Some research points are also worth reading, such as cold start, deviation and correction, sequence recommendation, interpretability, privacy protection and so on. These studies are very interesting and enlightening, which will help broaden your research ideas.
The following classification is mainly based on my own judgment when reading the topic or abstract, according to the research direction of recommendation system, recommended technology and reproducible articles with special experimental nature. There may be omissions and erroneous attribution. Please correct me a lot.
Information cocoon room/echo room)/filter bubble, these three concepts are similar, and there are different opinions at home and abroad. Generally, it refers to the use of social media and information apps with algorithmic recommendation function, which may lead us to only see what we are interested in and agree with, and then let everyone live in their own small world, which is difficult to agree and communicate. The concept of this part can be found in Zhihu's article: /p/7 184428 1. There are four articles discussing such problems.
There are also many discussions about exploration and utilization in this conference, such as dobby slot machine and Google's new work, namely client exploration.
It involves the correction of sorting learning and the excavation of user deviation.
The depolarization of implicit feedback can be interpreted as pairwise sorting.
Khalil Damak, Sami Genis and Orfa Nasraoui
Reducing confusion and deviation in recommendation through information bottleneck
Liu,,,,, He Xiuqiang, Pan Weike and
User bias in ultra-precision measurement of recommendation algorithm
Ningxia wanghe
Using graphic learning and representation learning to do cold start.
Cold start is similar to the ranking of artists and gravity-inspired automatic encoders.
Guillaume Sarha-Galban, Roman Hennekun, Benjamin Chapus, Chen Yuean and Michalis Wazirjanis.
Shared neural project representation of complete cold start problem
Ramin Raziperchikolaei, Guannan Liang and Young-joo Chung
It involves the design of unified indicators such as offline or online evaluation methods, accuracy and diversity.
Evaluation of non-policy evaluation: sensitivity and robustness
Saito Fujimori Hideki, Lany Toma, Haruka Kiyohara, Shigeki Tomokawa, Narita Tomsuke and Kanno Kei.
Fast multi-step evaluation of recommendation system based on VAE
Diego Antognini and Boyd faltings.
Suggest an online evaluation method of causal effect.
Masahiro Sato
Unified measurement of accuracy and diversity for recommendation system
Javier Palapar and Philip Radlinski
Short sequence recommendation involving conversation dimension; Using transformers commonly used in NLP to discuss and solve the gap of sequence recommendation, I am still very interested in this work, and I will read it carefully in the future!
Privacy protection is combined with federal research.
Extracting Black Box Attacks on Sequential Recommenders by Data-free Model
Yue Zhenrui, He Zhankui, Ceng Huimin and Julian Macaulay
Large-scale interactive session recommendation system
Ali Montazelegm, james allen and philip thomas.
Example 3: Explanatory Attribute-aware Item Set Suggestions
Xian Yikun, Li Jing, Jim Chan, Andrey Kan, Xin Luna Dong, Christos Fa Roussos, George Karypis, S. Muthukrishnan and Zhang Yongfeng.
Source alignment variational model for cross-domain recommendation
Aghiles Salah, Thanh Binh Tran and Hady Lauw
Use visual information to make suggestions.
Ambareesh Revanur, Vijay Kumar and Deepthi Sharma
Chen Huiyuan, Lin Yu III, Faye Wong and hao yang.
The interactive design of multi-user intention recommendation system in gourmet scene is discussed.
"Serve every user": Support different dietary goals through multi-list recommendation interface.
Alan Stark, Edith Asotich and Christopher Trattner.
Iteration involving traditional collaborative filtering and metric learning; Explore emerging graphics learning technology, federated learning technology and reinforcement learning technology.
The matrix decomposition for collaborative filtering is only to solve an adjoint potential Dirichlet distribution model.
Florian William
Improve the negative interaction of collaborative filtering: don't go deep, go deep.
Harald Staercke and Liang Dawen
Protocol F: Prototype collaborative filtering for a small number of project recommendations
Aravind Sankar, Junting Wang, Adit Krishnan and Hari Sundaram
The application of knowledge map and the integration of graphic embedding technology and situational awareness representation technology are of great interest to both working individuals.
Antonio Ferrara, Vito Walter Anelli, tommaso Di Noah and Alberto Carlo Maria mancino.
Marco polignano, Cataldo Musto, Marco de Demis, pasquale Lopez and Giovanni Semeraro.
It involves training, optimization, retrieval, real-time streaming and so on.
Jeremy lapaz, Julian Macaulay and Carl Arberry.
Reproducible papers can reproduce experimental articles, ***3 articles. The sampling evaluation strategies in sequence recommendation are discussed respectively. Comparison of generation and search methods in dialogue recommendation system: comparison between neural network recommendation system and matrix decomposition recommendation system.
By sorting and classifying the papers, the author also found some interesting research points, such as: recommending articles about the echo chamber effect of the system; The gap and solution of Transformers in sequence recommendation and NLP sequence characterization article: Transformers4Rec；; Fusion of graph embedding representation and context-aware representation: a comparative article on NCF and IF experiments;