nurnils/citizen-portal-with-artificial-intelligence

The main purpose of this web application is the conceptual design and development of a citizen portal with an integrated artificial intelligence during the 3rd and 4th semester at DHBW Stuttgart.

Citizen Portal

Welcome to the new intelligent citizen portal.

About

This project was developed by Nils-Christopher Wiesenauer (NurNils) on behalf of evia solutions GmbH during the 3rd and 4th semester as part of the module T2000 at DHBW Stuttgart. The main purpose of this web application is the conceptual design and development of a citizen portal with an integrated artificial intelligence.

💚 The system is based on the MEAN (MongoDB, Express.js, Angular, Node.js)-Stack with the CRUD (Create, Read, Update, Delete) functionality in the backend.

🧠 Brain.js is used for the integrated artificial intelligence. More info

🎓 The project was rated with a 1.5

Table of Contents

• Installation and Usage
  • Frontend
  • Backend
• Artificial Intelligence
  • Brain.js
• Technologies
  • MongoDB
  • Express.js
  • Angular
  • Node.js
• Documents
  • Documentation (T2000)
  • Presentation
  • Evaluation

Installation and Usage

Frontend

This Citizen Portal App was generated with Angular CLI version 12.0.2. More Info

1. Run npm install to download all needed packages and it's dependencies.

2. Run ng serve for a dev server. Navigate to http://localhost:4200/. The app will automatically reload if you change any of the source files.

3. Run ng build to build the project. The build artifacts will be stored in the dist/ directory.

Backend

This Citizen Portal API was created with Node.js version 12.16.x. More Info

1. Run npm install to download all needed packages and it's dependencies.

2. Go to the API folder and create a .env file with the following content (update if necessary):

// If PRODUCTION=true: Database connection with authorization below (user and password), otherwise it connects without authorization.
PRODUCTION=false 
PORT=3000
APP_NAME="Citizen Portal"
APP_DOMAIN="citizen-portal.com"
APP_MAIL="[email protected]"
JWT_SECRET="JWT_SECRET"
DB_PORT=27017
DB_HOSTNAME="localhost"
DB_DATABASE="database"
DB_USER="user"
DB_PASSWORD="password"
FILES_IMAGE="image/bmp,image/x-bmp,image/x-ms-bmp,image/cis-cod,image/cmu-raster,image/fif,image/gif,image/ief,image/jpeg,image/png,image/svg+xml,image/tiff,image/vasa,image/vnd.wap.wbmp,image/x-freehand,image/x-icon,image/x-portable-anymap,image/x-portable-bitmap,image/x-portable-graymap,image/x-portable-pixmap,image/x-rgb,image/x-windowdump,image/x-xbitmap,image/x-xpixmap"

3. Run npm start for a Node.js server. Navigate to http://localhost:3000/ or defined port in .env file. The app will automatically reload if you change any of the source files because of nodemon.

Artificial Intelligence

Brain.js

Brain.js is used for the integrated artificial intelligence. To archieve training of manual added tranings from database using strings with inputs and outputs a LSTM neural network was implemented: More Info

🛈 LSTM means long short-term memory and is a subcategory of recurrent neural networks (RNN) that are able to learn long-term dependencies. LSTMs were first introduced in 1997 by Sepp Hochreiter and Jürgen Schmidhuber. The idea of the two was to use three gates. An input gate, a forget gate and an output gate. In this way, the LSTM is able to remember events from earlier experiences compared to general RNNs, thus creating the long short-term memory.

// Initializes long short term memory (LSTM) neural network
const network = new brain.recurrent.LSTM();
// The maximum times to iterate the training data
const networkIterations = 20000;

// Loads and trains neural network
loadNeuralNetwork = () => {
  // Start date to check how long the operation took
  const startDate = new Date();
  // Path to save trained neural network
  const networkPath = './assets/neural-network.json';

  // Checks if neural network already exists
  if (fs.existsSync(networkPath)) {
    const networkData = JSON.parse(fs.readFileSync(networkPath));
    network.fromJSON(networkData);
  }

  // Trains manual added untrained searches from database
  TRAINING.getNotTrainedSearches((err, training) => {
    // Checks if untrained data are available
    if (!err) {
      if (training && training.length > 0) {
        console.log(`Training neural network with ${training.length} new searches...`);
        const trainingData = training.map((item) => ({
          input: item.search,
          output: item.category,
        }));
        network.train(trainingData, {
          iterations: networkIterations,
        });
        fs.writeFileSync(networkPath, JSON.stringify(network.toJSON(), null, 2));
      }
    } else {
      console.log(`Unable to load training data: ${err.message}`);
    }
    // Updates training data to status "trained"
    TRAINING.updateNotTrainedSearches((err, training) => {
      if (!err) {
        const endDate = new Date();
        const minutes = (endDate.getTime() - startDate.getTime()) / 1000 / 60;
        console.log(
          `Successfully loaded neural network in ${minutes} minutes with ${training.nModified} searches`
        );
      } else {
        console.log(`Unable to update not trained searches`);
      }
    });
  });
};

For the intelligent search an endpoint /search was created. To get a category based on a search term, a HTTP GET request must be sent:

router.get(`/`, (req, res) => {
  const data = req.query;
  if (data && data.search && data.lang) {
    let output = network.run(data.search);
    console.log('output', output);
    if (!categories.includes(output.toLowerCase())) output = 'corona';
    ARTICLE.search(output, (err, article) => {
      if (!err && article) {
        res.status(200).send({ status: 'success', data: { article, output } });
      } else {
        res.status(409).send({ status: 'error', message: 'Unable to search' });
      }
    });
  }
});

Technologies

The system is based on the MEAN (MongoDB, Express.js, Angular, Node.js)-Stack with the CRUD (Create, Read, Update, Delete) functionality in the backend.

MongoDB

MongoDB is a document-oriented NoSQL database used for high volume data storage.

Express.js

Express.js is the most popular Node web framework and is the underlying li-brary for several other popular Node web frameworks. It provides many mechanisms.

Angular

Angular is a TypeScript based front-end framework which is published as open source software.

Node.js

Node.js is a JavaScript free and open source cross-platform for server-side programming that allows users to build network applications quickly.

Corona Data

Matching the current situation, the citizen portal additionally displays the current situations of Baden-Württemberg and the associated urban and rural districts. For data in the portfolio of the Bundesinstitut auf dem Gebiet der Krankheitsüberwachung und Krankheitsprävention, the Robert Koch-Institut (RKI) is a federal institute in the field of disease surveillance and disease prevention, and thus also the central federal institution in the field. More Info

The /corona subpage displays statistics of currently available data to citizens. Anybody can quickly get to the desired information.

For the overview of Covid-19 data of Baden-Württemberg, the citizen portal uses data from the Center for Systems Science and Engineering (CSSE) of Johns Hopkins University. Current data from around the world is provided by the CSSE on a daily basis in a GitHub repository in the form of Excel files. More Info

The citizen portal automatically downloads these files every day, generating complete statistics of all available data for the state of Baden-Württemberg.

Documents

Documentation (T2000)

The project documentation (T2000) was written in latex and packaged as a .zip file. A .pdf document was created for submission.

Presentation

A PowerPoint presentation (.pptx) was created for the oral exam. The exam itself took place online.

Procedure:

• 15 minutes presentation of the project work and presentation of the own work.
• 15 minutes of questions from the examination board on the topic of the project work, including the associated basics.

Evaluation

The overall grade for the module subject consists of 50% of the grade for the documentation and 50% of the oral examination (1/2 presentation and 2/2 question paper). Both were rated with a 1.5.