Page 121 - Journal of Library Science in China, Vol.45, 2019

120 Journal of Library Science in China, Vol.11, 2019

defined citation intensity as the “the number of citations cited or mentioned in the body of a
citing literature”. He calculated citation intensities of 10,382 citations in 350 academic articles in
the Journal of Informetrics, and found that the number of references and the citation intensity in
academic articles were power-law distribution. Meanwhile, he compared the number of references
and citations to illustrate universality of multiple citations. Hassan, Akram, and Haddawy (2017)
randomly selected 465 articles from the Association for Computational Linguistics anthology,
and analyzed 106,509 citation contents. They pointed out that a reference was more important to
a literature if it was cited more times in the literature. S.B. Liu and K. Ding (2013) proposed an
evaluation metric for evaluating citation quality, namely:
Q citation = specific citation frequency in literature/citation frequency in references
They held that higher Q citation means the higher citation quality.
Based on the existing research, this paper introduced the concept of book citation intensity,
which is the ratio of the number of times a book was cited to the number of citing literatures, so as
to analyze distributions of citation intensity in different disciplines.

1.3 Related work about citation length and citation sentiment

Citation length is the string length of a citation sentence. It can reveal citation characteristics in
each discipline by exploring distributions of citation lengths. C.Z. Zhang, Wang, and C. Lu (2017)
analyzed citation behaviors of 39 English monographs. They found that the average length of
citation content was mainly distributed in 100-200 characters, and longer citations were mainly
distributed in the first half of a monograph. Meanwhile, there were obvious differences in the
average length of citation content in different disciplines.
Citation sentiments of citation content indicate authors’ attitudes to references. Obviously,
positive attitudes and negative attitudes have different effects on evaluation of book impacts. S.B.
Liu and K. Ding (2013) classified citation sentiments into positive citations, negative citations
and neutral citations. They used clue words to annotate sentiments of 147,817 citations in BMC_
Bioinformatrics. The results showed that 62.88% of citations were neutral citations, and only 3.53%
were negative citations, which indicated that researchers prefered to take positive attitudes toward
references. Athar and Teufel (2012) selected 1,741 citations of 20 articles in the ACL Anthology
and used different methods to conduct sentiment annotation. They found that the number of
negative citations was increased by three times in citation contexts compared with citation contents
only, which indicated that ignoring citation contexts might lose sentiment information. Abu-Jbara,
Ezra, and Radev (2013) adopted supervised learning methods to conduct sentiment annotation
on 3,500 citation contents in 30 academic papers in the ACL Anthology Network corpus. As
more than half of the citations were neutral, they used a two-step method for classification. In the
first step, the citations were classified as subjective and objective. In the second step, subjective
citations were divided into positive ones and negative ones. They found this method was more