Automatic classification of policy and control acts on government

Use case ID: 001

Author: IT Department, Chamber of Deputies of Italy

Date: 22 May 2024

Objective: 

Provide an automatic classification, according to the EuroVoc thesaurus, of policy and control acts on government, using a machine learning algorithm for classification.

Actors: 

• Business units
• Classification system

Prerequisites:

• Existing database of documents as training data
• Knowledge of the EuroVoc thesaurus
• AI classification model trained to assign labels to the documents

Scenario:

1. A new policy and control act is submitted and inserted inside the database.
2. The classification model suggests the most suitable EuroVoc labels for that act.
3. The user can either accept or modify the list of labels.
4. The labels associated with the act are stored as metadata.

Expected results:

• The process of assigning EuroVoc labels to documents is more efficient.

Potential challenges:

• Unbalanced dataset
• Non-interpretable model (e.g. neural networks)
• Long training time (weakly scalable in training data size)
• Some kinds of models could not be robust to outliers
• Avoid overfitting

Data requirements:

• Policy and control acts database

Integrations with other systems:

• Policy and control acts database

Success metrics:

• Reduction of the time needed to classify policy and control acts
• Percentage of policy and control acts classified without any modification made by humans to the labels
• Average response time needed to classify