Section: New Results
Participants : Quang Vinh Dang, Claudia-Lavinia Ignat, Francois Charoy, Olivier Perrin, Mohammed Riyadh Abdmeziem, Hoang Long Nguyen.
Trust between users is an important factor for the success of a collaboration. Users might want to collaborate only with those users they trust. We are interested in assessing users trust according to their behaviour during collaboration in a large scale environment. In order to compute the trust score of users according to their contributions during a collaborative editing task, we need to evaluate the quality of the content of a document that has been written collaboratively. We investigated how to automatically assess the quality of Wikipedia articles in order to provide guidance for both authors and readers of Wikipedia. Most existing approaches for quality classification of Wikipedia articles rely on traditional machine learning with manual feature engineering, which requires a lot of expertise and effort and is language dependent. We proposed an approach that addresses the trade-off between accuracy, time complexity and language independence for the prediction models . Our approach relying on Recurrent Neural Networks (RNN) eliminates disadvantages of feature engineering, i.e. it learns directly from raw data without human intervention and is language-neutral. Experimental results on English, French and Russian Wikipedia datasets show that our approach outperforms state-of-the-art solutions.
Rating prediction is a key task of e-commerce recommendation mechanisms. Recent studies in social recommendation enhance the performance of rating predictors by taking advantage of user relationships. However, these prediction approaches mostly rely on user personal information which is a privacy threat. We proposed dTrust , a simple social recommendation approach that avoids using user personal information. It relies uniquely on the topology of an anonymized trust-user-item network that combines user trust relations with user rating scores for items. This topology is fed into a deep feed-forward neural network. Experiments on real-world data sets showed that dTrust outperforms state-of-the-art in terms of Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) scores for both warm-start and cold-start problems.
One dimension of our work is dedicated to ensure consistency of the key server. We design Trusternity, which is a secure, scalable auditing mechanism using a blockchain to replace the gossiping mechanism of transparent log system. We have implemented Trusternity as a proof-of-concept, and we have led some evaluation about the detection of malicious behavior on the blockchain network.
Securing P2P collaborative system remains a critical issue for its widespread adoption. One of our goals is to ensure that communication between collaborating partner is secure from end to end. We need to encrypt exchange of operations among partners. For that we propose to rely on group keys management. One of the issue is that the composition of the partnership can change and this require to change the group key. Since we don't want a central server to manage keys, that would break the p2p nature of our approach we need to propose group key management protocols that are resilient to change in groups, even in group of large size.