| Evaluation of scientific publication is usually related to the number of citations of an article. However several problems need attention when considering the number of citations as a measure of the quality of a scientific article. For example, authors can write many articles and cite themself and increase artificially the number of their citations. Pioneering articles have usually few citations although they represent very important steps before a major discovery.
Here are the results of a new self-consistent benchmarking method that takes into account the number of citations but assign different self-consistent weights to each citation.
 |
In this figure, three different benchmark methods are compared for eight fictitious articles: the red bars represent the number of citations in other articles. The green bar the benchmark that takes into account the weigthed number of citations. The blue bar is the same as the green bar but without self-citations.
Note the better rank of articles 1 and 2 with the weighted benchmarking: the pioneering work is now rewarded! | This method has many advantages with respect to standard citations couting: - Fairness: pioneering articles that are cited by important articles are now rewarded.
- Differentiation: articles with a same number of citations have now a different ranking.
- Long term stability: articles whose aim is only to cite other articles have automaticaly a lower inluence. Hence it it hopeless to try to influence by artificial methods the ranking of an article.
- Comparison: countries or groups using a large number of citations in order to promote their work become now comparable to groups with low number of citations.
In the next example, 25 fictitious articles are studied, three groups have been build: group A: articles 1 to 5 are precursor work. Group B: Articles 6 to 15 form a closed group where the number of citations is high. Group C: Articles 16 to 25 form a closed group with few citations. We can observe that the average citations number of group B is 3.66 whereas it is 1.88 for group C. With the self-consistent ranking, the average benchmark of group B is 2.41 and the average benchmark of group C is 1.90. The difference between the two groups B and C is reduced by the weights. Moreover best ranked articles from goup C become better than many articles from group B with more citations! The algorithm takes into account the relative importance of citations and not the absolute number of citations. As in the first example the pioneering work of article 1 to 5 are rewarded.
 |
| In this figure, 25 fictitious articles are reported: the red bars represent the number of citations coming from other articles, and the blue bars are obtained with the self-consistent weigthed method. |
Bibliometrics on Wikipedia: http://en.wikipedia.org/wiki/Bibliometrics, http://en.wikipedia.org/wiki/Impact_factor Cibermetrics, International Journal of Scientometrics, Informetrics and Bibliometrics: http://www.cindoc.csic.es Scientometrics: http://www.springerlink.com/content/101080/ |