Tech*Research

This is an account to tweet scientific articles/books/papers I've read.

[Tech*Research @TechResearch5]

# Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5)
Large scale pretrained language modelを言語タスクに変換した， recommendation task でfine-tune... 大規模モデルは何でもお見通し理論．

[Tech*Research @TechResearch5]

# Countering Popularity Bias by Regularizing Score Differences
A regularization term to reduce popularity bias without losing accuracy too much.

#recsys2022 #fairness #privacy

[Tech*Research @TechResearch5]

# Bundle MCR: Towards Conversational Bundle Recommendation
Dialog base recommendation. 質問を繰り返すことで，BERTで商品の不明なattを穴埋め．

[Tech*Research @TechResearch5]

keynote
# Co-designing ML Models and Data Activists
femicideとかをデーテベース化して活動に活かす．
https://t.co/SkOGb9et1c

#recsys2022

[Tech*Research @TechResearch5]

# Multi-Modal Dialog State Tracking for Interactive Fashion Recommendation
In a dialog-based recommendation task, they proposed an RNN-based model matching image and its description(multi-modal).

[Tech*Research @TechResearch5]

# Modeling Two-Way Selection Preference for Person-Job Fit
A graph neural network for person-job matching considering two-way of candidates and employers.

#recsys2022

[Tech*Research @TechResearch5]

# Reusable Self-Attention Recommender Systems in Fashion Industry Applications
Instead of item embeddings, the model learns item-feature embeddings like color=red so that these embeddings are reusable. They used a 2-month history of user data because of GDPR limitations.