557 Exploratory Analysis of Citation Counts and Altmetrics Indicators of Chinese Academic Books and the Implications Hot! Ming Li, Jiang Li, Ming Chen, Shi Jin DOI: 10.3772/j.issn.1000-0135.2019.06.001 This study performed an exploratory analysis of the relationship between citation counts and altmetrics indicators of Chinese academic books, thereby providing a factual reference for the rational construction of a comprehensive impact evaluation system for Chinese academic books under the new academic exchange environment. The selected samples were three categories of top-ten cited books from CBKCI statistical reports, and their altmetrics data were collected from the Douban Reading platform. All data were analyzed by correlation analysis and principal component analysis. Our study found that citation counts and altmetrics indicators reflect different impacts of the sample books and that they are independent and complementary. From the perspective of the data’s internal correlation and clustering, additional factors must be extracted from the altmetrics indicators. The comprehensive evaluation of the impact of Chinese academic books can be conducted horizontally from both academic and social dimensions. Vertically, the level of communication and value judgment should be distinguished. In addition, clarifying the relationship between altmetrics and traditional evaluation methods can make the overall study more systematic. The introduction of altmetrics provides unique value in improving the usefulness of an evaluation system for Chinese academic books. 2019 Vol. 38 (6): 557-567 [Abstract] ( 229 ) HTML (164 KB)  PDF (809 KB)  ( 788 )

568 Literature Metadata Integration Hot! Qiujing Ding, Jianxun Zeng DOI: 10.3772/j.issn.1000-0135.2019.06.002 The development in digital publishing has gradually reduced the advantages of bibliographic data integration in libraries and information institutions that have acquisition and cataloguing as their core functions. They lag behind publishers and integrators scale, depth and application of data. Publishers and integrators strengthen the integration of literature metadata through fine-grained data processing, semantic association, and large-scale storage. The modern information service industry has undergone significant changes. Owing to the challenges of lagging metadata management and limited copyright of full-text service, libraries and information institutions need to pay greater attention to the local accumulation and application of literature metadata on the basis of the construction of full-text resources. The development of a modern information service system is supported by the collection and acquisition of literature metadata from multiple sources, integrations and semantic organization, based on various literature elements. Knowledge discovery services can be carried based on the integrated metadata. In the end， the content and significance of information resources can be fully reflected, and metadata development strategy can be practiced. 2019 Vol. 38 (6): 568-577 [Abstract] ( 177 ) HTML (75 KB)  PDF (1050 KB)  ( 804 )

578 Research on Hotspots and Trends of Domestic Text Mining Based on Cluster Analysis Hot! Tan Zhanglu, Peng Shengnan, Wang Zhaogang DOI: 10.3772/j.issn.1000-0135.2019.06.003 Understanding research hotspots and trends in the field of domestic text mining has immense significance in mastering the development and changes in domain content and promoting further development of the related research. First, this study uses the research literature of 1155 text mining related topics in CNKI database from 1998 to 2017 as the sample and the word frequency matrix of the article keywords as the data. It employs the SPSS software for cluster analysis. Further, the chi-square statistics are used to extract high-degree keywords to interpret the clustering results. According to the clustering results, the literature in the text mining field is divided into 13 categories from the macroscopic level to grasp the research hotspots and trends of domestic text mining. The results show the following: (i) The research on basic research of text mining, text big data preprocessing, and text mining application field are hot topics, (ii) the amount of applied research literature related to association rules, text clustering, and text classification is small, and (iii) text topic analysis, text big data preprocessing, and web text mining research are likely to become new research trends in the future. 2019 Vol. 38 (6): 578-585 [Abstract] ( 659 ) HTML (109 KB)  PDF (946 KB)  ( 1673 )

586 Research on Scenario Element Extraction from Intelligence Resources on Hazardous Chemical Accidents Hot! Feng Yang, Yao Leye DOI: 10.3772/j.issn.1000-0135.2019.06.004 Emergency management organization must use existing scenarios to simulate risk instances. Scenario element extraction can comprehensively analyze the composition and evolution of scenarios, which is important for objectively analyzing and making scientifically based decisions in emergencies. Extracting scenario elements from the existing collection of intelligence resources on hazardous chemical accidents can provide basic materials to formulate possible combinations of future events and effectively drive sensitive intelligence resources. This paper selects 120 hazardous chemical accidents as a basic intelligence resource, using grounded theory to extract scenario elements. Based on the 209 concepts derived from the qualitative analysis of the original coded data, 37 categories, 13 main categories, and 3 core categories were identified. These comprise scenario elements of hazardous chemical accidents. The research may be used as a reference for subsequent representation and construction of hazardous chemical accident scenarios. 2019 Vol. 38 (6): 586-594 [Abstract] ( 297 ) HTML (114 KB)  PDF (798 KB)  ( 791 )

595 Construction and Visual Analysis of Academic Paper-Linked Data Based on In-depth Mining Hot! Qu Jiabin,Ou Shiyan,Ling Hongfei DOI: 10.3772/j.issn.1000-0135.2019.06.005 Since Linked Data was proposed, it has become the mainstream method of publishing structured data on the Web. With the rapid increase in linked data sets, the effective consumption and utilization of linked data has become the focus of researchers. This study intended to explore the mining and visual analysis of linked data. Firstly, we conducted in-depth mining of implicit information hidden in the metadata of academic papers in the geological field using text mining techniques. We then transformed the metadata and mined information into RDF-based semantic representation to construct the linked data of academic papers based on a newly designed “academic paper-scholar” ontology. On this basis, five visual analysis modules were designed to visualize the macro- and micro-knowledge of academic paper-linked data from multiple perspectives. The results showed that (1) the linked data constructed based on in-depth mining can deeply and comprehensively display knowledge hidden in the metadata of academic papers and (2) the visual analysis of linked data can intuitively present macro- and micro-knowledge in the form of graphics and thus facilitate users’ rapid consumption and utilization of linked data. 2019 Vol. 38 (6): 595-611 [Abstract] ( 237 ) HTML (110 KB)  PDF (6299 KB)  ( 708 )

612 A Deep-Learning Model Based on Attention Mechanism for Chinese Comparative Relation Detection Hot! Zhu Maoran, Wang Yilei, Gao Song, Wang Hongwei, Zheng Lijuan DOI: 10.3772/j.issn.1000-0135.2019.06.006 There are a massive number of comparative opinions in online reviews, containing users’ assessments of their experience with different products or services. Businesses can gain insight into their market competitiveness by identifying useful user-generated comparison information from among a mass of low-quality comments. We were thereby motivated to study comparative sentence recognition in Chinese comments. Instead of using the pattern recognition method as past studies did, we addressed the recognition task through a hierarchical multi-attention network based on deep learning. Our model outperforms both the traditional classification model and the deep-learning-based text classification model in term of accuracy, with the F1-score reaching 81%. The proposed hierarchical multi-attention network model is end-to-end, thus avoiding the design of a large number of artificial features, and greatly reducing human involvement for comparative comment recognition. 2019 Vol. 38 (6): 612-621 [Abstract] ( 189 ) HTML (107 KB)  PDF (1198 KB)  ( 1544 )

622 An Ontology Fusion Method Based on Binary Similarity Calculation Hot! Lou Wen, Wang Hui, Ju Yuan DOI: 10.3772/j.issn.1000-0135.2019.06.007 Heterogeneous ontology causes redundancy in knowledge retrieval. Therefore, knowledge fusion based on heterogeneous ontology is necessary. However, because of the massive capacity and complicated processes required for semantic similarity computing, knowledge fusion has become less simple. In this paper, we propose an ontology fusion method based on binary metrics of semantic similarity calculation. In the fusion process, there will be only binary matching, thus aiming to further simplify the calculation of fusion from semantic similarity. Thus, the present research represents a shift from methods locating computing progress at the beginning of original ontology construction. We adopted three experiments to test the usability of our approach, from the perspectives of (1) actual library resources, (2) a small dataset, and (3) a large dataset. In experiment one, bibliographic data from Wuhan University Library were used to test our proposal s feasibility and capabilities. Results showed that our approach can completely merge two ontologies into a single theme. The second and third experiments both verified that our approach has the ability to accurately detect merging couples and decrease time cost. The tests demonstrated a good overall fusion result; nevertheless, recall requires future improvement. This method is expected to extend the implementation of expert ontology and aid in cost reduction of ontology construction. 2019 Vol. 38 (6): 622-631 [Abstract] ( 184 ) HTML (122 KB)  PDF (1205 KB)  ( 591 )

632 A Novel Recommendation Approach of Electronic Literature Resources Combining Semantic and Social Features Hot! Yang Chen, Liu Tingting, Liu Lei, Niu Ben, Sun Jianshan DOI: 10.3772/j.issn.1000-0135.2019.06.008 With the arrival of the era of information expansion, the load on electronic literature databases will dramatically increase, and it will become increasingly difficult for users to search for their required pieces of literature. In response to this issue, the development of recommendation systems to assist the management of electronic literature databases has received extensive attention from researchers. Currently, one commonly used recommendation technique for literature databases is collaborative filtering. However, the traditional collaborative filtering algorithms, which only consider the similarity of users’ search-history, ignore several important factors, such as the users’ semantic similarity and social relationships. In this paper, we integrated a text content similarity based on topic model as well as two kinds of user similarities based on social relationships (user tag similarity and personal group similarity) into the user collaborative filtering recommender system by utilizing an unsupervised integration strategy. The experiment on the real data set shows that by adding the multiple source features, there is an enhancement and promotion effect on the recommendation accuracy, which provides strong implications for related electronic literature resource recommendation research in the future. 2019 Vol. 38 (6): 632-640 [Abstract] ( 202 ) HTML (109 KB)  PDF (1020 KB)  ( 708 )

641 Research on the Domain Knowledge Alignment Model Based on Deep Learning: The Knowledge Graph Perspective Hot! Yu Chuanming, Wang Feng, An Lu DOI: 10.3772/j.issn.1000-0135.2019.06.009 To solve the problems of redundancy and inconsistency in the process of domain knowledge fusion, this paper studies domain knowledge alignment from the perspective of the knowledge graph. A novel knowledge graph alignment (KGA) model is proposed based on knowledge graph deep-representation learning. To verify the validity of the model, comparative experiments are conducted on the datasets of heterogeneous knowledge graphs and cross-lingual knowledge graphs. On heterogeneous datasets, the experimental results show that the Hits@1 value of the model is increased by 6.40% and the MRR value is increased by 6.30% over the traditional MTransE and IPTransE. On cross-lingual datasets, the experimental results show that the Hits@1 value of the model is increased by 9.66% and the MRR value is increased by 9.60%. The experimental results show that the effect of the KGA model on domain knowledge alignment is better than the traditional domain knowledge alignment methods. These research results are of great significance for improving the alignment effect of knowledge graph entities, improving the coverage and the correct rate of domain knowledge, and promoting the performance of knowledge graphs in the information field. 2019 Vol. 38 (6): 641-654 [Abstract] ( 217 ) HTML (166 KB)  PDF (1289 KB)  ( 1786 )

655 Exploring UsersInformation Source Selection and Use Strategies in Learning Related Search Hot! Song Xiaoxuan, Liu Chang DOI: 10.3772/j.issn.1000-0135.2019.06.010 With the volume of information increasing and online quality in question, it becomes more difficult for users to acquire appropriate information that satisfies their needs and enriches their knowledge by using search engines. When searching, usersinformation selection and subsequent use depend largely on information sources. A user experiment was conducted in order to design two kinds of tasks, one related to receptive learning and the other to critical learning. This study focused on the differences in the distribution of information source types during the usersprocess of source selection and use. It also explored the relationship between usersinformation source strategies and their learning outcomes. The results showed that users were more dependent on the information sources they ultimately selected rather than those recommended by search engines. According to the degree of dependence, four types of information source strategies were identified: total dependence, non-dependence, selective dependence, and use-dependence. We also found that usersinformation source strategies had significant impact on learning outcomes. Users adopting non-dependent strategies for receptive learning-related tasks, as well as those adopting total dependence strategies for critical learning-related tasks, could achieve better learning outcomes. This study contributes by clarifying the characteristics of usersinformation sources during selection and subsequent use in order to optimize search engine systems to improve userslearning outcomes by providing accurate and efficient guidance. 2019 Vol. 38 (6): 655-666 [Abstract] ( 182 ) HTML (118 KB)  PDF (1276 KB)  ( 784 )