1015 Research on Text Matching Model Based on Deep Interaction Hot! Yu Chuanming, Xue Haodong, Jiang Yifan DOI: 10.3772/j.issn.1000-0135.2021.10.001 For the application of text matching in information retrieval, text mining, and other research fields, a deep interactive text matching (DITM) model with good generalization ability is proposed. Based on the matching-aggregation framework, the encoder layer, co-attention layer, and fusion layer are used as an interaction module. The interaction process is iterated multiple times to obtain the in-depth interaction information. Finally, information is extracted through multi-perspective pooling to predict the relationship between text pairs. Compared with baseline methods, the proposed approach has achieved best results on four text matching tasks, namely opinion retrieval, answer selection, paraphrase identification, and natural language inference. The experimental results are of great significance to promote the practice of text matching models in the field of information. 2021 Vol. 40 (10): 1015-1026 [Abstract] ( 235 ) HTML (160 KB)  PDF (1493 KB)  ( 901 )

1027 Knowledge Evolution Analysis of ESI Research Fronts Based on Knowledge Element Migration Hot! Sun Zhen, Leng Fuhai DOI: 10.3772/j.issn.1000-0135.2021.10.002 Based on previous research, this paper proposes a method for analyzing the knowledge evolution of ESI research fronts based on knowledge element migration. Through quantitative analysis of the knowledge element transfer phenomenon and calculation of migration degree, the mechanism of the evolution of ESI (essential science indicators) research is further explored from the perspective of semantic analysis and knowledge computing. With the help of named entity recognition, bag of words model, PLDA (parallel latent Dirichlet allocation) topic model, information entropy algorithm, and other text semantic mining and natural language processing technologies, this paper explores the migration rule of knowledge elements by designing contribution index CVI and migration index MVI. The results show that by taking individual knowledge element in the front topic as the analysis object, it is possible to mine the changing laws of the inherent knowledge structure characteristics of ESI research front over time from a direct and fine-grained perspective. The method can not only reveal the evolution status of domain knowledge elements in different periods but also identify and track changes in the development of the research front in a more in-depth manner. This study provides a methodological reference for the scientific and technological intelligence work focusing on identifying domain research fronts. 2021 Vol. 40 (10): 1027-1042 [Abstract] ( 199 ) HTML (229 KB)  PDF (2156 KB)  ( 376 )

1043 Predicting Potential Technology Partners and Competitors of Enterprises: A Case Study on Fuel Cell Technology Hot! Li Bing, Ding Kun, Sun Xiaoling DOI: 10.3772/j.issn.1000-0135.2021.10.003 The rapid increase in the number of patents has made it more time-consuming and labor-intensive for companies to evaluate and screen potential technology partners and identify and avoid competitors in a large industry. The method of accurately and quickly narrowing the search scope and locating potential relationships becomes crucial and meaningful. In this study, a company-patent heterogeneous network is constructed based on the bipartite graph theory. The research method uses the link prediction algorithm based on the SimRank indicator of random walk, and a predictive analysis is conducted on the company's potential technology partners and competitors. The network representation method is used to identify the context information of patents and calculate the similarity of the patent representation vector to measure the technical difference between the target company and the competing object to determine the competitive relationship. Finally, an empirical study in the field of fuel cell technology is carried out to verify the effectiveness of the research theory and method and provide a reference for the development of enterprises. 2021 Vol. 40 (10): 1043-1051 [Abstract] ( 242 ) HTML (91 KB)  PDF (3070 KB)  ( 397 )

1052 Diachronic Semantic Mining and Visualization of Chinese Words: A Knowledge Discovery Perspective Hot! Pan Jun, Wu Zongda DOI: 10.3772/j.issn.1000-0135.2021.10.004 Mining knowledge from diachronic word semantic shifts has become an increasingly important problem in word temporal analysis. To this end, this paper aims to design a scalable framework for knowledge mining in the diachronic corpus, which is based on a loosely-coupled and service-oriented configurable architecture. The bottom layer of the framework provides data level services such as data cleansing, data normalization, and diachronic word vectors learning, among others. The middle layer defines customized data extraction strategy and user interface generation through the configuration files in xml format. The top layer uses various services to fulfill specific requirements of knowledge discovery and visualization. This study also implements a framework focusing on word semantic shifts of People’s Daily and identifies possible approaches in the application of diachronic word vector to digital humanities and social computing research. The proposed framework and its implementation are highly scalable, which can be used as a basis for researchers to further develop applications for diachronic word semantic knowledge mining and can also be extended to other diachronic corpora. 2021 Vol. 40 (10): 1052-1064 [Abstract] ( 203 ) HTML (120 KB)  PDF (5049 KB)  ( 457 )

1065 How Citation Dynamics Change: The Effect of Literature Content Characteristics Hot! Li Lingying, Min Chao, Yan Xiaoran DOI: 10.3772/j.issn.1000-0135.2021.10.005 The citation performance of an article is determined by its internal characteristics and external environment. This study focuses on literature characteristics, such as research quality, innovation level, and content diversity. Citation peak is the most influential stage in the process of citation; hence, apart from traditional citation counts, we also explored the impact of literature characteristics on citation peak. We conducted four regression models on biomedical literature in PubMed, in which the dependent variables were total citation numbers, peak counts, peak arrival time, and peak height. Research quality was measured by a peer review database, Faculty Opinions (F1000), and the innovation level were determined by experts in F1000. Content diversity was expressed by the distance of MeSH terms. The study results indicate that research quality, innovation level, and content diversity contribute to total citation numbers. Diversity can promote article gain more citation peaks. Peak arrival time is significantly influenced by research quality and content diversity. Research quality can decrease peak height, which makes citation curve smoother. 2021 Vol. 40 (10): 1065-1078 [Abstract] ( 260 ) HTML (174 KB)  PDF (1309 KB)  ( 453 )

1079 Distribution Features of Facebook Altmetrics of Scholarly Outputs Hot! Yu Houqiang, Zhang Wei, Cao Xueting DOI: 10.3772/j.issn.1000-0135.2021.10.006 In this paper, a statistical analysis of more than 420,000 Facebook altmetrics from July 2018 to June 2019 and a preliminary exploration of their numerical distribution characteristics were conducted. The study found that the relative coverage rate of Facebook altmetrics is 8.1%, indicating a relatively low level. The immediacy rate of Facebook altmetrics is 74%, which is better than News, Twitter, Weibo, and Policy document altmetrics, which instructs Facebook users to pay more attention to the latest achievements. The distribution of academic publications is relatively even: as based on the number of independent users, 20% of academic publications are mentioned by only 37% of Facebook altmetrics. The distribution of academic sources is in accordance with Bradford's Law, and 140 core sources are calculated, of which the core sources are The Conservation, Nature, and Science. As for the discipline-level distribution of Facebook altmetrics, medical and health sciences is in the dominant position, reaching as high as 61%, and biological science, psychology, and cognitive science also receive relatively high mentions. These conclusions will provide a reference for the further application of Facebook altmetrics. 2021 Vol. 40 (10): 1079-1091 [Abstract] ( 148 ) HTML (149 KB)  PDF (1885 KB)  ( 391 )

1092 Research on the Construction of an Intelligence Process Model for Strategic Decision-Making Hot! Li Pin, Yang Jianlin DOI: 10.3772/j.issn.1000-0135.2021.10.007 Information Science and information work always emphasize supporting decision-making, and the intelligence process is an important way to realize the function of information in supporting decision-making. Although most current intelligence processes are built to meet the decision-making needs, the differences in the characteristics of information needs in multiple types of decision-making are ignored, and there has been little focus on intelligence processes specifically for strategic decision support. In this study, the factors which influence strategic decision-making are systematically analyzed, and the functions of intelligence in strategic decision-making are explored. Based on classical theories such as the system theory, an intelligence process model for strategic decision-making is proposed. Such a model attempts to overcome the defects of traditional intelligence process as well as solve the problems of weak pertinence and low application effectiveness of intelligence process in strategic decision support. 2021 Vol. 40 (10): 1092-1107 [Abstract] ( 222 ) HTML (139 KB)  PDF (1856 KB)  ( 559 )

1108 Filter Bubbles—Induced by Personalized Recommendation Algorithms: A Review of Related Research Hot! Jiang Tingting, Xu Yanrun DOI: 10.3772/j.issn.1000-0135.2021.10.008 Personalized recommendation has engendered in the cyberspace numerous filter bubbles where users can only receive a low diversity of information. This study conducted a systematic review on 61 research articles on filter bubbles, which were published between 2010 and 2020. The review of the theoretical and technological foundations of filter bubbles and related research foci led to the following major findings: (1) researchers have taken different perspectives and adopted different rationales to determine whether filter bubbles existed; (2) filter bubbles have negative impacts on the development of individuals and the society in most cases; and (3) in coping with filter bubbles, researchers have attempted to reduce the effects of filter bubbles in virtue of various information filtering visualizations and to break or prevent filter bubbles by improving personalized recommendation algorithms. This study pioneered the in-depth interpretation of the concept of “filter bubble,” emphasizing its close relationship with personalized recommendation algorithms and the low diversity of information as one of its defining characteristics. 2021 Vol. 40 (10): 1108-1117 [Abstract] ( 259 ) HTML (133 KB)  PDF (1215 KB)  ( 1807 )

1118 Review of Application Research on Network Analysis in Informetrics Hot! Wu Jiang, Wang Kaili, Dong Ke, Yang Yujie, Yi Mengxin DOI: 10.3772/j.issn.1000-0135.2021.10.009 There are several important achievements of network analysis research under informetrics, and interested researches have reached the adjustment stage for the innovation at present. Summarizing the current relevant research can provide a reference for the sustainable development of this field. Starting from the development of network analysis application in informetrics, this study puts forward the research framework and discusses the essence of network analysis; subsequently, the study separately reviews the existing application researches from the macro, medium, and micro levels. Finally, possible future research directions in different dimensions are described. 2021 Vol. 40 (10): 1118-1128 [Abstract] ( 276 ) HTML (178 KB)  PDF (928 KB)  ( 710 )