CinePulse: OTT Platform Recommendation System Project

Abstract

This paper presents CinePulse, an intelligent machine learning–based OTT recommendation system developed to assist viewers in efficiently discovering movies and web series that closely align with their tastes and preferences. The system analyzes rich content metadata, including genres, cast, crew, descriptions, release year, and audience ratings, and applies a cosine similarity–based content filtering approach to identify meaningful relationships between titles in a multidimensional feature space. A robust hybrid architecture is implemented where the React.js frontend delivers an interactive, responsive user interface, while FastAPI acts as the communication layer between the client application and a Python-driven recommendation engine. The model is trained on a structured dataset compiled from multiple OTT sources, enabling the generation of highly relevant suggestions without requiring user login credentials or historical viewing patterns. CinePulse supports features such as real-time search, metadata-based filtering, dynamic content suggestions, and seamless UI interactions, ensuring an engaging discovery experience for users. The results demonstrate that similarity-based recommendation techniques can effectively generate precise and scalable OTT recommendations using metadata alone, making CinePulse a lightweight yet powerful solution for modern content discovery platforms.

Introduction

With the rapid growth of OTT platforms, users often struggle to find relevant content due to the overwhelming number of available movies and web series. CinePulse addresses this problem by providing a machine learning–based recommendation system that suggests content based on metadata such as genre, cast, crew, release year, and ratings, rather than relying on user history.

The system uses a cosine similarity–based content filtering approach to identify similar titles and deliver accurate recommendations. It features a modern architecture with a React.js frontend, a FastAPI/Node.js backend, and a Python-based machine learning engine. This modular design ensures efficient data processing, scalability, and smooth user interaction.

CinePulse follows a structured workflow where user input is processed through API layers, transformed into feature vectors, and analyzed using a KNN-based similarity model to return the top recommendations. The system avoids the cold-start problem and works effectively even for new users.

Experimental results show that CinePulse provides fast (≈210 ms latency), accurate, and scalable recommendations. It performs especially well when metadata features are descriptive, proving that metadata-driven models can effectively replace more complex user-based systems. Overall, CinePulse enhances content discovery by reducing search effort and improving recommendation relevance in OTT platforms.

Conclusion

The CinePulse project presents an efficient and accurate metadata-driven approach to OTT content recommendation. By using structured features such as genre, cast, crew, release year, and audience ratings, the system delivers relevant Top-5 recommendations without requiring user profiles or viewing history. The cosine similarity model, combined with an optimized preprocessing pipeline, ensures strong relevance across diverse content categories. With an average response latency of 215ms, CinePulse supports smooth real-time content discovery for modern OTT platforms. The system’s modular architecture built with a React frontend and a FastAPI backend powered by a Python recommendation engine offers scalability, responsiveness, and seamless communication. Evaluation results show that CinePulse outperforms traditional keyword-based and single-attribute filtering systems in both accuracy and latency. User interaction analysis further confirms its practicality and engaging experience. Overall, CinePulse serves as a lightweight and interpretable recommendation framework suitable for OTT streaming environments. Although the current system relies solely on metadata, it provides a solid foundation for enhancements such as semantic text embeddings, personalized user modeling, and approximate nearest-neighbor retrieval. CinePulse represents a meaningful step toward building accessible and intelligent OTT recommendation systems.

References

[1] F. Ricci, L. Rokach, and B. Shapira, Recommender Systems Handbook, 2nd ed. New York, NY, USA: Springer, 2015. [2] B. Sarwar, G. Karypis, J. Konstan, and J. Riedl, “Item-based collaborative filtering recommendation algorithms,” in Proc. 10th Int. Conf. World Wide Web, 2001, pp. 285–295. [3] T. Cover and P. Hart, “Nearest neighbor pattern classification,” IEEE Trans. Inf. Theory, vol. 13, no. 1, pp. 21–27, Jan. 1967. [4] S. Das, “Goodreads Books Dataset,” Kaggle, 2017. [Online]. Available: https://www.kaggle.com (Dataset reference for book metadata). [5] L. Rokach and O. Maimon, Data Mining With Decision Trees: Theory and Applications, 2nd ed., World Scientific, 2014. [6] F. Pedregosa et al., “Scikit-learn: Machine Learning in Python,” J. Mach. Learn. Res., vol. 12, pp. 2825–2830, 2011. [7] M. Abadi et al., “TensorFlow: Large-scale machine learning on heterogeneous systems,” 2015. [Online]. Available: https://www.tensorflow.org/ [8] [8] D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” in Proc. 3rd Int. Conf. Learning Representations (ICLR), 2015. [9] [9] E. Alpaydin, Introduction to Machine Learning, 4th ed. Cambridge, MA, USA: MIT Press, 2020. [10] [10] S. Russell and P. Norvig, Artificial Intelligence: A Modern Approach, 4th ed. Pearson, 2021. [11] M. Grinberg, Flask Web Development, 2nd ed. O’Reilly Media, 2018. [12] Node.js Foundation, “Node.js Documentation,” 2024. [Online]. Available: https://nodejs.org [13] J. J. Carroll, “Metadata and semantic enrichment for digital libraries,” J. Inf. Sci., vol. 45, no. 3, pp. 365–379, 2019. [14] X. Amatriain and J. Basilico, “Recommender systems in industry: A cross-industry analysis,” ACM Queue, vol. 14, no. 1, pp. 58–75, 2016. [15] Y. Koren, R. Bell, and C. Volinsky, “Matrix factorization techniques for recommender systems,” IEEE Computer, vol. 42, no. 8, pp. 30–37, Aug. 2009. [16] Y. Bengio, A. Courville, and P. Vincent, “Representation learning: A review and new perspectives,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 35, no. 8, pp. 1798–1828, Aug. 2013. [17] J. Leskovec, A. Rajaraman, and J. Ullman, Mining of Massive Datasets, 3rd ed. Cambridge University Press, 2020.

Copyright

Copyright © 2026 Dr. G Ganapathi Rao, Bavandla Sathwik, Noomuri Sathwik, Gopu Rishika. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.