suggestions and reflections based on the combination of qualitative and quantitative research [92–95].

This study established a dataset of AI policy analysis. First, based on the word frequency statistics of AI documents, it identified and constructed the list of initial AI keywords, which were mainly about "AI", and "deep learning", "big data", and "cloud computing", etc. The research team further asked experts in the corresponding field to judge and screen the keywords. Based on the keywords, the pre-search was conducted in the Government Documents Information System (GDIS) of the School of Public Policy and Management of Tsinghua University, and the title and text of policy documents containing the above-mentioned search terms were selected and screened. A total of 262 pieces of China's central-level or national AI policy documents were retrieved. The relevant policy documents were initially sorted and classified according to their years of publication, and the policy trend as well as a stage division map was drawn (shown in Figure 4).

**Figure 4.** The trends and stages of China's national-level AI policy (2009–2018). Source: According to References [19,51].

The collection of policy documents and reports, especially the China AI Development Report 2018 and Domestic and Foreign AI Policy Analysis Report 2018 was combined with the three-stage distribution of AI international research hotspots discussed in Section 2. In order to maintain the consistency and contrast of the analysis, this study roughly sorted the national AI policy into three groups: (1) the initial exploration stage of 2009–2012, where there were fewer national-level policies for AI; (2) the steady rising stage of 2013–2015, when the development of AI gradually increased to the level of national strategy, and policy documents were issued more quickly and steadily; and (3) the rapid development stage of 2016–2018, representing the upsurge of AI development, in which the national-level AI policies became more comprehensive, the top-down design was further enhanced, and the follow-up policies and plans at all levels in various fields were more targeted and specific.
