Search

We exploit the Twitter platform to create a dataset of news articles derived from tweets concerning COVID-19, and use the associated tweets to define a number of popularity measures. The focus on (potentially) biomedical news articles allows the quantity of biomedically valid information (as extracted by biomedical relation extraction) to be included in the list of explored features. Aside from forming part of a systematic correlation exploration, the features – ranging from the semantic relations through readability measures to the article’s digital content – are used within a number of machine learning classifier and regression algorithms. Unsurprisingly, the results support that for more complex articles (as determined by a readability measure) more sophisticated syntactic structure may be expected. A weak correlation is found with information within an article suggesting that other factors, such as numbers of videos, have a notable impact on the popularity of a news article. The best popularity prediction performance is obtained using a random forest machine learning algorithm, and the feature describing the quantity of biomedical information is in the top 3 most important features in almost a third of the experiments performed. Additionally, this feature is found to be more valuable than the widely used named entity recognition.

Many automatic semantic relation extraction\ud tools extract subject-predicate-object triples\ud from unstructured text. However, a large quantity of these triples merely represent background knowledge. We explore using full texts\ud of biomedical publications to create a training\ud corpus of informative and important semantic\ud triples based on the notion that the main contributions of an article are summarized in its\ud abstract. This corpus is used to train a deep\ud learning classifier to identify important triples,\ud and we suggest that an importance ranking for\ud semantic triples could also be generated.

Background\ud Literature based discovery (LBD) automatically infers missed connections between concepts in literature. It is often assumed that LBD generates more information than can be reasonably examined.\ud \ud Methods\ud We present a detailed analysis of the quantity of hidden knowledge produced by an LBD system and the effect of various filtering approaches upon this. The investigation of filtering combined with single or multi-step linking term chains is carried out on all articles in PubMed.\ud \ud Results\ud The evaluation is carried out using both replication of existing discoveries, which provides justification for multi-step linking chain knowledge in specific cases, and using timeslicing, which gives a large scale measure of performance.\ud \ud Conclusions\ud While the quantity of hidden knowledge generated by LBD can be vast, we demonstrate that (a) intelligent filtering can greatly reduce the number of hidden knowledge pairs generated, (b) for a specific term, the number of single step connections can be manageable, and (c) in the absence of single step hidden links, considering multiple steps can provide valid links.

Background: The quantity of documents being published requires researchers to\ud specialize to a narrower field, meaning that inferable connections between\ud publications (particularly from different domains) can be missed. This has given\ud rise to automatic literature based discovery (LBD). However, unless heavily\ud filtered, LBD generates more potential new knowledge than can be manually\ud verified and another form of selection is required before the results can be passed\ud onto a user. Since a large proportion of the automatically generated hidden\ud knowledge is valid but generally known, we investigate the hypothesis that non\ud trivial, interesting, hidden knowledge can be treated as an anomaly and identified\ud using anomaly detection approaches.\ud \ud Results: Two experiments are conducted: (1) to avoid errors arising from\ud incorrect extraction of relations, the hypothesis is validated using manually\ud annotated relations appearing in a thesaurus, and (2) automatically extracted\ud relations are used to investigate the hypothesis on publication abstracts. These\ud allow an investigation of a potential upper bound and the detection of limitations\ud yielded by automatic relation extraction.\ud \ud Conclusion: We apply one-class SVM and isolation forest anomaly detection\ud algorithms to a set of hidden connections to rank connections by identifying\ud outlying (interesting) ones and show that the approach increases the F1 measure\ud by a factor of 10 while greatly reducing the quantity of hidden knowledge to\ud manually verify. We also demonstrate the statistical significance of this result.\ud \ud Keywords: literature based discovery; anomaly detection; unified medical\ud language system

As the quantity of publications increases daily, researchers are forced to narrow their attention to their own specialism and are therefore less likely to make new connections with other areas. Literature based discovery (LBD) supports the identification of such connections. A number of LBD tools are available, however, they often suffer from limitations such as constraining possible searches or not producing results in real-time. \ud \ud \ud We introduce HiDE (Hidden Discovery Explorer), an online knowledge browsing tool which allows fast access to hidden knowledge generated from all abstracts in Medline. HiDE is fast enough to allow users to explore the full range of hidden connections generated by an LBD system. The tool employs two novel approaches to LBD: a graph-based approach which allows hidden knowledge to be generated on a large scale and an inference algorithm to identify the most promising information. \ud \ud \ud Available at http://nlp.shef.ac.uk/hide

Background: Literature based discovery (LBD) automatically infers missed\ud connections between concepts in literature. It is often assumed that LBD\ud generates more information than can be reasonably examined.\ud \ud Methods: We present a detailed analysis of the quantity of hidden knowledge\ud produced by an LBD system and the effect of various filtering approaches upon\ud this. The investigation of filtering combined with single or multi-step linking term\ud chains is carried out on all articles in PubMed.\ud \ud Results: The evaluation is carried out using both replication of existing\ud discoveries, which provides justification for multi-step linking chain knowledge in\ud specific cases, and using timeslicing, which gives a large scale measure of\ud performance.\ud \ud Conclusions: While the quantity of hidden knowledge generated by LBD can be\ud vast, we demonstrate that (a) intelligent filtering can greatly reduce the number\ud of hidden knowledge pairs generated, (b) for a specific term, the number of single\ud step connections can be manageable, and (c) in the absence of single step hidden\ud links, considering multiple steps can provide valid links.\ud