Image Recommendation

Description

Using the Maximal Marginal Relevance (MMR) algorithm, this project ranks images by semantic content, drawing from textual annotations. It features an experimental comparison between Word2Vec and GloVe embeddings to optimize recommendation quality.

Installation

git clone https://github.com/hikariakio/Image-MMR-Recommendation.git

Backend Setup

1. Direct to PyBackend.

2. Create and go inside to the virtual environment (python 3.11.6)

3. Install the packages from requirements.txt

4. Download Precomputed Caches

Sim_Matrix_Glove.csv
https://drive.google.com/file/d/1ChvRpdPInB1SdZOp0nr6JUhew6zal7eW/view?usp=sharing

Sim_Matrix_Word2Vec.csv
https://drive.google.com/file/d/1N-3Nt4Tp2DhkfzFoe9ovRzYSYO0-jj2e/view?usp=sharing

5. Download GoogleNews-vectors-negative300

6. start app.py

FrontEnd Setup

1. Direct to NodeFrontend.

2. Install node modules

npm install

3. Start the client

npm start

Image Server Setup

Download dataset and create an http server at port 5001.

http://images.cocodataset.org/zips/val2017.zip

python -m http.server 5001

Screenshot