Unsupervised Keyphrase Extraction from Scientific Publications

AbstractWe propose a novel unsupervised keyphrase extraction approach that filters candidate keywords using outlier detection. It starts by training word embeddings on the target document to capture semantic regularities among words. then uses minimum covariance determinant estimator model distribution of non-keyphrase vectors, under assumption these vectors come from same distribution, indicative their irrelevance semantics expressed dimensions learned vector representation. Candidate keyphrases only consist words are detected as outliers this dominant distribution. Empirical results show our outperforms state-of-the-art and recent methods.KeywordsUnsupervised extractionOutlier detectionMCD