Automatic identification of responsible authorities for citizen appeals

📌 Project Description

This system automatically determines responsible government entities for citizen complaints by analyzing problem descriptions, categories, and geographic locations to identify the appropriate authority (federal or state level).

Contents

This project contains a simple webserver with a http api that can assign issues to authorities based on the problem region, description and category.

There is also a utility in csv_processor directory that applies the API method localy to a csv file from the example. The guide for using it is located in the directory in the csv_preprocessor/readme.md

Algorithm Overview

Hybrid approach combining:

1. Rule-based matching:
   • Category-to-entity competency mapping
   • Geographic filtering (state/region)
2. Semantic text analysis:
   • Vector embeddings of texts and competencies
   • Cosine similarity calculations of pairwise semantic

Installation and Usage of the webserver

1. Install Dependencies (it is advised to use a virtual environment)

pip3 install -r requirements.txt

2. Run webui (It will take quite a long time on a cold start):

python3 main.py

Now you can open the webui on localhost: http://localhost:7860

Formal solution as required in the problem statement

The solution from the problem is in the folder processed_solution

Evaluation Metric

Impossible to calculate because of a broken dataset (train, validate) since the authority assigned does not come from the federal state defined in the field state and may or may not have a mismatched category. (explained thoroughly in the presentation).

Credits

Hey, there is no commit history but the thanks should be given where its due, to all people who contributed to this repository:

Hlib Zabudko
Eriks Spaks
Sergey Sarkisyan
Nurzhan Zhukesh (me)