摘要：In example-based searching, users look for the set of most similar publications to a given one. This requires estimating similarities between publications. A tf.idf formula can be used to compute publication-to-publication text-based similarity, e.g., the Okapi BM25 formula. Studies show that augmenting the importance of search terms in the BM25 formulae improve similarity scores. To this end, we introduce a term-ranking technique and use it for improving publication similarity scores. The proposed term-ranking algorithm is a slight modification of the TextRank algorithm that utilizes the well-known PageRank algorithm to identify the important term/phrases within texts. The proposed approach considers the length of sentences to identify links between terms rather than considering fixed window size. We experimentally found that the proposed approach works well when paired with Okapi BM25.
