Deep Learning For Implicit Feedback-based Recommender Systems

• Thesis title in Czech: Deep learning pro doporučování založené na implicitní zpětné vazbě
• Thesis title in English: Deep Learning For Implicit Feedback-based Recommender Systems
• Key words: deep learning, doporučovací systémy, implicitní zpětná vazba, recurrent neural networks
• English key words: deep learning, recommender systems, implicit feedback, recurrent neural networks
• Academic year of topic announcement: 2019/2020
• Thesis type: diploma thesis
• Thesis language: angličtina
• Department: Department of Software Engineering (32-KSI)
• Supervisor: Mgr. Ladislav Peška, Ph.D.
• Author: hidden - assigned and confirmed by the Study Dept.
• Date of registration: 10.10.2019
• Date of assignment: 11.10.2019
• Confirmed by Study dept. on: 19.11.2019
• Date and time of defence: 16.09.2020 09:00
• Date of electronic submission: 30.07.2020
• Date of submission of printed version: 26.07.2020
• Date of proceeded defence: 16.09.2020
• Opponents: RNDr. Štěpán Balcar

Guidelines

The thesis topic lies on an intersection of two important and intensively studied areas: deep learning and recommender systems. First, student should get sufficient knowledge in both areas. In deep learning domain, the main focus should be on recurrent neural networks (e.g., GRU, LSTM), Siamese network variants and embeddings generation networks (word2vec, doc2vec etc.). In recommender systems, the main focus should be on content-based recommendation and variants implicit user feedback.

Within these general concepts, the main aim of the thesis is to propose/adapt and evaluate suitable algorithms (expectably some RNN or Siamese networks variants) that can incorporate multimodal implicit user feedback (with a specific attention towards causality issues, implicit negative feedback and/or utilizing per-user object's visibility) and optionally also content-based information in the recommendation process. It is expected that proposals will be evaluated experimentally against some existing recommending approaches.

References

Ricci, F. et al (Eds): Recommender Systems Handbook, Springer, 2011
Jannach, D. et al (Eds): Recommender Systems: An Introduction, Cambridge University Press, 2011
Alexandros Karatzoglou, Balázs Hidasi: Deep Learning for Recommender Systems (Tutorial). RecSys 2017: 396-397
Massimo Quadrana, Alexandros Karatzoglou, Balázs Hidasi, Paolo Cremonesi: Personalizing Session-based Recommendations with Hierarchical Recurrent Neural Networks. RecSys 2017: 130-137
Dietmar Jannach, Malte Ludewig: When Recurrent Neural Networks meet the Neighborhood for Session-Based Recommendation. RecSys 2017: 306-310
Elena Smirnova, Flavian Vasile: Contextual Sequence Modeling for Recommendation with Recurrent Neural Networks. DLRS@RecSys 2017: 2-9
Hanjun Dai, Yichen Wang, Rakshit Trivedi, Le Song: Recurrent Coevolutionary Feature Embedding Processes for Recommendation. arXiv.org abs/1609.03675 (2016)
Peška L., Vojtáš P.: Using Implicit Preference Relations to Improve Recommender Systems, in Journal On Data Semantics, ISSN: 1861-2032
Ladislav Peska:Using the Context of User Feedback in Recommender Systems. MEMICS 2016: 1-12
Marius Kaminskas, Derek Bridge, Franclin Foping and Donogh Roche: Product-Seeded and Basket-Seeded Recommendations for Small-Scale Retailers, Journal on Data Semantics, pp.1-12, 2016.

Preliminary scope of work

Doporučovací systémy (doporučování objektů uživateli na základě jeho předchozích akcí) získaly v poslední dekádě velkou pozornost jak ve vědecké komunitě, tak i v komerčním prostředí. V současné době probíhá na poli doporučovacích systémů a algoritmů “revoluce” v podobě zapojení různých deep learning algoritmů na tvorbě doporučení. Jedním ze slibných směrů vývoje je využití recurrent neural networks (RNN) pro tvorbu session-based, případně session-aware doporučení (doporučení na základě dat z právě probíhající session uživatele).

Cílem práce je prozkoumat možnosti rozšíření RNN, případně podobných architektur pro využití různorodé implicitní zpětné vazby od uživatele, případně informací o jednotlivých objektech při tvorbě doporučení. Práce je svým charakterem především vědecká, částečně implementační. Dosažené výsledky by měly být v případě zájmu publikovatelné na mezinárodních konferencích nebo ve vědeckých časopisech.

Preliminary scope of work in English

Recommender systems gained serious attention recently both in research and industry. Currently, we face an ongoing revolution in the domain of recommender systems by applying various deep learning techniques into the process of recommendation. One promissing research direction is to apply recurrent neural networks (RNN) on the task of session-based or session-aware recommendation (the recommended objects are derived based on the feedback from current user's session).

The aim of the thesis is to focus on RNN or its derivates/extensions and use them for session-based/session-aware recommendations enhanced by the user's implicit feedback and/or content-based description of objects. The thesis is mostly research-based and partially implementational - the results (if interesting) could be published in scientific conferences/journals.