ABSTRACT
Objectives The sense of smell is important as a warning system, in social communication and in guiding food intake. Impairment is common, and cases are increasing following COVID-19. Olfactory dysfunction may lead to decreased quality of life. There are several established ways to assess olfaction including the "Sniffin' Sticks" which are a validated test for healthy and diseased populations. Methods The odor threshold is traditionally determined using a single staircase procedure, with narrow or wide step. We investigated a Bayesian adaptive algorithm (QUEST) to estimate olfactory threshold in a hyposmic population compared with a healthy control group. Thresholds were measured using the three procedures in two sessions (Test and Retest). Results All the tested methods showed considerable overlap in both groups: there was a positive correlation between the QUEST procedure and classic staircase method (r = 0.88), and high test-retest reliability for all three methods used (Sniffin' Sticks narrow: r = 0.81;Sniffin' Sticks wide: r = 0.95;QUEST: r = 0.80). Conclusions Results from these approaches exhibit considerable overlap with all of them being suitable for clinical use. An advantage of the QUEST method can be the defined number of trials needed to determine an odor threshold.
ABSTRACT
The year 2020 marked an important moment when the COVID-19 pandemic promoted Internet as a necessity even more than before, especially for school activities and businesses. This increased usage emphasized the importance of cybersecurity, a frequently overlooked subject by the common users, which in return plays a crucial role in safe Internet browsing. This paper introduces an approach grounded in Natural Language Processing techniques to identify the main trends in security news and empowers the analysis of vulnerable products, active attacks, as well as existing methods of defence against new attacks. Our dataset consists of 2264 news articles published on cybersecurity dedicated websites between January 2017 and May 2021. The RoBERTa language model was used to compute the texts embeddings, followed by dimensionality reduction techniques and topic clustering methods. Articles were grouped into approximately 20 clusters that were thoroughly evaluated in terms of importance and evolution. © 2021 IEEE.