Image-Knn recommender

Image-based recommendation model developed within Bachelor's thesis of Martin Pavlicek in RecombeeLab at FIT CTU in Prague.

Project structure

ReadMe.txt ............................................................ Brief overview of archive and program manual
|
|-- src................................................................................................ Source codes
|
|-- Bins .......................................................................... Runable version of program files
| |
| |-- 1) ExtractVectors.cmd .............................................. Extracts feature vectors from item images
| |
| |-- 2) ComputeImageKnn.cmd ...................................................... Computes KNN neighbours for item
| |
| |-- 3) ShowSimilar.cmd ............... Starts interactive dialog presenting similar items based in precomputed KNN
| |
| |-- 4) DownloadInteractions.cmd ......................................... Downloads interaction data from database
| |
| |-- 5) RecommendForUser.cmd .......................................... Starts interactive recommendation for users
| |
| |-- DeepRecommender.exe .................................................... Binary version of reccomender program
| |
| |-- VGG16_FeatureExtractor.py .................................... Python script for item image feature extraction

Requirements

.NET Framework 4.6.1 or newer (Windows) OR .NET Core 2.0 or newer (Linux)
Keras neural network framework python package
Any Keras python backend (Tensorflow preferred)
For GPU accelerated computing CUDA 8 or newer is required and TensorflowGPU python package

How to use

Batch files are provided to assist you in bootstrap process.
For help run "DeepRemonneder.exe help".
To show info about specific command run "DeepRecommender.exe [command] help".

Recommendation process is divided into 3 steps:

1. Extraction of feature vectors from images of recommended items
2. Computation of NN neighbours for each recommended item
3. Recommendation for specific user based on Image-KNN algorithm

For quick start follow steps below:

1. Edit provided batch files accoding to your configuration (directories, paths, DB connection string, ...)
2. Run batch files one by one in numbered order
3. For advanced usage run "DeepRecommender.exe [command] (args)"

For more info about this recommendation model see:
Pavlíček, Martin. 2018. Doporučovací modely založené na obrázcích. České vysoké učení technické v Praze, Fakulta informačních technologií