Les systèmes de recommandation à base de confiance

Abstract : Recommender systems (RS) exploit users' behaviour to recommend to them items they would appreciate. Users Behavioral divergence on the web results in a problem of performance fluctuations to (RS). This problem is observed in the approach of collaborative filtering (CF), which exploites the ratings attributed by users to items, and in the trust-based approach (TRS), which exploites the trust relations between the users. We propose a hybrid approach that increases the number of users receiving recommendation, without significant loss of accuracy. Thereafter, we identify several behavioral characteristics that define a user profile. Then we classify users according to their common behavior, and observe the performance of the approaches by class. Thereafter, we focus on the TRS. The concept of trust has been discussed in several disciplines. There is no real consensus on its definition. However, all agree on its positive effect. Subjective logic (LS) provides a flexible platform for modeling trust. We use it to propose and compare three trust models, which aims to predict whether a user source can trust a target user. Trust may be based on the personal experience of the source (local model), or on a system of mouth (collective model), or the reputation of the target (global model). We compare these three models in terms of accuracy, complexity, and robustness against malicious attacks
Document type :
Theses
Complete list of metadatas

Cited literature [175 references]  Display  Hide  Download

https://hal.univ-lorraine.fr/tel-01751172
Contributor : Thèses Ul <>
Submitted on : Thursday, March 29, 2018 - 1:09:08 PM
Last modification on : Tuesday, April 24, 2018 - 1:30:10 PM

File

DDOC_T_2014_0203_ALCHIEKH_HAYD...
Files produced by the author(s)

Identifiers

  • HAL Id : tel-01751172, version 1

Collections

Citation

Charif Alchiekh Haydar. Les systèmes de recommandation à base de confiance. Autre [cs.OH]. Université de Lorraine, 2014. Français. ⟨NNT : 2014LORR0203⟩. ⟨tel-01751172⟩

Share

Metrics

Record views

174

Files downloads

237