Linked Data Visualization. Laura Po

      Figure 2.1: Process for the visualization of data.

      The information is usually present in various formats depending on the source; the appropriate data extraction and analysis technique must be selected to transform the raw information into a more semantically rich, structured format. A set of data processing techniques are then applied to enhance the quality of the collected data; these include cleaning data inconsistencies, filling in missing values and detecting and eliminating duplicates. After that, the data is enriched and customized with visual characteristics, meaningful aggregations, and summaries which facilitate user-friendly data exploration; finally, proper indexing is performed to enable efficient searching and retrieval. The details for each step are presented below.

      Data Retrieval, Extraction. The first step concerns the retrieval and extraction of the data to be visualized. Raw data exists in various formats: e.g., books and reports describe phenomena in plain text—unstructured information, websites, and social networks contain annotated text and semi-structured data, whereas open data sources and corporate databases provide structured information. Data must first be retrieved in a digital format that is appropriate for processing (e.g., digital text files from newspapers). The core modeling concepts and observations are then extracted. Especially when the source data is in plain text, this is usually performed in an iterative human-curated way that refines the quality of the extracted data. For structured data sources, the process involves the extraction of the source concepts and their mapping to the target modeling.

      Linked Data usually comes in structured formats, i.e., conforms to well-defined ontologies and schemas. Linked data are usually the result of a data extraction and transformation task, which has turned raw data into semantically rich data representations. Therefore, LD visualization usually starts from the next step, that of preparing already structured data for visualizing them.

      Data Preparation. Input data are provided either in databases or data files (e.g., .csv, .xml). This step involves identifying all the concepts within the input datasets and representing them in a uniform data model that supports their proper visualization and visual exploration. For example, the multidimensional model is largely employed in social and statistical domains and represents concepts as observations and dimensions. Observations are measures of phenomena (e.g., indices) and dimensions are properties of these observations (e.g., reference period, reference area). Thus, a first step is to analyze the datasets and identify the different types of attributes they contain (date, geolocation, numeric, coded lists, literal). Each attribute is mapped to the corresponding concept of the multidimensional model, such as dimension, observation, coded list, etc. In addition, data processing requires a set of quality improvement activities that eliminate data inconsistencies and violations in source data. For example, missing or inconsistent codes are filled in for coded list attributes, date and time attributes are transformed to the appropriate format, and numerical values are validated so that wrong values can be corrected. Moreover, input data from multiple sources usually contain duplicate facts and a deduplication of the dataset must be performed. Deduplication is the process of identifying duplicate concepts within the input dataset based on a set of distinctive characteristics. A final task concerns the enrichment or the interlinking of the data with information from external sources. For example, places and locations are usually extracted as text; they can subsequently be annotated and enriched with spatial information (e.g., coordinates, boundaries) from external web services or interlinked with geospatial linked data sources (e.g., geonames.org) for their proper representation on maps.

      Visual Preparation. This set of tasks involves the enrichment and customization of the data with characteristics that enable the proper visualization of the underlying information. These characteristics extend the underlying data model with visual information. For example, colors can be assigned to coded values and different types of diagrams can be bound to different types of data, timelines to date attributes and maps to geographical values. Thus, customization and the building of the visual model are necessary tasks before data visualization. In addition, the production of visual summaries and highlights is a common task, especially in the visualization of very large datasets. Summaries provide the user with overviews of the visualized data, and visual