Data mining tools combat COVID-19 misinformation and identify symptoms

UC Riverside machine scientists are processing tools to assistance way and show COVID-19 symptoms and to sift done misinformation astir the illness connected societal media.

Using Google Trends data, a radical led by Vagelis Papalexakis, an subordinate prof successful the Marlan and Rosemary Bourns College of Engineering; and Jia Chen, an adjunct prof of teaching, developed an algorithm that identified 3 symptoms unsocial to COVID-19 compared to the flu: ageusia—loss of the tongue's sensation function—shortness of breath, and anosmia, oregon nonaccomplishment of smell. The algorithm was developed successful collaboration with 2 postgraduate students, Md Imrul Kaish and Md Jakir Hossain, astatine the University of Texas Rio Grande Valley.

"Much of the enactment utilizing Google Trends for flu has focused connected forecasting the flu season," Papalexakis said. "We, connected the different hand, utilized it to spot if we could find a needle successful a haystack: symptoms unsocial to COVID-19 among each the flu-like symptoms radical hunt for."

The researchers located symptoms connected Google Trends for 2019 and 2020 and utilized a method they called nonnegative discriminative analysis, oregon DNA, to extract presumption that were unsocial to 1 dataset comparative to the other.

"We assumed that symptom searches successful 2019 would pb to influenza oregon different respiratory ailments, portion searches for the aforesaid symptoms successful 2020 could beryllium either," Chen said. "Using DNA, we were capable to find the quality betwixt the 2 datasets. This happened to beryllium presumption clinicians person already identified arsenic unsocial to COVID-19, showing that our attack works."

Papalexakis and Chen expect their enactment volition assistance epidemiologists and different nationalist wellness experts way and show COVID-19 utilizing Google Trends arsenic a proxy for infirmary data.

"Google trends information is precise noisy, but infirmary information is not publically available. People mightiness hunt for symptoms due to the fact that they are experiencing them oregon due to the fact that they person heard of them and privation to cognize more," Papalexakis said. "Searches bespeak involvement successful symptoms amended than radical actively experiencing them, but fixed the deficiency of different data, we deliberation this instrumentality could assistance researchers recognize symptoms better."

Chen said that the algorithm is elemental and casual to instrumentality arsenic portion of a imaginable instrumentality that tin assistance scientists researching different diseases larn astir imaginable symptoms.

The paper, "COVID-19 oregon Flu? Discriminative Knowledge Discovery of COVID-19 Symptoms from Google Trends Data," was presented astatine epiDAMIK 2021, a store connected data mining for advancing epidemiological knowledge. The store was organized arsenic portion of the largest yearly information subject conference, the Association for Computing Machinery's, oregon ACM, Special Interest Group connected Knowledge Discovery and Data Mining.

Papalexakis and UC Riverside doctoral pupil William Shiao are besides processing a instrumentality that not lone identifies COVID-19 misinformation but shows wherefore the accusation is flagged arsenic mendacious successful narration to a database of technological articles astir probe connected coronaviruses.

Papalexakis and Shiao utilized 90,000 articles from the COVID-19 Open Research Dataset Challenge (CORD-19) prepared by the White House and a conjugation of probe groups, and collected 20,000 articles "in the wild" with misinformation astir the caller coronavirus. Using a similarity matrix-based embedding method they called KI2TE, the articles were linked to a acceptable of notation documents and interpreted. The documents utilized for notation were a acceptable of world papers connected coronavirus probe included successful the CORD-19 dataset.

When tested connected articles that had been labeled by humans arsenic mendacious oregon identified by Google Fact Check arsenic false, their method not lone correctly identified the mendacious stories but besides pointed to the technological sources that corroborated the system's decision.

"We are not funny successful censoring what radical see. We privation to spell beyond hiding thing altogether oregon simply showing a informing label," Papalexakis said. "We privation to besides amusement them sources to amended them."

Although the instrumentality developed by Papalexakis and Shiao is simply a prototype nether progressive probe development, it could yet beryllium incorporated into a smartphone app oregon into societal media platforms similar Facebook.

---

---