Author Disambiguation

The ability to identify scholarly authors is central to bibliometric analysis. Efforts to disambiguate author names using algorithms or national or societal registries become less effective with increases in the number of publications from China and other nations where shared and similar names are prevalent. This work analyzes the adoption and integration of an open source, cross-national identification system, the Open Researcher and Contributor ID system (ORCID), in Web of Science metadata. Results at the article level show greater adoption, to date, of the ORCID iD in Europe as compared with Asia and the US. Focusing analysis on individual highly cited researchers with the shared Chinese surname "Wang," results indicate wide scope for greater adoption of ORCID. The mechanisms for integrating ORCID iDs into articles also come into question in an analysis of co-authors of one particular highly cited researcher who have varying percentages of articles with ORCID iDs attached. These results suggest that systematic variations in adoption and integration of ORCID into publication metadata should be considered in any bibliometric analysis based on it.

Track 2 in KDD Cup 2013 aims at determining duplicated authors in a data set from Microsoft Academic Search. This type of problems appears in many large-scale applications that compile information from different sources. This paper describes our solution developed at National Taiwan University to win the first prize of the competition. We propose an effective name matching framework and realize two implementations. An important strategy in our approach is to consider Chinese and non-Chinese names separately because of their different naming conventions. Post-processing including merging results of two predictions further boosts the performance. Our approach achieves F1-score 0.99202 on the private leader board, while 0.99195 on the public leader board.

Bibliometric methods depend heavily on the quality of data, and cleaning and disambiguating data are very timeconsuming. Therefore, quite some effort is devoted to the development of better and faster tools for disambiguating of the data (e.g., Gurney et al. 2012). Parallel to this, one may ask to what extent data cleaning is needed, given the intended use of the data. To what extent is there a trade-off between the type of questions asked and the level of cleaning and disambiguating required? When evaluating individuals, a very high level of data cleaning is required, but for other types of research questions, one may accept certain levels of error, as long as these errors do not correlate with the variables under study. In this paper, we present an earlier case study with a rather crude way of data handling as it was expected that the unavoidable error would even out. In this paper, we do a sophisticated data cleaning and disambiguation of the same dataset, and then do the same analysis as before. We compare the results and discuss conclusions about required data cleaning.

Knowledge transfer between science and technology has been studied at micro-and macrolevels of analysis. This has contributed to the understanding of the mechanisms and drivers, but actual transfer mechanism and process, be they through codified or tacit sources, have very rarely been mapped and measured to completeness and remain, to a large extent, a black box. We develop a novel method for mapping science-technology flows and introduce 'concept clusters' as an instrument to do so. Using patent and publication data, we quantitatively and visually demonstrate the flows of knowledge between academia and industry. We examine the roles of exogenous and endogenous knowledge sources, and of co-inventors and co-authors in the application of university-generated knowledge. When applied to a stylised case, we show that the method is able to trace the linkages between base knowledge and skill sets and their application to a technology, which in some instances span over twenty-five years.

Knowledge transfer between science and technology has been studied at micro-and macrolevels of analysis. This has contributed to the understanding of the mechanisms and drivers, but actual transfer mechanism and process, be they through codified or tacit sources, have very rarely been mapped and measured to completeness and remain, to a large extent, a black box. We develop a novel method for mapping science-technology flows and introduce 'concept clusters' as an instrument to do so. Using patent and publication data, we quantitatively and visually demonstrate the flows of knowledge between academia and industry. We examine the roles of exogenous and endogenous knowledge sources, and of co-inventors and co-authors in the application of university-generated knowledge. When applied to a stylised case, we show that the method is able to trace the linkages between base knowledge and skill sets and their application to a technology, which in some instances span over twenty-five years.

Key to accurate bibliometric analyses is the ability to correctly link individuals to their corpus of work, with an optimal balance between precision and recall. We have developed an algorithm that does this disambiguation task with a very high recall and precision. The method addresses the issues of discarded records due to null data fields and their resultant effect on recall, precision and F-measure results. We have implemented a dynamic approach to similarity calculations based on all available data fields. We have also included differences in author contribution and age difference between publications, both of which have meaningful effects on overall similarity measurements, resulting in significantly higher recall and precision of returned records. The results are presented from a test dataset of heterogeneous catalysis publications. Results demonstrate significantly high average F-measure scores and substantial improvements on previous and stand-alone techniques.

Key to accurate bibliometric analyses is the ability to correctly link individuals to their corpus of work, with an optimal balance between precision and recall. We have developed an algorithm that does this disambiguation task with a very high recall and precision. The method addresses the issues of discarded records due to null data fields and their resultant effect on recall, precision and F-measure results. We have implemented a dynamic approach to similarity calculations based on all available data fields. We have also included differences in author contribution and age difference between publications, both of which have meaningful effects on overall similarity measurements, resulting in significantly higher recall and precision of returned records. The results are presented from a test dataset of heterogeneous catalysis publications. Results demonstrate significantly high average F-measure scores and substantial improvements on previous and stand-alone techniques.

Key to accurate bibliometric analyses is the ability to correctly link individuals to their corpus of work, with an optimal balance between precision and recall. We have developed an algorithm that does this disambiguation task with a very high recall and precision. The method addresses the issues of discarded records due to null data fields and their resultant effect on recall, precision and F-measure results. We have implemented a dynamic approach to similarity calculations based on all available data fields. We have also included differences in author contribution and age difference between publications, both of which have meaningful effects on overall similarity measurements, resulting in significantly higher recall and precision of returned records. The results are presented from a test dataset of heterogeneous catalysis publications. Results demonstrate significantly high average F-measure scores and substantial improvements on previous and stand-alone techniques.

Key to accurate bibliometric analyses is the ability to correctly link individuals to their corpus of work, with an optimal balance between precision and recall. We have developed an algorithm that does this disambiguation task with a very high recall and precision. The method addresses the issues of discarded records due to null data fields and their resultant effect on recall, precision and F-measure results. We have implemented a dynamic approach to similarity calculations based on all available data fields. We have also included differences in author contribution and age difference between publications, both of which have meaningful effects on overall similarity measurements, resulting in significantly higher recall and precision of returned records. The results are presented from a test dataset of heterogeneous catalysis publications. Results demonstrate significantly high average F-measure scores and substantial improvements on previous and stand-alone techniques.