The app displays the session protocols of the german Bundestag from 1949 till 2017.
Besides that the app provides an Ngram Viewer that displays word frequencies over time for all those protocols. Th Ngram Viewer and its functionality is similar to the [Google Ngram Viewer](https://books.google.com/ngrams).
The n-gram data and the protocols have been created using this software (also part of the same masterthesis): https://gitlab.ub.uni-bielefeld.de/sporada/bundesdata_markup_nlp_software
1. First install `docker` for your OS according to this guide: [https://docs.docker.com/install/](https://docs.docker.com/install/)
2. After that install `docker-compose` for you system according to this guide: [https://docs.docker.com/compose/install/](https://docs.docker.com/compose/install/)
3. Clone this reposiory with `git clone https://gitlab.ub.uni-bielefeld.de/sporada/bundesdata_web_app.git` to a location of your choice.
## Build the app/start the app with docker
1. Navigate into the repository with `cd path/to/the/repository`.
5. Start the web app with `docker-compose up`. Doing this the first time will take some time because all the needed images (postgres, python and nginx) will have to be downloaded. Also all the packages defined in the Pipfile will be installed in the according image. This terminal will stay open for now to show the log messages of the three running containers.
7. The website should be showing up. But it looks kind of broken. To fix this collect the needed static files (css, images, javascripts etc.). Open a second terminal in the same location as the running one and execute `docker-compose run web python manage.py collectstatic`.
1. Befor importing the data we have to setup the tables in the PostgreSQL database. Do this with `docker-compose run web python manage.py makemigrations` followed by `docker-compose run web python manage.py migrate`.
13. Change the owner rights of all files in the repository. This has to be done because every process inside a docker container is always executed with root privilage. Thus the created volumes are not accessable anymore. Change the rights with `sudo chown -R $USER:$USER .` This is only needed for linux systems.
12. Download the folders *MdB\_data* and *outputs* from the link mentioned in [this repository](https://gitlab.ub.uni-bielefeld.de/sporada/bundesdata_markup_nlp_data) and copy those into the folder *input_volume* which is located inside the web app repository on the root level. If the downloaded folders are inside an archive extract the folders first. This folder is a volume which is mounted into the web app container. The contianer is able to read every data inside that volume. Note that the volume is accessed with the path */usr/src/app/input_data* not */usr/src/app/input_volume*.
13. First we have to import the speaker data. This will be done by executing following command `docker-compose run web python manage.py import_speakers /usr/src/app/input_data/MdB_data/MdB_Stammdaten.xml` in the second terminal.
14. After that we can import all the protocols and thus all speeches for every person. The command to do that is `docker-compose run web python manage.py import_protocols /usr/src/app/input_data/outputs/markup/full_periods` (Importing all protocols takes up to 2 days. For testing purposes *dev\_data/beautiful\_xml* or *test\_data/beautiful\_xml* can be used.)
15. Now the n-grams can be imported by using `docker-compose run web python manage.py import_ngrams_bulk 1 /usr/src/app/input_data/outputs/nlp/full_periods/n-grams/lm_ns_year/1_grams lm_ns_year`. This command imports the alphabetically splitted n-grams into their according tables. First parameter of this command is *1*. This tells the function to import the n-grams from the input path as 1-grams. Therefore the second parameter is the inputpath */usr/src/app/input_data/outputs/nlp/full_periods/n-grams/lm_ns_year/1_grams* where the 1-grams are located. The last part of the input path clearly identifies the n-grams as 1-grams. Finally the third parameter identifies what kind of n-grams are being imported. In this case the parameter is set to *lm_ns_year* which means the ngrams are based on lemmatized text without stopwords counted by year. An example to import 2-grams would look like this `docker-compose run web python manage.py import_ngrams_bulk 2 /usr/src/app/input_data/outputs/nlp/full_periods/n-grams/lm_ns_year/2_grams lm_ns_year`. To import 3-grams from a different corpus the command for example should look like this: `docker-compose run web python manage.py import_ngrams_bulk 3 /usr/src/app/input_data/outputs/nlp/full_periods/n-grams/tk_ws_speaker_\(1-3\)/3_grams tk_ws_speaker`. Be careful when importing the n-grams. **If the parameters are set wrong, the n-grams will be imorted into the wrong tables and thus leading to incorect findings using the Ngram Viewer.** If you did something wrong you can reset the database with `docker-compose run web python manage.py flush` and start the data import again. It is possible to import different n-gram sets at the same time using multiple commands in multiple terminals. Just keep an eye out on the CPU and RAM usage. There is also an optional fourth parameter to set the batch size of one insert. The default is set to read 1 million rows from the csv and insert them at once into the database. The parameter `-bs 10000000` would set it to 10 million. Increasing that value also increases the RAM usage so be careful with that.
16. Repeate the step above for every kind of n-gram data you want to import. Importing 1-grams will only take some minutes while importing 5-grams will take several hours. (For testing purposes the n-grams from *dev\_data* can be used.)
A live version of the app is running under http://129.70.12.88:8000/ inside the University Bielefeld network. You have to access the university network via VPN to be able to use the live version. (https://www.ub.uni-bielefeld.de/search/vpn/)
