Data Processing (What Works When for Whom?)

Data processing scripts of the e-Mental Health project. The scripts deal with two data sets: OVK and Tactus. Data related to our paper "De-identification of Dutch Medical Text" can be found in the data repository

OVK mails

The OVK data consist of Word files with e-mails. First, each Word file was converted to text: abiword -t text 1234.docx # creates file 1234.text

Next, extra lines have been added to the text files to indicate where a new email started. A typical separator line was "Date: 1 Jan 2000": ovkPrepare.py 1234.text > 1234.prepared # expects a numeric file name The program tries to guess the author of an email but this sometimes fails. The missed names are indicated by the string "???" and need to be added manually.

The files contain personal information, which needs to be removed. Therefore all names and numbers were removed from the file. This process contains two steps: 1. named entity recognition (ner), and 2. anonymization: ner.py < 1234.prepared > 1234.ner # requires frog, see comment below anonymize.py 1234.ner # creates file 1234.ner.out, see comment below

For named entity recognition, we rely the program frog, which is part of the LaMachine package. After installing the package, we run it as follows: docker run -p 8080:8080 -t -i proycon/lamachine lamachine@abcd1234:/$ frog -S 8080 --skip mcpla After starting LaMachine like this, the ner.py program is able to process the texts.

The anonymization process (anonymize.py) is interactive. Each new entity will be displayed on the screen and the user is required to enter the entity type (like PER, ORG or LOC) or an empty string (which stands for: no entity).

In the file 1234.ner.out, all names and numbers have been replaced by dummy strings (like PER for person names). Mail headers signalling the start of a new mail, start on a new line containing a capitalized word followed by a space and a colon (like: "Date :"). All other lines contain a single sentence.

Finally, the output files of the anonymization process can be converted to a csv table which can be used for analysis: ovk2table.py 1234.ner.out ... > all.csv The program ovk2table.py assumes that the files contain emails in chronological order. The local file reversed.txt contains the names of files with the emails in reversed chronological order.

The Jupyter notebooks pca.ipynb and pca-results.ipynb can be used for analyzing the data in the csv tables. pca.ipynb uses principle-component analysis for visualizing the e-mails on a two-dimensional space. It also displays words which are specific for subsets of the emails. pca-results.ipynb performs exactly the same task but next to building models from the words in the emails, it also displays information on the treatment progress.

Tactus mails

The Tactus data consist of XML files with emails. These can be converted to csv table: tactus2table.py 1234.xml ... # creates file ../output/emails.csv

The texts of the mails need to be anonymized. There is no automatic solution for this yet.

The data in the csv table can be analysed with the Jupyter notebooks pca-tactus.ipynb and pca-tactus-results.ipynb, in the same way as the OVK data.

Metadata

The OVK metadata is stored in the SPSS file opve.sav. The file was converted to csv with R: library(foreign) data <- read.spss("file.sav") write.csv(data,file="file.csv") The Tactus metadata is stored in the XML file of each client. The program tactus2table.py extracts these and stores them in five files in the directory ../output (Intake.csv and Lijst.*)

Installation

clone the repository
git clone git@github.com:eriktks/data-processing.git change into the top-level directory
cd data-processing

Dependencies

• Python 3.6

License

Copyright (c) 2018, Erik Tjong Kim Sang

Apache Software License 2.0

Contributing

Contributing authors so far: * Erik Tjong Kim Sang