We searched for research papers in the field of Library Science & Information Science on the Web of Science (WOS) platform through the “advanced search” function with the search formula “WC = INFORMATION SCIENCE LIBRARY SCIENCE”, the type of literature was limited to “Article, Review and Proceedings Paper”, and the year was limited to 2016-2020. 25,153 research papers in the field of Library Science & Information Science were retrieved, and the search period was February 2021.

The common sources of Altmetrics metrics data are ImpactStory, Altmetric.com, Plum Analytics, PLOS ALM and other databases, among which Altmetric.com has the most specific evaluation objects and relatively complete and diverse data sources [7], so this study uses Altmetric.com to collect the Altmetrics metrics data of the papers. The DOI numbers of 25,153 papers exported from the WOS platform were searched in Altmtrics.com, and a total of 12,466 papers with Altmetrics index data were obtained.

The Web of Science platform identifies papers in the “WC” field according to its disciplinary classification system, and each paper belongs to one or more disciplines, and this disciplinary classification system is mostly used in previous studies to determine the disciplinary affiliation of papers [8]. Therefore, this study defines papers involving two or more disciplines at the same time as interdisciplinary research papers based on the disciplines marked in this platform. Among 12,466 papers in the field of library and intelligence, a total of 8,541 papers involved both library and intelligence disciplines and other disciplines (including computer science, management, communication, medical informatics, etc.), and the number of interdisciplinary research papers reached 68.51%.

The descriptive statistics of 17 Altmetrics indicators of 8541 interdisciplinary research papers were shown in Table 1. In terms of concentration trends, the 75%

quartiles of most indicators were 0, and the median was more different from the mean, indicating that the concentration of these indicators was low. In terms of dispersion, the standard deviation of each indicator was much larger than the mean, and the extreme deviation is also larger, indicating that the data dispersion was very high. From the dispersion degree, the standard deviation of each indicator was much larger than the mean value, and the extreme deviation is also larger, indicating that the data were very discrete. In terms of the distribution pattern, the skewness of each indicator was greater than 0, and the kurtosis is greater than 3, showing a long-tailed right-skewed spike curve. In general, the non-zero scores of most indicators were concentrated in a very small number of literature, which obviously did not obey normal distribution.

The coverage ratio in Table 1 was the ratio of the number of papers with non-zero values under a certain indicator to the total number of papers, which could reflect to some extent the value of the indicator available in the evaluation of papers, and also to see where the influence of papers mainly comes from. The seven indicators with the highest coverage of interdisciplinary research papers in the field of Library Science & Information Science were, in order, Mendeley readership, Citation, Twitter mentions, Facebook mentions, Blog mentions, News mentions, and Policy mentions. Therefore, these seven indicators were chosen to analyze the influence of interdisciplinary research topics.

After importing 8541 interdisciplinary research papers into Incites, the micro themes of each paper were identified through the “citation theme” function, and a total of 587 micro themes were obtained, among which the 20 micro themes with the largest number of papers are shown in Table 2. The themes were listed in descending order of the number of papers. The keywords were extracted from 4634 papers for word frequency statistics, and a total of 9312 keywords were obtained, among which 98 high-frequency keywords with a word frequency of 14 or more. In this study, a “topic-high-frequency keyword” (20 rows * 98 columns) matrix was constructed to calculate the TF-IDF value of each high-frequency keyword on each topic, and then the three keywords with the highest TF-IDF values were extracted from each topic to characterize the content of each topic. For example, the main feature keywords of theme 1 “bibliometrics” were Bibliometrics, Altmetrics, and Citation Analysis, and the theme names and feature keywords show a strong consistency, which further verified the reasonableness of “citation themes” in classifying the topics of papers.

The rank mean values of Altmetrics indicators of the 20 research topics were counted, and the results were shown in Table 3. To further analyze the influence composition characteristics of each research topic, the data of the 7 indicators in Table 3 were imported into Matlab, a matrix of 20 rows and 7 columns was built, and the SOFM neural network was trained with SOM Toolbox: the input layer was set to 7 neurons, corresponding to the 7 research topic influence indicators; the competitive layer is set as a 2 × 2 two-dimensional structure, all research topics were classified into 4 categories by 4 neurons. The number of iterations (epochs) was set to 1000, and the final training results of the SOFM network were shown in Figure 1.

The neurons were represented by blue hexagons in Figure 1, and the neurons corresponding to each of the 20 study topics were labeled. The connection color between neurons indicated the distance: the closer the connection color between neurons was to yellow, the closer the distance between two neurons, the more

similar the influence distribution characteristics of the study topic; the closer the connection color between neurons was to black, the more distant the distance between two neurons, and the more different the influence distribution characteristics of the study topic. The connection between the first neuron and the third neuron was black, which means the influence of the two corresponding research topics was very different; the connection between the first neuron and the second neuron was yellow, which means the influence of the two corresponding research topics was closer.

According to the results of clustering of SOFM network, the influence characteristics of 20 research topics could be divided into 4 categories (see Table 4), namely, comprehensive attention hotspots, online social hotspots, academic research hotspots, and potential attention hotspots.