Block Stars - AI Match Mapping
An AI-based Match Mapping System that automatically maps scraped sports matches to their corresponding standard API matches. The system compares team names, match time, and other relevant details to identify the same event with high accuracy, reducing manual effort and minimizing incorrect mappings.
Block Stars
Category:
AI & ML
My Role:
AIML Engineer
The objective of this project is to develop an AI-powered Match Mapping System that automatically identifies and maps the same sporting event from two different data sources: a scraped sports feed and a standard API feed. Currently, this mapping is performed manually by comparing team names, match time, and tournament details, which is time-consuming and prone to human error.
The proposed solution uses Natural Language Processing (NLP), semantic similarity models, and rule-based validation to automate this process. The system compares team names, match timings, and other relevant metadata to determine whether two records represent the same real-world match. If the confidence score is sufficiently high, the system automatically maps the records; otherwise, it flags them for manual review.
Approach in this project
I started the project by collecting scraped match data and standard API match data, followed by data preprocessing to normalize team names, match times, and other relevant information. Applied Rule-based validations, such as time tolerance, blacklist keyword filtering, and home-away swap detection, to narrow down potential match candidates and eliminate invalid comparisons.
The filtered matches are then evaluated using a combination of string similarity and AI-based semantic similarity techniques to determine how closely they match. Based on the calculated confidence score, high-confidence matches are automatically mapped, while low-confidence cases are flagged for manual review. This hybrid approach improved the mapping accuracy while significantly reducing the manual effort required.
Developed an AI-assisted match mapping system. The solution reduces manual effort, improves mapping accuracy, and identifies uncertain matches for manual review, making the overall mapping process more efficient and reliable. The initial implementation is developed as an offline batch-processing application, where historical match data is processed from CSV or database exports. Once validated, the same matching engine can later be integrated into a real-time pipeline for live sports data processing.
Related Projects
