#15: Podcast Recommendations in the ARD Audiothek with Mirza Klimenta
Apr 27, 2023
auto_awesome
A senior data scientist discusses podcast recommendations in the ARD Audiothek, exploring algorithms, mitigating bias, collaborative filtering, and content-based recommendations. He also talks about responsibility in providing diversified content suggestions and shares insights on becoming a novelist.
Start with simple methods for recommender systems before advanced techniques to prioritize diversified content over precision.
Combining embeddings from USC, transcripts, and images enhances recommender system accuracy, diversity, and novelty metrics.
Utilize inverse propensity weighting and logarithmic functions to reduce popularity bias and enhance content diversity in recommendations.
Deep dives
Importance of Simple Approaches in Recommender Systems
Emphasizing the significance of starting with simple methods before advanced techniques in recommender systems to fully utilize their potential. Specifically, highlighting the need to prioritize diversified content over precision, utilizing embeddings to calculate inter-list similarity for content diversity.
Drifting Through Educational Background and Career Start
Exploring Mirza Clemente's academic journey from a fast-tracked PhD in graph embedding to real-world data science experience, including postdoctoral research and work in recommender systems. Detailing challenges and successes in algorithm efficiency and award-winning contributions to graph drawing conferences.
Enhancing User Engagement Through Algorithm Tweaking
Discussing the enhancement of the recommender system's accuracy, diversity, and novelty metrics by combining embeddings from USC, transcripts, and images. Experimenting with personalized user embeddings based on browsing history and weighted averages to optimize recommendations based on confidence intervals for improved user engagement.
Key Points: Enhancing Diversity and Honing in on Popularity Bias
To improve content diversity in recommendations, the podcast delved into utilizing inverse propensity weighting to re-rank items based on popularity scores. By implementing logarithmic functions on probabilities, they aimed to keep the reorder results aligned with the original while reducing popularity bias. Business rules, like genre diversity constraints, also played a role in diversifying content. The discussion extended to utilizing inverse propensity scoring as a post-processing step to address popular debiasing in comparison to other approaches.
Key Points: User Group Studies and Novelty Exploration
Exploring user demographics for underserved groups and preferences based on content duration, the podcast highlighted efforts to cater to various user behaviors. Strategies included adaptive recommendations for users with diverse durations and interests, with an aim to tackle the cold start problem. Future plans involved building user simulation systems to optimize offline evaluations and align them with A.B. testing results, enabling efficient testing and validation of recommender strategies.
In episode 15 of Recsperts, we delve into podcast recommendations with senior data scientist, Mirza Klimenta. Mirza discusses his work on the ARD Audiothek, a public broadcaster of audio-on-demand content, where he is part of pub. Public Value Technologies, a subsidiary of the two regional public broadcasters BR and SWR.
We explore the use and potency of simple algorithms and ways to mitigate popularity bias in data and recommendations. We also cover collaborative filtering and various approaches for content-based podcast recommendations, drawing on Mirza's expertise in multidimensional scaling for graph drawings. Additionally, Mirza sheds light on the responsibility of a public broadcaster in providing diversified content recommendations.
Towards the end of the episode, Mirza shares personal insights on his side project of becoming a novelist. Tune in for an informative and engaging conversation.
Enjoy this enriching episode of RECSPERTS - Recommender Systems Experts.