Skip to Main content Skip to Navigation
New interface
Theses

Towards diversified recommendations

Abstract : Recommender systems (RS) have been widely applied in real life scenarios to constantly provide personalised recommendation to satisfy users’ need. In classical top-N recommendation models, historical user-item interactions are collected and exploited to learn and predict for each user a top-N item list. But we are also aware that both user-side and item-side auxiliary information may help to improve the recommendation performance. At the same time, in terms of recommendation performance, accuracy has been the main research focus in recommender systems, though works have pointed out that an optimal accuracy is not equal to an optimal satisfaction of users towards recommendation. An accuracy-centric recommendation model may create an isolate, singular and redundant atmosphere when providing the service, thus it is essential to bring other goals in RS to alleviate these problems. In this dissertation, we focus on bringing diversity along with accuracy as recommendation goals, as we argue that a diversified recommendation helps alleviate the problems suffered by accuracy-centric RS. We also take item-side auxiliary information into account for enhancing accuracy. Thus we propose diversity-aware top-N recommendations based on knowledge graph embedding to aim at achieving both high accuracy and high diversity in recommendation lists for users. Our first contribution is DivKG, a diversified recommendation framework that combines knowledge graph embedding and determinantal point processes (DPP). We propose a new personalised DPP kernel matrix construction method that uses knowledge graph embedding results for DPP diversification. Our second contribution is EMDKG, a diversified recommendation framework which encodes semantic diversity into item representations and achieve better trade-off compared to state-of-the art methods in terms of accuracy and diversity.
Document type :
Theses
Complete list of metadata

https://tel.archives-ouvertes.fr/tel-03814902
Contributor : ABES STAR :  Contact
Submitted on : Friday, October 14, 2022 - 11:39:29 AM
Last modification on : Saturday, October 15, 2022 - 4:33:20 AM

File

these.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-03814902, version 1

Citation

Lu Gan. Towards diversified recommendations. Information Retrieval [cs.IR]. Université de Lyon, 2022. English. ⟨NNT : 2022LYSEI047⟩. ⟨tel-03814902⟩

Share

Metrics

Record views

0

Files downloads

0