EV Charging Station Recommendation System

Introduction

Our project aims to build a search engine for electric vehicle charging stations with personalized recommendations. The system consists of a naive ranker by distance, and a LLM-based personalization system. The performance of each ranker combination is evaluated with query results from Google Maps API processed with a customized relevance function.

Start the Server

First, initialize the database by running the following command in the terminal:
flask --app electrifind init-db

To start the server, run the following command in the terminal:
flask --app electrifind run --debug

To start the server with auto reload, run the following command in the terminal:
waitress-serve --call 'electrifind:create_app'

The query contains the following fields:

latitude: The latitude of the user.
longitude: The longitude of the user.
user_id: The user ID.

The system will return the top electric vehicle charging stations based on the user's preferences.

Environment

To use the syste, you need to first create a Python virtual environment:
python3 -m venv .venv

Switch to the virtual environment:
source .venv/bin/activate

Install the required packages:
pip install -r requirements.txt

Key Features

Data scraping using GoogleMap/Serp/NREL API
NLP comments using ChatGPT API & prompt engineering
LLM-based embedded vector space personalization.

Project Layout

/project2-team7
├── .venv/
├── data/
│   ├── encoded_station.npy
│   ├── encoded_user_profile.npy
│   ├── GPT_analysis_AA_DTW.csv
│   ├── mergd_reviews_and_predicted_dataset.csv
│   ├── NREL_RAW.csv
│   ├── row_to_docid.txt
│   ├── station_personalized_features.csv
├── src/
│   ├── __init__.py
│   ├── models.py: Models for the system
│   ├── pipeline.py: Pipeline for the whole system, use it as main!
│   ├── ranker.py: Rank the documents
│   ├── utils.py: Utility functions
│   ├── vector_ranker.py: Vector ranker
├── notebook/
|   ├── Deep Learning Factorization model.ipynb
│   ├── evaluation.ipynb
|   ├── Google_Map_Find_API.ipynb
│   ├── Google_Map_Review_API.ipynb
│   ├── GPT_prompt.ipynb
│   ├── GPT_sentiment_analysis.ipynb
│   ├── Joint_Data_Model.ipynb
│   ├── NREL_processing_numerical.ipynb
│   ├── NREL_processing.ipynb
│   ├── Review_Feature_Extractor.ipynb
│   ├── user_profile.ipynb
├── .gitignore
├── README.md
├── requirements.txt

Methodology

Data Collection

encoded_station.npy: This file contains the encoded data of the electric vehicle charging stations in the United States.
encoded_user_profile.npy: This file contains the encoded data of the user profiles.
GPT_analysis_AA_DTW.csv: This file contains the metadata and score of the electric vehicle charging stations in the United States.
mergd_reviews_and_predicted_dataset.csv: This file contains the merged data after LLM and GPT analysis.
NREL_raw.csv: This file contains raw data from NREL API of electric vehicle charging stations in the United States.
row_to_docid.txt: This file contains the mapping of the row index to the document ID.
station_personalized_features.csv: This file contains the personalized features of the electric vehicle charging stations in the United States.

Notebooks

Google_Map_Review_API.ipynb: This notebook contains the code for scraping the reviews of the electric vehicle charging stations in the United States using the Google Map API.
GPT_prompt.ipynb: This notebook contains the code for analyzing the reviews of the electric vehicle charging stations in the United States using the ChatGPT API.

priest-yang/Electrifind