A Step-by-Step Guide to the Microsoft News Recommendation Competition | Towards Data Science

Abstraction: Walkthrough of NRMS algorithm and MIND competition submission process

Key points:

• NRMS (Neural Recommendation with Multi-Head Self-Attention) is a content-based algorithm using multi-head self-attention for word/news representations plus additive attention for importance weighting
• Five competition baselines available: DKN, LSTUR, NAML, NPA, and NRMS; NRMS outperforms others on MIND in the ACL 2020 evaluation
• MIND dataset: MINDlarge for competition, MINDdemo/MINDsample for prototyping; training/dev/test phases on CodaLab
• Submission: train model using provided Jupyter notebook, save predictions to zipped folder, upload to CodaLab; register via email to mind@microsoft.com
• Potential improvements noted: add positional encoding, model long-term user preferences across impressions, apply Graph Neural Networks on user-behavior graphs

Connections: Microsoft · Microsoft Recommenders · News Recommendation · Recommendation Systems · Attention Mechanism

Source: https://towardsdatascience.com/a-step-by-step-guide-to-the-microsoft-news-recommendation-competition-700ab00831a