121 Research on the Evolution of the Scientific Collaboration Network and the Growth of the High-Impact Author in the Life Cycle Phase Wang Yuefen, Li Dongqiong, Yu Houqiang DOI: 10.3772/j.issn.1000-0135.2018.02.001 In order to further study scientific collaboration and reveal the law of scientific development, research on the evolution of the scientific collaboration network, especially of individual authors and component networks was conducted on the New Energy field of CNKI in China, based on the life cycle theory and by means of mathematical statistics and complex networks. Results show that except for the burgeoning phase, collaboration networks in the growing phase, booming phase, and transforming phase are scale-free networks and follow the power law distribution. Collaboration networks in each phase present different features. This study explores the growth characteristics of the top 10 high-impact scholars in the new energy field in different life cycle stages of the scientific collaboration network from three points of view: entering the collaboration pattern, the evolution model, and the network type of a high-impact author. Results show that the collaboration network in each phase presents distinctive features. Other important factors include the number of co-authors, whether the cooperation is with other high-impact authors, and the time interval when authors enter a new energy field. Four types of entering collaboration patterns are identified. Through the degree distribution of high-impact authors in different stages, four evolution patterns of individual high-impact authors are detected, namely a steadily growing pattern, growth and diminish pattern, keep leading pattern, and remain normal pattern. Three types of component networks are recognized, namely mobile collaboration component networks, leading growth collaboration component networks, and multiple core collaboration component networks. 2018 Vol. 37 (2): 121-131 [Abstract] ( 254 ) HTML (1 KB)  PDF (1561 KB)  ( 910 )

132 Predicting Research Collaborations Based on Network Embedding Zhang Jinzhu, Yu Wenqian, Liu Jingjie, Wang Yue DOI: 10.3772/j.issn.1000-0135.2018.02.002 In order to improve the efficiency and effect of predicting research collaborations in a large data environment, correlations among researchers should be learned and discovered automatically from massive datasets. Firstly, the co-authorship network is built from a massive dataset where research collaborations are indicated by co-authorship. Then, the researchers’ context in the network is learned by network embedding based on the deep machine learning method, and each researcher’s dense, low-dimensional vector is formatted. Finally, the semantic similarities among authors are calculated through the vector similarity indices for research collaboration prediction. Experiments in the field of Library and Information Science verify that the method can improve the accuracy and efficiency of research collaboration prediction. This method enriches and expands the information analysis methods based on complex networks from the perspective of data science. 2018 Vol. 37 (2): 132-139 [Abstract] ( 306 ) HTML (1 KB)  PDF (1992 KB)  ( 1087 )

140 Study of Scientific Tweet Author’s Behavior Pattern and Geographic Distribution Yu Houqiang, Wang Yuefen, Wang Feifei, Chen Bikun DOI: 10.3772/j.issn.1000-0135.2018.02.003 Statistical and visualization analyses were conducted on 2.63 million authors of 20.69 million scientific tweets, to reveal the authors’ behavior patterns including the number of scientific tweets, followed sources, and followed disciplines, as well as their geographic distribution at both a country and a city level. Results will provide reference for further understanding of the meaning of Twitter altmetrics and for future applications. Results show: (1) the distribution of authors’ productivity is highly skewed; 10% of the authors produced 80% of scientific tweets, and 91% of the authors tweeted no more than 10 scientific tweets. This means that most authors only occasionally disseminate and discuss academic products on Twitter. Meanwhile, there is a small percentage of extremely active authors. (2) Core sources that attract most authors, such as Nature, The Conversation and PLoS ONE, take up 6% of all publication sources and account for 77% of scientific tweets. Furthermore, 62% of authors follow only one source. (3) Disciplines that attract most authors are Medicine, General and Social Science, with 71% of authors following only 1 discipline while approximately 8% of authors follow over 3 disciplines. (4) Authors of scientific tweets are distributed all over the world, but are especially dense in USA and Europe. In East Asia, Japan is the most prominent country whereas in South America, Brazil is the most prominent. Authors are concentrated in cities like London and New York. These results show that Twitter altmetrics based on pure number of scientific tweets are not effective enough and future practical indicators need to combine author context as an important factor. 2018 Vol. 37 (2): 140-150 [Abstract] ( 222 ) HTML (1 KB)  PDF (2413 KB)  ( 665 )

151 Internationalization of References in Journal Papers: An Empirical Study on Library and Information Science Gong Kaile, Xie Juan, Cheng Ying, Meng Fansai DOI: 10.3772/j.issn.1000-0135.2018.02.004 In the context of the “Double First-Class” plan for Chinese higher education, the internationalization of discipline becomes an inevitable trend. The “Prosperity Project of Philosophy and Social Science (2011—2020),” which was formulated by Ministry of Education (MOE) and Ministry of Finance (MOF), also put forward the “go global” and “bring in” strategies. The internationalization of references is not only an important means to realize the strategy of “bring in” but also an indicator to measure scholars’ international vision. Therefore, this paper does an empirical study on Library and Information Science (LIS). Sixteen journals indexed by the Chinese Social Sciences Citation Index (CSSCI) since 1998 are selected to explore the internationalization of references in LIS journal papers and to reveal the scholars’ international vision. The paper then explores the factors affecting the internationalization of references and puts forward some specific suggestions for the internationalization of LIS in China. 2018 Vol. 37 (2): 151-160 [Abstract] ( 282 ) HTML (1 KB)  PDF (801 KB)  ( 695 )

161 Meta-Analysis of the Relationship between Individual Cognitive Absorption and Virtual Community Participation Zhang Ning, Yuan Qinjian, Zhu Qinghua DOI: 10.3772/j.issn.1000-0135.2018.02.005 Cognitive absorption is a term used to describe a state of deep involvement or participation. However, there are inconsistent findings regarding the relationship between individual cognitive absorption and virtual community participation according to current literature. The present study discusses the relationship between cognitive absorption and virtual community participation using the meta-analysis method and the random-effect model. Thirty-five studies, which included a total of 37 independent samples and 10210 participants, met the criteria for inclusion in the meta-analysis. The results showed overall that the correlation between cognitive absorption and virtual community participation was strong (r=0.433). Additionally, results indicated that the cognitive absorption measure used, the virtual community types, and subjects’ features all moderated the positive relationship between cognitive absorption and virtual community participation. The findings provide an accurate estimate of the relationship between cognitive absorption and virtual community participation, and can therefore guide future research. 2018 Vol. 37 (2): 161-171 [Abstract] ( 223 ) HTML (1 KB)  PDF (462 KB)  ( 1040 )

172 Design and Implementation of an Automatic Briefing System Based on the Work Thinking of Intelligence 3.0 Liu Ru, Zhang Huina, Du Liping, Li Menghui, Wu Chensheng DOI: 10.3772/j.issn.1000-0135.2018.02.006 An automatic briefing system makes intelligence services more real-time、accurate、and personalized. Not only can it present valuable information, but it can also support the daily decision-making of managers and communication connections between the intelligence department and the decision-making level. An automatic briefing system plays an important role in improving the service quality of intelligence 3.0. In this paper, the framework of an automatic briefing system based on the work thinking of intelligence 3.0 is designed, a management system is constructed, an automatic briefing model is defined, and an automatic generating path is described. Lastly, through case studies of briefing from scientific papers, we summarize the automatic briefing service paradigm based on the work thinking of intelligence 3.0. 2018 Vol. 37 (2): 172-182 [Abstract] ( 299 ) HTML (1 KB)  PDF (1772 KB)  ( 970 )

183 Research on the Automatic Word Segmentation of The Book of Songs under Multi-dimensional Domain Knowledge Wang Shanshan, Wang Dongbo, Huang Shuiqing, He Lin DOI: 10.3772/j.issn.1000-0135.2018.02.007 The Book of Songs is the earliest anthology of poetry in China: it is one of the thirteen classic books of Confucian tradition. The Book of Songs is ranked the first of the ancient canonical Five Classics. The Five Classics include Yijing (“Classic of Changes”), the Shujing (“Classic of History”), The Book of Songs, the Collection of Rituals, and the Chunqiu (“Spring and Autumn Annals”). The connotations of The Book of Songs are abundant, reflecting all aspects of social life in the Zhou Dynasty, such as labor and love, war and corvee oppression and rebellion, customs and marriage, ancestor worship and banquets, and even astronomy, geomorphology, animals, and plants. It is a mirror of Zhou Dynasty society, known as The Life Encyclopedia of Ancient Society. Moreover, The Book of Songs is the textbook of ancient Chinese political ethics, aesthetic education, and naturalism. With the extensive application of humanities computing, this paper combines the Sinological Index Series with the domain knowledge of the Mao Shi Index, and studies the automatic word segmentation of The Book of Songs using the machine learning method. Based on the corpus of the manual word segmentation of The Book of Songs, the method of combining the Guang Yun and statistical analysis was used to get 23 sets of feature templates that fuse different characteristics knowledge and then producing machine learning segmentation model by training. The performance of each word segmentation model is analyzed, and it is found that lexical features have the greatest influence on the word segmentation effect of The Book of Songs, and the harmonic mean F value of the word segmentation model can be up to 97.42%. Finally, the paper uses the domain glossary of the Mao Shi Index to carry out the post-processing of the long word correction with the test performance optimum segmentation model, and obtains the word corpus of The Book of Songs that fuses the expert vocabulary knowledge of the Mao Shi Index. This article integrates knowledge into the multi-dimensional domain to realize the automatic segmentation of The Book of Songs, which provides reference for the related research of the Pre-Qin poetry. Moreover, it inspires the study of the automatic word segmentation of Pre-Qin Classics. The word corpus of The Book of Songs, as part of the Pre-Qin Classics word corpus, has a supporting role to further realize the knowledge mining of the Pre-Qin Classics. 2018 Vol. 37 (2): 183-193 [Abstract] ( 279 ) HTML (1 KB)  PDF (872 KB)  ( 745 )

194 Deep Neural Networks Language Model Based on CNN and LSTM Hybrid Architecture Wang Yi, Xie Juan, Cheng Ying DOI: 10.3772/j.issn.1000-0135.2018.02.008 The language model is one of the most important domains in natural language processing. It is a bridge for the computer to identify and comprehend human language, and it is also a sign of Artificial Intelligence development. The language model is popular in Speech Recognition, Machine Translation, Information Retrieval, and Knowledge Mapping. With the rapid expansion of technology and hardware, the language model has experienced a transformation from statistical model to neural network model and then to the deep neural network model. The wide application of depth learning makes language modeling more extensive, complex, and expensive. This paper combines the personalized input, convolutional neural network (CNN) coding, and the technique of union gate, cooperating with long short-term memory (LSTM) mechanism to improve the language model. The dynamic integration of LSTM and CNN is called Gated CLSTM. In the experiment, we used the deep learning framework Tensorflow to achieve a Gated GLSTM architecture. Besides, some classical optimization techniques, such as noise contrastive estimation and recurrent projection layer, were adopted in the experiment. We tested the performance of the Gated CLSTM under an open and big scale corpus set and trained a signal-layer model and a three-layer model to observe how network depth influences the performance. The single-layer model has 4 days of training experience and reduced the perplexity to 42.1 in four GPU console environment. The three-layer model reduced the perplexity to 33.1 in 6 days. Compared with some classical benchmark models, significant improvements have been made by Gated CLSTM considering both hardware and time complexity and perplexity. 2018 Vol. 37 (2): 194-205 [Abstract] ( 206 ) HTML (1 KB)  PDF (1368 KB)  ( 3595 )

206 Mechanism and Empirical Research on Forecasting Influenza Epidemic Fused with Baidu Index Wang Ruojia DOI: 10.3772/j.issn.1000-0135.2018.02.009 This study explores the internal mechanism and possibility of forecasting an influenza epidemic based on both search queries and actual influenza data. First, the logical relationship is explored between online information searches and conventional surveillance data based on the concepts of information behaviors, information seeking behaviors, and so on. Then, the range selection method and cross-correlation analysis are used to select keywords according to the theoretical framework. Finally, three models are established and compared. The results show that (i) the empirical research proves the logical rationality of the theoretical framework: the keywords that could reflect flu trends ten weeks in advance are related to influenza vaccines; those a week in advance are related to influenza symptoms; and most of the simultaneous keywords are frequent terms related to influenza; (ii) all three models can predict influenza effectively, and support vector machine yields the most accurate forecasting result. 2018 Vol. 37 (2): 206-219 [Abstract] ( 306 ) HTML (1 KB)  PDF (1489 KB)  ( 2250 )

220 A Review on the Search Engine Result Page (SERP) Wu Dan, Tang Yuan DOI: 10.3772/j.issn.1000-0135.2018.02.010 This paper uses synthesis methods to examine foreign conference and journal articles that study search engine result page (SERP), and analyzes the research progress in this field. With respect to the research available on the layout design of SERPs, the results show that since the modern search engine aggregates the various types of information resources, the SERP is vertical and diversified in nature, and the embedded elements of the SERP affect the user’s satisfaction, attention, experience and evaluation of the SERP; on the other hand, using user-behavior data that includes eye movement, cursor movement, gestures, acoustics and other interactive data, it is possible to analyze the user-fixation areas and attention-distribution characteristics of the SERP to establish a user-attention prediction model, and to predict user intentions, attention, and the results of bias. Future directions mentioned in the literature include SERP layout design research extending to voice interaction, social networking, and other related fields, as well as more comprehensive interactive data support for user-seeking behavior modeling and attention prediction. 2018 Vol. 37 (2): 220-230 [Abstract] ( 180 ) HTML (1 KB)  PDF (1736 KB)  ( 939 )