14th Recommender Systems Netherlands meetup @ RTL

Recommender Systems Netherlands - RecSysNL
Recommender Systems Netherlands - RecSysNL
Public group

Barend en Van Dorpweg 2

Barend en Van Dorpweg 2 · Hilversum

How to find us

RTL is just a short walk from Hilversum Media Park station. Follow the passageway over the road. About halfway, go down the winding staircase that looks like an emergency exit. Turn left and follow the signs indicating the main pedestrian route.

Location image of event venue


After a long break, RecSys is back!

We look forward to welcoming you to the 14th RecSysNL meetup, hosted by RTL in Hilversum. We'll have our usual program with one talk from academia by Jay Kim (TU Delft), and one from industry, by Anca Dumitrache (FD Mediagroep). In addition, Marijn Koolen will give us a quick flashback and summary of last October's RecSys2019 conference.

Doors open at 18:00, talks start at 18:30.
Drinks and Pizza will be provided.

Talk 1:
"Towards Seed-Free Music Playlist Generation: Enhancing Collaborative Filtering with Playlist Title Information" by Jay Kim (TU Delft).

In this presentation, Jay proposes a hybrid Neural Collaborative Filtering (NCF) model trained with a multi-objective function to achieve a music playlist generation system. The proposed approach focuses particularly on the cold-start problem (playlists with no seed tracks) and uses a text encoder employing a Recurrent Neural Network (RNN) to exploit textual information given by the playlist title. To accelerate the training, we first apply Weighted Regularized Matrix Factorization (WRMF) as the basic recommendation model to pre-learn latent factors of playlists and tracks. These factors then feed into the proposed multi-objective optimization that also involves embeddings of playlist titles. The experimental study indicates that the proposed approach can effectively suggest suitable music tracks for a given playlist title, compensating poor original recommendation results made on empty playlists by the WRMF model.

Jaehun(Jay) Kim currently is seeking his Ph.D. under the supervision of Dr. Cynthia C. S. Liem and Prof. Dr. Alan Hanjalic at TU Delft. The main research interest of him is the interpretability of the complex recommender systems. Specifically, highly non-linear systems such as deep learning-based models are the main challenge for his projects. He is an almost-retired freelancer music writer, which fuels and motivates his research more oriented mostly on the music recommendation. He received his MSc in 2015 from Seoul National University and BA in 2013.

Talk 2:
Intermezzo by Marijn Koolen (Royal Netherlands Academy of Arts and Sciences), who gives a summary of the ACM RecSys 2019 conference that was held in September in Copenhagen, Denmark.

Talk 3:
"Experiments with Recommending Financial News" by Anca Dumitrache from FD Mediagroep.

This talk will discuss a series of experiments to build a recommender system for news articles in the financial domain. Recommending news means a continuous cold-start problem, that can be tackled by issuing content-based recommendations. Other interesting challenges for this use case are learning from implicit feedback, and handling a large number of user interactions. The proposed approach uses Gradient Boosted Decision Trees (GBDT) to learn from a diverse set of features related to the article content, aggregated user reading behavior, and composite user-article features (e.g. the set overlap between the article's tags and the user's most read tags). The talk will discuss a selection of experiments used to select what model to use, what is the trade-off between data volume and recency, and feature ablation for the model. Finally, a description will be given of how the data pipeline was implemented to bring this model to production.

Anca Dumitrache is a Senior Data Scientist at FD Mediagroep, working on a Google DNI project on personalized news feeds. She graduated from her PhD at Vrije Universiteit Amsterdam, where she studied how to capture and interpret inter-annotator disagreement in crowdsourcing, and how to use this disagreement to get better training data for natural language processing models.

Looking forward to seeing you all there!