Title and abstract of the articles-keywords for distinctive competencies are based on the keywords assigned to each of the 80,000 clusters in the product. a distinctive competency is built up from a set of those clusters.the keywords for each of the clusters are generated as follows:we take all of the titles/abstracts of articles in a cluster (not just for your institution, but for any institution) and extract those 2-word phrases with the highest mutual information or entropy across the set of titles/abstracts. instead of doing it just once for each cluster, we do it many times using random subsets of the cluster. we then build up statistics for each phrase, leading to the weights. for example, if we take 20 separate subsets of ¼ of the cluster, and the phrase e. coli is one of the top entropy phrases in 19 of the 20 subsets, we give e. coli a weight of 0.95 for that cluster. keywords are weighted in a cluster, and now we will also apply an additional weighting based on the number of articles your institution has in each of the clusters, to have better keywords for the dc. this should increase the relevance of the keywords from generic for the combination of clusters, to more distinct/specific for your publication history in the clusters.
Answer Anonymously
Copy Link
