Graph Databases (GDBs) have been successfully used to build high quality, up-to-date and low latency recommendation applications in the server. When it comes to Graph Databases (GDBs), one may ask whether their power could be harnessed and exploited in mobile or embedded environments in the edge, in order to provide faster yet high-quality, reliable and private recommendations to the end user. Sparksee is the graph database solution for embedded and mobile applications, thanks to its compact and fast management of data and the privacy preserving features provided to the user.

We provide a number of use cases (developed in depth in other posts), where a graph based recommender system can be used in the edge to provide private yet high quality recommendations. For digital Tourist guides, the recommendation of Points of Interest (POIs) in a city, where the interests of the user or group of users travelling together are taken into account  In the case of drivers, to recommend Infotainment system actions to reduce the number of interactions and reduce the number of accidents. Recommending routes, which requires complex multipoint routing algorithms, that will be used in digital Tourist guides or in rescheduling failed deliveries on the fly.

In all those cases, the privacy of data, the lack of connectivity, and the reduced amount of computational resources available,  push even further for a use of a GDB in the mobile or embedded environment.

The technologies provided by Sparksee are compact thanks to the use of bitmaps which provide high performance in the computations of graph algorithms in the mobile or embedded environment. Sparksee is out of core, providing a persistent data store at the same time. Also, Sparksee provides special features like encryption at the disk level, which allows to preserve the privacy of the data in the case of theft, loss or sale of the device, and recovery of the database in the case of database corruption. Having a GDB in the mobile device allows for preserving the service to the user in the case of bad connectivity or off-line operation, and better latency in the case of saturated networks. Finally, saving cloud operation at the cost of mobile calculations may save a significant amount of money to the provider of the service.

Automotive predictive systems

Safety is a must for car manufacturers, and most of their customers consider this as one of the most important factors when buying a vehicle. Providing such safety is not only a matter of robustness, but also a matter of avoiding interactions with the car. Distractions caused by such interactions have turned to be the largest cause of accidents in the USA.

Thus, creating predictive systems that foresee the interaction of the driver with the car is a need. Some types of predictive systems are based on multivariable Knowledge Bases (KB) that record all the actions done by the user. The evolution of the actions over time, allow those KB systems predicting and correcting the actions of the driver, increasing the safety of the vehicle and its occupants.

However, managing those recommendations from the cloud entails different issues that can turn into problems, like the latency of the network which may delay predictions and imply higher accident chances, or the sharing of data with third parties that may be used against the user, causing data privacy breaching.

Sparksee mobile provides the ideal technology to provide a KB that is installed and used in the car, with no need for external cloud support, providing the fast and reliable predictions that increase the safety for the vehicle occupants. Sparksee mobile provides the first high performance Graph Management System for embedded and mobile environments, including encryption at the disk level, and very compact data and software footprints.

Tourist Guide Application

Tourist guide applications and other applications where the recommendation of points of interest (POIs) is a central feature, require the gathering and processing of user’s private data (e.g. where a user has been, what she has visited, etc.) to empower the recommendation engine in the cloud. These recommendation engines typically use expensive cloud batch processes to compute models that are later used to recommend the POIs to the end users. The main drawback of this approach is that: 1) the models are never up-to-date given their computational cost (they are usually recomputed periodically) making them to lag against recent trends and 2) they require the use of massive amounts of privacy sensitive data the user might not want to disclose.

Thanks to Sparksee Graph Database, one can create and deploy a real-time and high quality graph-based recommender engine in the mobile device, keeping the privacy sensitive data under the control of the user. With such data, a personalized context graph storing the different user interactions with the app such as liked interests, viewed posts or personal interests can be built. Given a list of POIs, these can be ranked by analysing how these are structurally connected with the personalized context graph of the user and extracting features to be used in conjunction to other ML techniques such as support vector machines or neural networks.

With this approach, privacy sensitive data is always kept in the device and never sent to the cloud, and the recommender engine is personalized for each user individually. With Sparksee, such graphs can be stored and queried efficiently, providing fast responses with a small memory footprint.

Ego network use

Social media apps are used by billions of users to consume content created by friends and other institutional sources. In order to recommend the best content to any user, applications record user’s interactions such as post reads, likes, messages to other users or friends, etc. which are then fed into a recommender systems.

Such an approach, entails two major issues. First, privacy sensitive data is sent to a third party (the application provider), which becomes at risk of being exploited for malicious uses (e.g. the Cambridge Analytica case). Second, applications do not share data between them, data that could be potentially combined for improved recommendation tasks. For instance, people use to interact with the same people using different apps.

Thanks to Sparksee Graph Database, one can create and deploy a real-time and high quality graph-based recommender engine on the mobile device, that combines data from multiple sources/applications to create an ego-network around the user. Such an ego network can then be used to filter content or predict its relevance to the user based on her preferences, her past interactions with different social media apps, messages to users, etc. Since data is kept in the mobile device, it is controlled by the end user and not disclosed to any third-party.

With Sparksee, such graphs can be stored and queried efficiently, providing fast responses with a small memory footprint.

Send a message


Barcelona, Spain