I’m making an interactive app for Paediatric COVID-19 academic publications.
The main component is a filterable and sortable datatable of publications, ranked by relevance to Paediatric Clinicians.
i’m using a Naive Bayes classifier on the full text of each document to get a ranking based on the log odds of belonging to the Paediatric Class (based on 4000 manually classified documents) but I would also like the model to improve based on additional classified documents from users of the app.
To do this i thought it would be possible to use the muti row_select option as input in a callback to create a list of indexes i can add to the model for a certain class.
I want users of the app to select the documents that should not be present in the data_table. These then get added to a list of articles to review. If an article is marked inappropriate 3 times then it should be added to the model as a non-paediatric class document.
Is there a reasonable way to accomplish this so that the list of indexes is sent as a text file to a repository that i can retrieve and feed back in to the model?
If anyone is keen to help with this project i’d also be keen to hear any ideas.
I am a Paediatrician in the UK, working on an MD in text and datamining, reasonable python programming but new to dash and plotly.
The end product should be a living review of paediatric papers with routine Pubmed Searches, full text download, classification and biomedical entity extraction for filtering.
here is the basic format of the table and the row selection i want to use