The QROWD Analytics Optimisation Toolbox (AOT) comprises five tools aimed at improving the performance -efficiency and accuracy- of the analytics required to deliver the QROWD mobility services:
- Crowd feedback-aware link prediction: Link prediction engine for city staff, capable of iterative improvement of generated datasets using data quality assessment from the crowd.
- Analytics with crowd feedback: Algorithms Optimiser Tool: A great share of machine learning algorithms performing analytics to gain insights into bigger amounts of data require training data. Having great amounts of high quality training data is key for performing meaningful analytics. To bootstrap such analytics algorithms, or to adapt to changing conditions one can make use of the knowledge of the crowd and let users generate the training data, or correct predictions made by the analytics algorithms in cases where the achieved accuracy is insufficient.
- Hybrid Analytics Workflow: In terms of machine learning and analytics there is a plethora of different approaches to gain insights or make predictions based on streams of numeric data. Even though these techniques in general are robust and perform very well they usually cannot make use of background knowledge, e.g. expressed be means of Semantic Web technologies. To improve the outcome of the analytics tasks we combine these numeric and knowledge-based approaches to form a hybrid analytics workflow.
- Spatio-temporal Analytics: Apart from background knowledge explicitly expressed, e.g. by means of Semantic Web technologies, especially in terms of spatial and temporal data, there is also a vast amount of knowledge usually not stored explicitly. This concerns spatial relations like ‘near’, ‘inside’, or temporal relations like ‘before’ and ‘after’. However such relations are useful to express human-understandable patterns in data. To make use of such implicit information we exploit special purpose spatial and temporal data stores to support our analytics tasks.
- Crowd feedback allows starting and continuously improving analytics tasks without the need for vast amounts of initial training data
- Combination of stable and robust machine learning approaches working on streams of numeric data and (background) knowledge-based analytics method
Areas of Application:
- Smart Cities
- Urban mobility
- Transport logistics
Market Trends and Opportunities:
- Leveraging existing technology stacks
- Pioneering algorithms show weaknesses which hamper their accuracy and render low levels of trust
- Improves/makes applicable algorithms that otherwise might perform poorly
- Enables integrated analytics based on static, semantically rich background data and data streams.
- Spatial and temporal analytics on semantic, descriptive data which is capable of providing human understandable classifiers describing certain events/phenomena
- Spatio-temporal analytics
- Knowledge graph embedding-based link prediction
INSTITUT FUR ANGEWANDTE INFORMATIK (INFAI)
BDV Reference categories:
Data processing architectures
Data Visualisation and Interaction