Reflective random indexing for semi-automatic indexing of the biomedical literature

Abstract

The rapid growth of biomedical literature is evident in the increasing size of the MEDLINE research database. Medical Subject Headings (MeSH), a controlled set of keywords, are used to index all the citations contained in the database to facilitate search and retrieval. This volume of citations calls for efficient tools to assist indexers at the US National Library of Medicine (NLM). Currently, the Medical Text Indexer (MTI) system provides assistance by recommending MeSH terms based on the title and abstract of an article using a combination of distributional and vocabulary-based methods. In this paper, we evaluate a novel approach toward indexer assistance by using nearest neighbor classification in combination with Reflective Random Indexing (RRI), a scalable alternative to the established methods of distributional semantics. On a test set provided by the NLM, our approach significantly outperforms the MTI system, suggesting that the RRI approach would make a useful addition to the current methodologies.

A semi-automatic indexing system based on embedded information in HTML documents

“…However automatic systems also present problems because of the complexity of natural language processing (Sinkkilä et al, 2011). Consequently the semi-automatic indexing approach is a good solution, because in addition to obviating the problems of the automatic indexing system it facilitates the the task of indexers by providing suitable term suggestions (Vasuki and Cohen, 2010).…”

Leveraging output term co-occurrence frequencies and latent associations in predicting medical subject headings

“…Vasuki and Cohen [10] use an interesting approach that employs reflective random indexing to find the nearest neighbors in the training dataset and use the indexing based similarity scores to rank the terms from the neighboring citations. A recent effort by Jimeno-Yepes et al [11] uses a large dataset and uses meta-learning to train custom binary classifiers for each MeSH term and index the best performing model for each terml for usage on new testing citations; we request the reader to refer to their work for a recent review of machine learning approaches used for MeSH term assignment.…”

“…We experiment with two public datasets used by Huang et al [1]. The NLM2007 dataset has 200 test citations and is used by other recent studies on this subject [10]. The L1000 dataset is curated by Huang et al by random selection for the purposes of their work to test their methods on a larger dataset that spanned a large number of years.…”

“…For a detailed analysis of other RI variants and a thorough introduction, please see [6]. We note that Vasuki and Cohen [10] use TRRI to obtain the nearest neighbors of a testing citation and rank the neighbors’ terms using the citation similarity score sums as discussed in Section 5.3.1. Using this approach they obtain results better than the MTI method.…”

Indexation automatique de documents en santé : évaluation et analyse de sources d’erreurs

“…Les travaux sur le MeSH en anglais ont utilisé le modèle probabiliste [4] et les techniques d'apprentissage automatique telles que le réseau bayésien [5] et les k-plus proches voisins pour la classification des documents [2,[6][7][8]. Aronson et al [2] exploitent également l'outil MetaMap [9] et la méthode de tri-gram (cette méthode permet de déterminer la similarité entre deux phrases) pour l'extraction des concepts Unified Medical Language System (UMLS 1 ) qui sont ensuite restreints aux concepts MeSH.…”