Amazon ML Challenge 2024

Entity Value Extraction

In this hackathon, our goal is to develop a machine learning model that can extract entity values (such as weight, dimensions, etc.) directly from product images. This is especially useful in fields like healthcare, e-commerce, and content moderation, where accurate product information is crucial for digital stores.

Data Description

The dataset consists of the following columns:

• index: A unique identifier (ID) for the data sample.
• image_link: Public URL where the product image is available for download. Example link - https://m.media-amazon.com/images/I/71XfHPR36-L.jpg. To download images, use the download_images function from src/utils.py. See sample code in src/test.ipynb.
• group_id: Category code of the product.
• entity_name: Product entity name. For example, “item_weight”.
• entity_value: Product entity value. For example, “34 gram”.

⚙️ Approach Overview

---

1. 🔍 Text Extraction using PaddleOCR:

   • We use PaddleOCR to extract text from the images.
   • This tool helps retrieve essential textual information from images accurately.

2. 🧹 Text Preprocessing:

   • After extraction, the text is cleaned and preprocessed.
   • We remove any irrelevant characters and inconsistencies to make it easier to recognize entities.

3. 📑 Named Entity Recognition (NER):

   • A custom-trained NER model is used to identify key entity values such as weight, voltage, and dimensions.
   • The model predicts both the entity_value and the corresponding entity_name by locating their start and end indices.

4. 🧮 Rule-based Recognition:

   • If the NER model fails, we fall back to Rule-based Recognition.
   • This uses regular expressions (regex) to detect entities based on patterns (e.g., numerical values followed by units like "5.0 kg" or "220 volts").

5. ✅ Final Entity Extraction:

   • The extracted entities are finalized

Flowchart

---

LeaderBoard 🔗:

Team Algorithm Alchemists :

• Akshay Nagamalla @AkshayNagamalla
• Darsh Agrawal @DarshAgrawal14
• Areeb Akhter @Areeb-Ak
• Ayush Reddy @RahZero0