Problem Statement Description

Build a "Spam Buster" for user generated IMAGE content based on predefined text, links identified as "known bads", based on which the image should be classified as spam or not-spam. The "known bads" can be updated anytime. The "Spam Buster" would need to analyze at image level, identify text components inside image and build appropriate logic to give a score to the image.

A similar used case is employed by Facebook Ads technology to identify if the image ad contains too much text or contains any objectionable text."

---

Solution

We build a flask based micro-service which gives a "Spaminess score" for any uploaded image. The score is derived by aggregating the scores from two models:

1. Text based spam classifier - Text is extracted from the image and a pre-trained model runs on the text to classify text in respective bucket
2. Image based spam classifier - Convolution neural network which classifies image as spam or not based on the image features

---

Instructions

Work with the spam classifier models

Image to Text

Image Spam Classifier

Text Spam Classifier

Run locally

Clone the repo and cd into predict folder

pip3 install -r requirements.txt

cd predict

python3.6 app.py 

Now, Open this link in your browser...  
http://localhost:5000/spam_buster/api/v1/model

Run within docker

Clone the repo and cd into predict folder

Run the image and bash into it to start the server

docker run --rm -itd \
 --name spam_buster \
 -v image-spam/predict:/predict \
 -p 5000:5000 \
 kayush206/img_spam:v2 bash

cd /predict

python3.6 app.py

Now, Open this link in your browser...  
http://localhost:5000/spam_buster/api/v1/model

---