text classification application

upload your dataset, select your input and target columns, train model, and predict.

app workflow

upload page

data pipeline + classification workflow

upload page

app preview

upload page

preview page

summary page

report page

predictions page

model summary + findings

algorithm	accuracy	training time	architecture	incremental learning	performance	pros	cons
multinomial nb	good	fast	probability-based	✔	efficiently handles large datasets	fast	least accurate
logistic regression	excellent	slow	sigmoid function	✖	efficiently handles large datasets	-	-
svm	excellent	slow	decision boundary	✖	computationally expensive	most accurate	long training
sgd	excellent	fast	decision boundary	✔	-	fast and accurate	-
xgboost	excellent	slow	decision trees	✖	efficiently handles large datasets	-	occasionally fails
cnn	excellent	very slow	1d cnn	✖	-	very accurate	hyperparameter tuning takes time
fasttext	excellent	fast	shallow nn model architecture. hierarchical softmax, tree-like structure	✖	efficiently handles large datasets	minimalistic; balancing not required; built-in multilabel	lacks some features

algorithms used

for this specific application, i chose:

multinomial nb
support vector machine (svm)
stochastic gradient descent (sgd)
fasttext by facebook research

as they were the most efficient and accurate models for a wide variety of datasets.

folder structure

.
└── text-app/
    ├── backend/
    │   ├── csv/
    │   ├── datasets/
    │   ├── downloads/
    │   ├── executions/
    │   ├── logger.py
    │   ├── server.py
    │   └── sharepoint.py
    ├── build/
    ├── env/
    ├── logs/
    ├── node_modules/
    ├── public/
    │   └── assets/
    ├── screenshots/
    ├── src/
    │   └── components/
    ├── .env
    ├── .gitignore
    ├── config.ini
    ├── package-lock.json
    ├── package.json
    ├── README.md
    └── requirements.txt

linking with sharepoint

this app can be configured to store the files in sharepoint or on the server (local). to switch to sharepoint storage, set sharepoint in config.ini under the section storage to true. similarly, you can set it to false to use server storage.

this is the redacted.

sharepoint folder structure

.
└── sites/
    └── text-classification/
        └── Shared Documents/
            ├── executions/
            ├── datasets/
            ├── csv/
            └── config.ini

note: do not change configuration during the execution. this will cause errors!

how to use

prerequisites

install node.js
install npm (node package manager)

create react app

change to project directory cd <repo-name>

create react app npx create-react-app .

copy over all the files in the bitbucket into the app you just created.

find more information on creating a react app.

install dependencies for react front-end

npm install @lottiefiles/react-lottie-player
npm install axios
npm install csv-parser
npm install fs
npm install papaparse
npm install react-csv-reader
npm install react-drag-drop-files
npm install react-tooltip

your package.json should look something like this:

"proxy": *redacted*,
"dependencies": {
    "@lottiefiles/react-lottie-player": "^3.5.4",
    "@testing-library/jest-dom": "^5.17.0",
    "@testing-library/react": "^13.4.0",
    "@testing-library/user-event": "^13.5.0",
    "axios": "^1.7.2",
    "csv-parser": "^3.0.0",
    "fs": "^0.0.1-security",
    "papaparse": "^5.4.1",
    "react": "^18.3.1",
    "react-csv-reader": "^4.0.0",
    "react-dom": "^18.3.1",
    "react-drag-drop-files": "^2.3.10",
    "react-scripts": "5.0.1",
    "react-tooltip": "^5.27.1",
    "web-vitals": "^2.1.4"
}

note: the versions above might be different to yours

install dependencies for flask back-end

pip install -r requirements.txt

starting the front-end (react)

npm start runs the app in the development mode. open http://localhost:3000 to view it in your browser.

starting the back-end (flask api)

by default, the app is connected to the api hosted on the server at redacted.

you can use a locally hosted api by running python backend/server.py and it will be hosted at http://localhost:5000. do note that you will have the change the endpoint url in each component file in src/components via the apiurl variable to point to this local host. this can be done through editing the global variable in the .env file.

notes:

make sure to use a semicolon “;” to comment out lines in the .env file
if using the server-hosted api, the config.ini file on the server is the one that will be used. otherwise, it will use config.ini on your local machine.

deployment/changing production code

update changes to the files on the server (changes made locally must be reflected on the server)
run npm run build in local machine
make sure to copy the latest build file to the server

you can find the deployed app at redacted

see the section about deployment for more information.