AdSpotlight
- #Marketplace
AdSpotlight – Gathers diverse commercial videos for insightful analysis.
- Computer Vision
Impact
- Successfully collected over 200,000 unique English commercial videos
- Dataset meets the 80% accuracy threshold for identifying and extracting commercial videos
- Provides a solid foundation for advanced analytics and in-depth analysis of commercial content
- Future potential for improvements and expansions
- Ongoing data collection and method refinement to enhance accuracy and diversity
- Enables more comprehensive analytics and valuable insights into commercial content
Services we provided
The solution to collect commercial videos for in-depth analysis and insights.
Tech Stack
Scikit-Learn
Youtube DL
Deepgram
NLP
Challenges and Solutions
🧐 Challenges
- Combining free sources and open datasets to collect commercial videos
- Developing methods and algorithms to accurately identify and extract commercial videos with a min. accuracy of 80%
- Addressing the challenge of limited training data for neural networks by using a TF-IDF-based classifier
- Conducting manual labelling to evaluate and validate the results, ensuring the quality and accuracy of the collected commercial videos
💡 Solutions
- Developed a robust solution for the project, utilizing data parsing, statistical analysis, TF-IDF-based classifiers, manual labelling, and speech-to-text solutions
- Platform collects and analyzes commercial videos
- It provides valuable insights for optimizing marketing strategies and campaigns
User flow
1. The system collects videos that might be commercials from input sources
2. The system analyses video and determines whether it is a commercial or not
3. The system collects additional data that is useful for analysis
4. As output user receives commercial videos with their stats and additional analysis