Resumen de Privacy-Preserving Crowdsourcing-Based Recommender Systems for E-Commerce & Health Services
Recommender systems have become a fundamental mechanism to provide users with useful selected information, which could be effective to optimise a large amount of decisions, for instance, in the e-commerce field. In this context, the Internet provides a wealth of information on a huge variety of products and services that may be useful to potential buyers. However, this wealth of information may become a problem rather than a solution because it can hinder the decision making. Collaborative Filtering (CF) is a recommender system that comprises a large family of methods. The aim of CF is to make suggestions on a set of items {I} (e.g. books, music, films or routes), based on the preferences of a set of users {U} that have already acquired and/or rated some of those items. Therefore, information about the users’ behaviours and preferences is required in order to provide profitable recommendations. In this sense, one of the most relevant problems faced by businesses is non-response. Moreover, the widespread use of CF on the Internet provides great opportunities and benefits to both companies and users, but there is a major drawback: the lack of users’ privacy. The importance of privacy in CF systems is emphasised by the growing pace at which information of each user is collected and stored. Careless management of personal information, besides being illegal, could lead to serious consequences for both users, whose information is stored, as well as businesses.
This dissertation contributes to the design of algorithms and systems that face the issues explained above. First, we provide an extensive background in CF, Statistical Disclosure Control and privacy-preserving CF methods. Second, we propose a set of imputation methods in order to deal with the lack of information in CF based recommender systems. Next, we present two privacy-preserving CF methods which protect the privacy of the users and achieve a remarkable recommendation’s accuracy.
In addition to the aforementioned contributions, we also focus on the urbanisation process that is taking place worldwide. Nowadays, smart cities are gaining importance and their infrastructure can be used to improve the healthcare services provided to citizens. This is the philosophy of the so-called Smart Health concept. In this context, wireless communication systems play a key role, as enablers of real time and location independent connectivity, increasing system functionality and decreasing operational costs. In this scenario, multiple wireless systems co-exist, which requires an in-depth radioplanning in terms of coverage/capacity ratios, with particular consideration of the impact of interferences. However, radiofrequency analysis, in terms of useful received signal levels for a given set of connections as well as for potential interfering connections, can be a challenging task due to: the large size of scenarios; the existence of multiple frequency dependent materials; and inherent variability of mobile connections, given by the movement of potential scatterers.
We have studied that information filtering techniques such as CF may augment the capabilities of smart cities and smart health. In order to support our vision, we propose the use of recommender systems integrated with the sensing infrastructure of smart cities to provide citizens with routes recommendations that take into account their health conditions and preferences. Moreover, a novel approximation based on the combination of an in-house developed 3D Ray Launching code and a CF technique is used to analyse the performance of wireless channels emulating context-aware scenarios. The aim of the proposed method is to provide optimal deployment strategies for massive wireless system and wireless sensor networks. In this sense, the results show a remarkable improvement in both accuracy of radiofrequency measurements and computational cost, compared with other methods of the state-of-the-art.
A summary of the main contributions and future work is presented in the next paragraphs:
In Chapter 3, we propose classical and new imputation methods to deal with incomplete data in CF datasets, which have specific characteristics such as a high dimensionality and a high percentage of null values. Moreover, we performed extensive experiments with three well-known datasets, showing that our proposed methods are able to obtain better quality recommendations and behavioural precision than well-known state-of-the-art methods. Future research in this topic will focus on two ways: (i) combine this imputation methods with privacy-preserving CF methods and (ii), use context-aware information to deal with CF’s inherent problems such as sparseness, cold user/item and scalability.
Collaborative Filtering is a recommender system used to perform automatic recommendations to users in multiple contexts. Despite its great advantages, we highlighted its downside regarding users’ privacy. In Chapter 4, we propose three PPCF methods in order to protect the privacy of the users involved in CF processes. The experimental results show that V-MDAV obtains better results and provides both more privacy and data usability than well-known methods such as MDAV and Gaussian noise addition. Future work in this topic will focus on two ways. First, to overcome the dimensionality issues and outlier/malicious users in order to obtain less biased results. Second, to improve the efficiency of our methods in order to enable implementation in a decentralised setting.
In Section 5.1, we proposed the idea of using recommender systems integrated with the sensing infrastructure of smart cities to promote citizen’s healthy habits in real time. In this field, future work will focus on open research topics related to s-health such as the privacy protection of citizens that use our proposal.
In Section 5.2, we proposed and tested a set of hybrid methods combining RL and CF techniques in order to estimate context-aware radio-planning tasks with high accuracy in a time-efficient way. Further work in radiopropagation analysis will focus on three directions: First, to explore the use of three-dimensional structures instead of two-dimensional sub-matrices. Second, to completely transform LD simulations into HD simulations in a time-efficient way. Third, to study the performance of this technology in large complex environments, such as smart cities.
