Skawr
- #E-Commerce
A system that accurately eliminates irrelevant results and partial result matching while predicting the most probable category for a given query. It ensures that only one relevant category is displayed for each query, reducing the need for multiple-category queries.
- Machine Learning
Impact
The improved search functionality helped our client to:
- Enhance User Experience: Users quickly find relevant products/services, increasing satisfaction and retention.
- Increase Conversion Rates: Easy access to desired items boosts purchase likelihood, driving revenue.
- Improve SEO: Relevant search results attract organic traffic, benefiting the website.
- Reduce Operational Costs: Fewer support requests due to better search functionality, leading to cost savings and improved efficiency.
Services we provided
Precise text search: Input words, get accurate database matches.
Refined Image Categories: Input images or Rekognition categories, get the top 5.
Tech Stack
Pytorch
Python
Linux (Ubuntu)
Tensorflow (GPU 2.3)
Huggingface
GitHub
GitLab
Flask
Streamlit
Challenges and Solutions
🧐 Challenges
- The system must grasp query context and dataset categories to achieve project goals effectively.
- Distinguishing relevant and irrelevant results, and matching queries to suitable categories without being too broad or narrow.
- Preventing retrieval of partial matches to avoid displaying irrelevant results and improve user experience.
💡 Solutions
We utilized advanced NLP for precise query interpretation and category matching, supplemented by ongoing machine learning to enhance prediction accuracy. The solution that we’ve come up with allows to:
- Eliminate irrelevant results such as accessories, complementary products, and services.
- Remove partial result matches that don’t match the complete phrase. – Provide the most probable predictions for results.
- Display only one category per query to avoid irrelevant categories.
User flow
Enhanced search
Categorization