Identifying COVID-19 cases and extracting patient reported symptoms from Reddit using natural language processing

Abstract

We used social media data from "covid19positive" subreddit, from 03/2020 to 03/2022 to identify COVID-19 cases and extract their reported symptoms automatically using natural language processing (NLP). We trained a Bidirectional Encoder Representations from Transformers classification model with chunking to identify COVID-19 cases; also, we developed a novel QuadArm model, which incorporates Question-answering, dual-corpus expansion, Adaptive rotation clustering, and mapping, to extract symptoms. Our classification model achieved a 91.2% accuracy for the early period (03/2020-05/2020) and was applied to the Delta (07/2021-09/2021) and Omicron (12/2021-03/2022) periods for case identification. We identified 310, 8794, and 12,094 COVID-positive authors in the three periods, respectively. The top five common symptoms extracted in the early period were coughing (57%), fever (55%), loss of sense of smell (41%), headache (40%), and sore throat (40%). During the Delta period, these symptoms remained as the top five symptoms with percent authors reporting symptoms reduced to half or fewer than the early period. During the Omicron period, loss of sense of smell was reported less while sore throat was reported more. Our study demonstrated that NLP can be used to identify COVID-19 cases accurately and extracted symptoms efficiently.

Figure 1

Classification model overview. The first row shows that we convert the original posts into chunks and feed them into the BERT-Large for Sequence Classification model to get chunk scores. The second row shows that the weights of chunk scores are trained with deep neural networks, and the final prediction is obtained by weighting chunk scores.

Figure 2

QuadArm model overview. Model inputs are positive posts and outputs are symptoms named by the UMLS. The model consists of four steps: BERT/BioBERT for question answering, dual-corpus expansion, adaptive rotation clustering, and mapping.

Figure 3

Daily trends in number of COVID-19 cases reported to the CDC and we extracted, for the corresponding three periods.

Figure 4

Symptom clustering and trending based on our model QuadArm with BioBERT. Panel (a) shows the clustering results of COVID-19 symptoms through t-SNE visualization for the SARS-CoV-2 early period, Delta period, and Omicron period, respectively. Panel (b) is a ThemeRiver plot showing the change of ten common symptom frequencies over time in the three COVID-19 periods.

Figure 5

Comparison of symptoms extracted by our model (QuadArm with BioBERT) in the two virus variation periods. Panel (a) shows the top 15 commonly reported symptoms for each period. Panel (b) includes two Chord diagrams showing the co-appearance relationship between symptoms for each period. The width of the connection between two symptoms represents the number of authors with both symptoms.

Figure 6

A closer look of the COVID-19 symptom corpus system (Omicron variant period). The left column is the COVID-19 key-word corpus obtained from our dual-corpus expansion method. The middle column is refined symptoms after applying the key-words corpus on the marked answers. The right column is the final standardize medical symptom names obtained by mapping to UMLS. The width of the connection line represents the number of corresponding authors.