No abstract

The present study aims to deduce bikeability based on a collective understanding and provides a methodology to operationalize its calculation based on open data. The approach contains four steps building on each other and combines qualitative and quantitative methods. The first three steps include the definition and operationalization of the index. First, findings from the literature are condensed to determine relevant categories influencing bikeability. Second, an expert survey is conducted to estimate the importance of these categories to gain a common understanding of bikeability and merge the impacting factors. Third, the defined categories are calculated based on OpenStreetMap data and combined to a comprehensive spatial bikeability index in an automated workflow. The fourth step evaluates the proposed index using a multinomial logit mode choice model to derive the effects of bikeability on travel behavior. The expert process shows a stable interaction between the components defining bikeability, linking specific spatial characteristics of bikeability and associated components. Applied components are, in order of importance, biking facilities along main streets, street connectivity, the prevalence of neighborhood streets, green pathways and other cycle facilities, such as rental and repair facilities. The mode choice model shows a strong positive effect of a high bikeability along the route on choosing the bike as the preferred mode. This confirms that the bike friendliness on a route surrounding has a significant impact on the mode choice. Using universal open data and applying stable weighting in an automated workflow renders the approach of assessing urban bike-friendliness fully transferable and the results comparable. It, therefore, lays the foundation for various large-scale cross-sectional analyses.

Information extraction from documents is a ubiquitous first step in many business applications. During this step, the entries of various fields must first be read from the images of scanned documents before being further processed and inserted into the corresponding databases. While many different methods have been developed over the past years in order to automate the above extraction step, they all share the requirement of bounding-box or text segment annotations of their training documents. In this work we present DocReader, an end-to-end neuralnetwork-based information extraction solution which can be trained using solely the images and the target values that need to be read. The DocReader can thus leverage existing historical extraction data, completely eliminating the need for any additional annotations beyond what is naturally available in existing human-operated service centres. We demonstrate that the DocReader can reach and surpass other methods which require bounding-boxes for training, as well as provide a clear path for continual learning during its deployment in production.