FIGURE 7 - uploaded by Felix Beierle
Content may be subject to copyright.
Source publication
Recommender systems recommend new movies, music, restaurants, etc. Typically, service providers organize such systems in a centralized way, holding all the data. Biases in the recommender systems are not transparent to the user and lock-in effects might make it inconvenient for the user to switch providers. In this paper, we present the concept, de...
Context in source publication
Context 1
... cspurl. This is the URL pointing to the publicly available dataset available at a Cloud Storage Provider (CSP). Fig. 7 shows the initialization of the device-to-device communication utilizing the workaround with a CSP. Alice authorizes access to her account with the CSP. In the MobRec MVP, we use Google Drive. The platform-independent JavaScript code then handles the authorization for Google Drive via OAuth 2.0 and uploads Alice's dataset dataset_a. ...
Citations
... System/Method Object Application [21] Logistic regression-based system In-game item selection Online game [44] The intelligent exergame-based rehabilitation system Patient rehabilitation Serious games [45] Recurrent neural networks Game item MOBA game [46] Feedforward neural network Game item and product Video games [47] Neural network In-game item selection Video games [48] ELECTRE Traveler itinerary Web-based [49] K-means + GA Tourism destinations Mobile [50] Opinion-mining technology Tourism destinations Web-based [51] Topic modeling + emotional analysis + haversine Tesser-known tourism place Not mentioned [52] MCRS using a fuzzy approach Tourism service Web-based [53] Destinations ratings-based MCRS Halal tourism destinations Desktop game [54] Weighted sum model-based MCRS Tourism destinations Web-based [55] Decentralized collaborative filtering Movies Not mentioned [56] Decentralized hybrid systems Advertising Mobile [57] MobRec Movies, music, restaurants, etc Mobile Y.M. Arif et al. ...
... The MobRec architecture consists of data collection through a data-sharing system between devices and a recommender system that works locally on each device, where recommendations are generated based on user interests and preferences. [57]. Several previous studies have succeeded in developing the decentralized concept in a recommender system, but a secure data-sharing system should certainly support the concept. ...
Tourism destinations serious game (TDSG) requires the ability to respond to players through recommendations for selecting appropriate tourist destinations for them as potential tourists. This research utilizes ambient intelligence technology to regulate the response visualized through a choice of serious game scenarios. This research uses the Multi-Criteria Recommender System (MCRS) to produce recommendations for selecting tourist destinations as a reference for selecting scenario visualizations. Recommender systems require a decentralized, distributed, and secure data-sharing concept to distribute data and assignments between nodes. We propose using the Ethereum blockchain platform to handle data circulation between parts of the system and implement decentralized technology. We also use the known and unknown rating (KUR) approach to improve the system's ability to generate recommendations for players who can provide rating values or those who cannot. This study uses the tourism theme of Batu City, Indonesia, so we use personal characteristics (PC) and rating of destinations attribute (RDA) data for tourists in that city. The test results show that the blockchain can handle decentralized data-sharing well to ensure PC and RDA data circulation between nodes. MCRS has produced recommendations for players based on the KUR approach, indicating that the known rating has better accuracy than the unknown rating. Furthermore, the player can choose and run the tour visualization through game scenarios that appear based on the recommendation ranking results.
... There have been demonstrations with dozens and even hundreds of clients, but the complexity of a federated learning system comprising millions of mobile devices is simply inconceivable. (2) Constrained resources on client devices: Most existing FedRecs [1,4] require the client devices to maintain and store the whole item embedding table, which is hard to work in practice due to the constrained on-device resources. (3) Cumbersome pipeline: To achieve privacy protection or data augmentation, some FedRecs require an additional third-party server to implement encryption services [21] or neighbor discovery [45], resulting in a cumbersome pipeline and also increasing the risk of privacy leakage. ...
... However, one major difference from our approach is that FedPerGNN requires a trusted third party to find users who have co-interacted items, which results in a cumbersome pipeline, and it is difficult to find such a trusted third-party server in practice. There are also some works allowing devices to collaboratively learn with its neighbor devices in a fully device-to-device (D2D) fashion, namely decentralized recommender systems [2,3,5,32]. However, such methods only rely on D2D collaborations and generally require that each device stores and maintains huge chunks of non-privacy item data, which is unnecessary and hard to work in practice due to the constrained resources. ...
Collaborative filtering (CF) based recommender systems are typically trained based on personal interaction data (e.g., clicks and purchases) that could be naturally represented as ego graphs. However, most existing recommendation methods collect these ego graphs from all users to compose a global graph to obtain high-order collaborative information between users and items, and these centralized CF recommendation methods inevitably lead to a high risk of user privacy leakage. Although recently proposed federated recommendation systems can mitigate the privacy problem, they either restrict the on-device local training to an isolated ego graph or rely on an additional third-party server to access other ego graphs resulting in a cumbersome pipeline, which is hard to work in practice. In addition, existing federated recommendation systems require resource-limited devices to maintain the entire embedding tables resulting in high communication costs. In light of this, we propose a semi-decentralized federated ego graph learning framework for on-device recommendations, named SemiDFEGL, which introduces new device-to-device collaborations to improve scalability and reduce communication costs and innovatively utilizes predicted interacted item nodes to connect isolated ego graphs to augment local subgraphs such that the high-order user-item collaborative information could be used in a privacy-preserving manner. Furthermore, the proposed framework is model-agnostic, meaning that it could be seamlessly integrated with existing graph neural network-based recommendation methods and privacy protection techniques. To validate the effectiveness of the proposed SemiDFEGL, extensive experiments are conducted on three public datasets, and the results demonstrate the superiority of the proposed SemiDFEGL compared to other federated recommendation methods.
In this chapter, we develop a mobile platform for decentralized recommender systems—MobRec. The core concept is that everything runs on smartphones. Due to decentralization, MobRec does not exhibit the lock-in effects present in centralized service providers for recommender systems. In MobRec, ratings and preferences are captured locally. During daily life, these ratings and preferences are exchanged with users in proximity in a device-to-device manner. Locally running recommender systems or third-party service providers can then recommend items based on own data and data received from users met before. We implement the platform for off-the-shelf smartphones for both Android and iOS. We implement device-to-device data exchange that can run in the background on both platforms. In our evaluation, we show the feasibility of our approach. Our implementation can serve as a blueprint for future research and software development in this field.
