• Nenhum resultado encontrado

views using keyword searches and text similarity metrics. Shrinivasan et al. [70] recom- mend related annotations based on the context of the current analysis. Finally, Chen et al. [80] use annotation context as a way to semi-automatically annotate facts belonging to the same analysis categories across visualizations.

Most existing VA work supports context aware annotations on a single visualization view.

Traditionally BI analysts conduct their analysis using complex visualization dashboards.

Dashboards visualize multi-dimensional datasets through a combination of linked charts [5], and are very widespread in BI (e.g. [42][43][44][35][36]). Annotations are supported in most dashboards, and are usually attached to data points (not the chart graphic).

Nevertheless, these systems do not adequately support several needs identified by BI ex- perts (detailed later in CHAPTER 4), such as annotations spanning multiple charts, or annotation transparency across data dimensions and visualizations (annotate once, see everywhere, as is done in text [81]). These advanced functionalities, such as multi-chart annotation, can be hard to achieve, as annotating multiple points is an involved process:

since only one visualization is shown at a time, an annotation can only be added to one query. To add more queries to an existing annotation, the user needs to continuously switch visualizations.

This body of research uses context aware annotations for different purposes. Our work extends it in two ways: based on expert users’ comments we identify desired behavior and possible uses of context aware annotations on visualization dashboards; and we explore challenges in their design and use, for example due to the dynamic nature of ”context”, discussed in CHAPTER 4 .

Research conducted to date has demonstrated the value of storytelling to improve organi- zational structure [86] and collaboration quality [87, 88], socialization and adaptation of new employees [89–91], organizational and financial success [92,93], innovation and new product development [94], and teaching and learning [95].

The majority of this work is a meta-analysis of the effect of storytelling within an organi- zation, rather than the development of tools to enhance the storytelling process as is our case. Moreover, the stories themselves discussed in this work, relate to the transmission of information and knowledgewithin an organization, mostly intextual or verbal form, rather than in visual form. The widespread use of visualization dashboards in the domain is a more recent development [5], and so is the transmission of knowledge between organiza- tions (dedicated BI analysis organizations and their clients or end-users). Thus storytelling needs in the domain have evolved and in our work we take a step at characterizing them more precisely.

2.5.2 Stories in sense making

Baber et al.[83] point out that contemporary theories of sense making rely on the idea of ’schema’, of a structure to organize and represent factual information, as well as the knowledge, beliefs and expectations of the people who are engaged in sense making. They can thus be considered as a collection of narratives. They further discuss the formalism of stories in sense making, and how the most effective stories are organized around the actors in the stories, their actions and rationale, events, their context, and most impor- tantly the relationships between these. As further argued by Bier et al.[96], for effective collaboration and communication we need to use less text, and organize knowledge around entities (people, places, things, times etc.) rather than free form text. Similarly, Pirolli

& Russell [97] propose the mapping of intelligence facts and insights into frames, that can be expressed in a variety of forms including stories, maps, organizational diagrams or scripts. It is thus clear that conducting intelligence analysis, communicating findings, and organizing knowledge in stories, has a strong visual component that represents entities and their connections.

2.5.3 Stories in data visualization

Often in narratives text or audio transmit the main story, while visualizations or images support the story or provide details. Comics and flowcharts are special types of narratives relaying mostly on visual components rather than text. Recently, we have seen an increase in integrating complex visualizations into narratives in many news organizations [98], journalism reports (e.g. New York Times, Washington Post, the Guardian), and television

reports [99]. Segel et al.[1] explore different aspects of narratives from a variety of sources and identify distinct genres of narrative visualization, see Figure 2.5.

Figure 2.5: Types of Narrative Visualization proposed by Segel et al. [1]

In the domain of business intelligence the main visualization tool is BI dashboards [28], collections of multiple visual components (e.g. charts, tables) on a single view [100].

Nevertheless, as pointed out by Wojtkowski and Wojtkowski [101], dashboards and other visualization tools used to analyze complex content cannot simply be used to tell stories.

They need to be ”tailored” to accommodate storytelling so as to better highlight items of importance within very large data resources [102], in a way that is not cumbersome for the storyteller and clear for the audience.

Some visualization systems began to integrate storytelling. For example GeoTime [103], a geo-temporal event visualization system, integrates a story system that shows events in space and time, hypertext linked visualizations, and visual annotations to create an envi- ronment for both analytic exploration, and story creation and communication Figure2.6.

Storytelling tools in the business domain are not yet as advanced. Systems like Sense.us [52] and Tableau [36] allow analysts to visualize their data, conduct their analysis, and store a history of the analysts exploration. This history can serve as a first step towards creating a story. Many Eyes [37], Tableau Public [104] and Sense.us [52] allow publishing of interactive visualizations online, and permit collaborative analysis through comments on a single visualization, creating an evolving analysis. Again this collaborative annotation can be seen as are a first step towards making a collaborative knowledge narrative, where an analysis story could be extracted from the visualizations and comments. Nevertheless, these tools do not provide explicit means to indicate story progression, and to highlight or explain relationships between multiple visualizations (seen in dashboards) which is key to intelligence analysis and communication [83].

Figure 2.6: Geotime shows events in time and space, in a X, Y and T coordinating space. [103]