Update README.md

README.md

@@ -19,4 +19,35 @@ task_categories:
 - text-classification
 task_ids:
 - multi-label-classification
----
+---
+## Data Description
+Long-COVID related articles have been manually collected by information specialists. As a certain amount of data is needed
+to train a deep learning-based model, data from two different
+sources have been merged in case of positive examples. To get
+further negative examples, we used a third resource.The first subset was provided by information specialists
+from the Robert Koch Institute and has been collected in the
+following way:
+Katharina
+The second subset is retrieved from the ”Long covid research
+library” released by Pandemic-Aid Networks, who collect
+”important papers that have been published on Long Covid” [2].
+We retrieved 195 articles on January 4th, 2022. As these are all
+positive examples, we needed further negative examples to have
+a balanced training data set. Therefore, we used the database
+LitCovid [5, 6] and filtered for non-long-COVID articles (query:
+NOT e condition:LongCovid) and retrieved further 62 articles.
+As we wanted to train a model on manually curated data rather
+than using semi-automatically classified data - as implemented
+in LitCovid - we did not include further documents from there.
+Preliminary experiments revealed that using more documents
+from LitCovid, also for both classes, lowers the performance,
+when evaluated on the manually curated data sets. The data
+sets have been merged, shuffled randomly and split into training,
+development and test sets.
+
+## Size 
+||Training|Development|Test|Total|
+|--|--|--|--|--|
+Positive Examples|215|76|70|345|
+Negative Examples|199|62|68|345|
+Total|414|238|138|690|