Urban Data Observatory

The Urban Data Observatory is the main entry point to the OrganiCity platform assets.

UDO

It provides data exploration functionalities across three main interface modules: the navigation modules, involving the spatial and filter views, the text search and the assets visualization. The three modules together are designed to perform as an integrated ecosystem, supporting users on the search and understanding of useful and valuable data assets

The map interface provides geographical exploration of assets at multiple scales. In order to maintain a low entry barrier the interface is modelled on other existing map services the users might be familiar as Google Maps. This includes the following key features: - Manually navigating the map by standard zoom and pan actions. - Browsing the map by clicking at specific assets. - Incremental search featuring simple places search. - Client geolocation in order to center the map on to the user location.

UDO Navigation

The top search bar allows to search for assets name and metadata with auto-complete features. Results are returned based on the recommendation engine. The search offers a simplified list of the available resources that users can quickly access without any intermediate process.

On the upcoming relases this will be completed by the Advanced search. This allowd using information related to assets metadata, such as provider, typology or update time. The result is then shown in a permanent list format allowing users to browser the obtained results in detail.

UDO Navigation

Assets Visualization

The Assets Visualization is designed following a system of horizontal blocks. The anatomy of the assets views contains the following sections:

UDO Navigation

[A] Data Location:

This module shows the geographical location of a resource on the map as described on the previous section. OC Assets always include location information as a common metadata. This can come in the form of:

[B] and [C] Data Visualization:

This module is designed to support users on assets data exploration:

[D] Assets details and metadata:

This supplies a detailed insight on the asset metadata. It provides information about the resource such as the provider, the service or the asset type, and thus supports users in getting a clear understanding about the data they are seeing.

[E] Provider details:

This module provides in-depth information about the provider of the asset. It is designed to give advance users as experimenters a clearer understanding of what asset they are browsing..

[F] Recommendations:

The recommendation section suggests four similar assets. Similarity is understood as: asset A and B are concidered similar if most users who previously viewed asset A also viewed asset B. A machine learning algorithm is handling this in the background, and the idea is that user will get suggestions on relevant alternative or supplementary assets to take into account when traversing available data sources in the OrganiCity facility.

Under the hood

The asset recommender is implemented leveraging Prediction.io (http://predictionio.incubator.apache.org), and we have used this recommender template: http://predictionio.incubator.apache.org/templates/similarproduct/quickstart/.

When an end user clicks on a specific asset in the UDO, a "view event" is send to the Predition.io core machine learning platform. The event sends information on which user clickes on the asset (only the userid), what asset was clicked (only asset id), and a timestamp for when the event happened. This behavior is shown with the "Event Data" arrow in the diagram below.

Prediction.io diagram

As a parallel action we are asking the recommender engine for recommendations that relate to the clicked asset. This is illustrated with the two arrows "Query via REST" and "Predicted Result" in the above diagram. Since all communication is happening between the browser (client) and a remote server, these actions can be delayed a few seconds. A consequense is that the end user might not see recommendations until a few seconds after they clicked a specific asset in the UDO. This action is happening asynchroniously, so there will be no sensation of the website being slow.

Over time we will get a huge amount of usage data, which will only make the recommendations better. This is due to the fact that machine learning algorithms need to be trained regularly in order to interpret input and produce a relevant output. In the case of the recommender engine, we train it once every night (around midnight) in order to keep recommendations as update as possible without slowing the recommendation engine down, thereby keeping a good end user experiencing.

[G] Comments:

This module allows users to comment on an asset.