The highly cited articles in each stage that counted by CNKI database were showed in additional file [see Additional file 1]. The deadline for citation counting was February 2, 2020. The citation frequency of high-cited articles varied greatly in each stage, and the theme represented by articles also varied greatly. In the 1980s, the most cited articles mainly focused on Anopheles; in the 1990s, the themes increased, but the antimalarial drugs and vectors were the main ones; in 2000s and 2010s, there were more citation to the epidemic analysis; but in the past ten years, retrospective and summary research articles received more attention.
Keyword frequency analysis
The word clouds of different stages (Fig. 3) reveal the following features: 1) falciparum malaria, vivax malaria were the main types of malaria in China; 2) Anopheles sinensis was main malaria vector; 3) imported cases, surveillance and elimination had come at the forefront of concerns in the fourth stage.
For the analysis of single keyword, the keyword ‘Plasmodium falciparum’, it maintained the fourth rank in the first three stages, but its’ K/A ratio had been in a state of decline, 17.05% in 1980s, 13.04% in 1990s, 9.86% in 2000s. The keyword ‘falciparum malaria’, it’s frequency rank rose from seventh in 1980s to fourth in 2010s, but its’ K/A ratio declined from 12.97% in 1980s to 7.82% in 2010s. These results suggested that there were some common patterns in the change of K/A ratio.
The heat map (Fig. 4a) showed two obvious change pattern of K/A ratio from an overall perspective. It clearly showed that the K/A ratio of some keywords continuously decreased, and some continuously increased (Fig. 4b and Fig. 4c). Keywords in continuous decrease of K/A ratio were ‘internal medicine’, ‘Plasmodium falciparum’, ‘vivax malaria’, ‘falciparum malaria’, ‘Anopheles sinensis’, ‘antimalaria drugs’, etc. Keywords in continuous increase included ‘imported case’, ‘surveillance’, ‘artemisinin’, ‘floating population’, ‘epidemiological characteristic’, ‘elimination’, etc. In essence, these two patterns were the manifestation of the research theme change.
Keywords co-occurrence network
Figures 5, 6, 7, and 8 were the maps of keyword co-occurrence network in four stages.
According to the strength of the co-occurrence relationship, the keywords in the four stages were respectively divided into 5 clusters in 1980s (Fig. 5), 6 clusters in 1990s (Fig. 6), 5 clusters in 2000s (Fig. 7), and 7 clusters in 2010s (Fig. 8). These clusters were considered as research themes, and each theme could be divided into sub-themes according to the subjects that were represented by specific keywords in the cluster.
From the perspective of whole network maps of four stages, in 1980s (Fig. 5), the blue cluster was centered on ‘Anopheles’ and included keyword ‘Anopheles sinensis’, ‘Anopheles anthropophagus’ and ‘Anopheles minimus’. And the same situation occurred in 1990s (Fig. 6, green cluster). This two clusters had obvious boundaries with other clusters in their own networks. This result indicated that researches on Anopheles had high independence. But in 2000s, this kind of independence became weak. And in 2010s (Fig. 8), this kind of independence was disappeared. The boundaries of clusters were difficult to identify. For the analysis of the network structure among the clusters, the boundaries between two clusters in one stage became more and more blurred, especially in 2010s (Fig. 8). Figure 8 also showed that many main nodes in one clusters are also intermediaries with other clusters. This result suggested that the relationship between research themes is no longer the weak connection due to sub-themes’ co-occurrence in the past, but a strong connection that emerged from the deep integration of the subjects and research methods.
From the perspective of the cluster in different network maps, in the blue cluster in 1980s (Fig. 5), the peripheral keywords, which were around the central keyword ‘Anopheles sinensis’, included ‘retention spray’, ‘ecological habit’, ‘life history’, etc. However, in the blue clusters in 2010s (Fig. 8), peripheral keywords around the central keyword ‘Anopheles sinensis’, included ‘surveillance’, ‘drug resistance’, etc. It was found that the keywords that represented the research object, such as ‘falciparum malaria’, ‘Plasmodium falciparum’, ‘Anopheles sinensis’, were always the central keywords in different stages. But peripheral keywords, which represented the research fields, around central keywords changed. For ‘Anopheles sinensis’, in 1980s, the research direction was entomology. In 2010s, the research direction was insect vector control. So this result indicated that the research direction around the central keywords was changing with the process of malaria elimination.
For the analysis at node level, in all four keyword co-occurrence network maps, some nodes in one cluster only had co-occurrence with the nodes inside this cluster, and other nodes had co-occurrence with multiple nodes outside this cluster. Under this common feature, there were differences in details, such as the link density between nodes. In Fig. 7, we clearly observed that the network of green clusters looks more complex than the red clusters on the premise that the number of nodes is not much different between the red clusters and the green clusters. This result indicated that the co-occurrence between the nodes in the green cluster was more divergent, while the co-occurrence relationship between the nodes in the red cluster was more directional. This meant that the sub-themes represented by the nodes in the red cluster have a high degree of independence.