Interactive Visual Exploration Of A Large Spatio Temporal

Published Date: 02 Nov 2017

Literature Survey

Breadth approach to explore research direction in information visualization

Yilin Gu

1. Introduction

In this paper, two articles with separate information goals are chosen to be compared and discussed. The first article is ��Interactive Visual Exploration of a Large Spatio-Temporal Dataset: Reflections on a Geovisualization Mashup�� by Jo Wood, which aims at developing a mashup approach to effectively visualize large geography based dataset. The second article is ��Extreme Visualization: Squeezing a Billion Records into a Million Pixels�� by Ben Shneiderman, which introduces various approaches to visualize billion records. The two articles will be discussed, summarized, and criticized below, and finally, this paper will compare these two articles and discuss which one is more close to my information goals.

2. Article discussion, summary and criticism

2.1 Article ��Interactive Visual Exploration of a Large Spatio-Temporal Dataset: Reflections on a Geovisualization Mashup��

? 2.1.1 Discussion and summary

In this paper, the author introduced a ��mashup�� approach, which integrates multiple open and freely-available sources to explore interactive visualization of large spatio-temporal dataset. Exploratory visual analysis can be regarded as an effective means of preliminary investigation. Compare to low level programming such as using C++ or Java which takes considerable development time, the mashup approach focuses effort on specification of visual encoding and interaction using high level KML markup and develop loose couplings between KML and a GIS to ensure the adaptability of visualization environment. In the approach, Google Earth and Google maps API are used since the technology are described as variously accessible, agile, adaptable and data rich; server side technology for retrieving information and generating content dynamically using PHP, and client side to enable user interaction using JavaScript and Ajax. MySQL is selected as database management system since it`s free and widely used for the storage and retrieval of large data sets and offers good integration with web services.[1]

The dataset explored here consists of 1.42 million requests made of a US-based mobile telephone service over a period of one month. The dataset could be explored in both time and space dimension. Several visual encodings and interactions are developed to demonstrate the effectiveness of this mashup approach. In the first example, placemarks were coloured according to Language ID in order to explore their spatial and thematic associations. Two issues are mentioned within it, the one is that high-resolution satellite imagery and aerial photography could obscures the detail of graphics described in KML, and the other issue is overplotting, which happen when crowded symbols need to be dealt with. Solutions are provided for these issues. In Google Earth, it automatically revealed as an explosion, while text culling and dithering could also reduce the consistency of mapping between geographical location and graphic. The second case is about the plotting locations with labels and data synthesis. Ancillary datasets like US population density data could be integrated with into the mashup, but too many points in the labeled query data were over plotted, and the author suggested two possibilities, one is that queries were made from a fixed position and the other is positions were simply assigned to the location of the nearest cell mast, like a zip code field.

The author also provide new visual encodings convey data at different spatial, temporal and attribute aggregations. Tag clouds, which are a visualization technique for summarizing the importance of words, are first introduced. Words size indicates its frequency of use, and clicking a word could result in the generation of KML that zooms in on the viewable area to which the word applies. While tag maps is similar to tag cloud, but the words get spatial attribute. Tag maps aggregate records at a scale appropriate to viewable window and a new request leads to a specific view. In this way, the tag map provides a multi-scale means to explore data at different spatial aggregations. Finally, the author provides a spatial analysis case for compare the population density and the mobile phone queries data, using LandSerf script, and as a result, different colors indicates different comparisons on the map as expected. [1]

In conclusion, the ��mashup�� approach in this paper identified patterns in the large data sets in time and space through tag maps, tag clouds and so on. LandSerf function was used to transform and aggregate data, and to calculate and smooth density surfaces. All the techniques and methods are proven to be effective to exploratory visual analysis. When it comes to flexibility, KML markup is implemented instead of low level development, so that it gains rapid development iteration and this is especially advantageous in the early iterations of visual exploration.

2.1.2 Criticism

First of all, hierarchical data organization may not be suitable for all data sets. Even though in this approach, KML is used to organized data according to space, time or attribute through multiple alternative hierarchies, other data may not be so amenable to this structure. Also, the dependent on APIs gets a degree of volatility over time, since API may evolve to different version and may not compatible with previous one.

2.2 Article ��Extreme Visualization: Squeezing a Billion Records into a Million Pixels��

? 2.2.1 Discussion and summary

This paper mainly introduced multiple methods to visualize billion record data sets. For millions of records, various approaches have been developed including implementing dynamic queries sliders to query and filter data, graphical selectors for query specification, and results displayed by information visualization techniques. These approached are proven to be effective for many tasks in which visual presentations enable discovery of relationships, clusters, outliers, gaps, and other patterns. When the data size comes from millions to billions level, rapid aggregation, meaningful coordinated windows, and effective summary graphics are needed, and the paper discuss three proposed solutions. [2]The first one is atomic visualizations, which means that one marker per data record. This method is effective when the records are ordered in useful way so that the image could reflect pattern within the data if each data record is represented by a marker. This method could also be applied to show the directory structure on a file server using treemap, and color encodes file time while area encodes file size. The second solution the author mentioned is aggregate visualization, which basically squeezes the information into million pixels and views it on a commonly available display. The author gives multiple figures to illustrate this concept. There are digital libraries, which offer a scalable approach where each axis is an expandable hierarchy and each grid cell contains colored dots indicating for different kind of documents. There are also scatter gram which using circles to indicate a baseball player, and the size is number of hits, and acts as an aggregate marker for their career data. For multivariate databases, it utilize matrix and each column represents one attribute, and each element in the matrix represents an aggregation of population which belongs to the crossing two attributes. So in this way, it is likely to get the strong positive linear correlations within the data set. The author also introduced tree structures, which is a general method to represent hierarchical structure. So are networks, which could allow users to see the main clusters, compare their sizes and understand the connectedness among clusters. [2] The third solution is density plot visualizations, which use color coded areas, and show users where to explore.

In conclusion, the author offers three methods to effectively visualize billions of data records, including atomic visualization, aggregate visualization, and density plot visualization, especially aggregate visualization, which shows the greatest promise since they promote the sense-making while keeping display complexity low.

2.2.2 Criticism

Even though this paper enumerates various methods for visualization of billions of records, it hardly gets any innovative ideals or approaches. All the methods mentioned here could be regarded as general ways of visualizing data, no matter for millions or billions of data records. In addition, the paper doesn`t explore and analysis a method deeply enough, like if certain type of data structure is better fit into one method than the other, or the relationship and difference between different methods. Finally, the structure is not clear enough, like the third part, density plot visualizations, is partly repeated the aggregation visualization, and some less relevant content is included like how to optimize a visualization approach on changing the contrast and darkness of an image, which is quite basic to information visualization.

3. Comparison and conclusion

These two papers both belong to information visualization area, while the former one focuses on exploration about agile method of developing visualization tools on large spatio-temporal dataset, and the latter one generally introduced the popular visualization methods when millions even billions of data records need to be visualized.

For the first article, the authors have effectively developed a mashup approach to exploratory visualization. Complicated and customizable spatial processing functionality is required in a geovisualization mashup to transform data and perform spatial and statistical comparisons. A flexible environment which could involve new techniques for visual representation is also critical. Traditionally, this adaptive feature requires low level programming, like using C++ or Java, and considerable development time. However, a sufficiently rapid development circle is need to allow prototyping, in this way, mashup approach, which integrate web-based technologies and create an application tailored to a specific task is more effective. So in the article, the author introduce each web technology used in this approach, and how this techniques come together to solve problems such as over plotting, interaction, flexibility. Also the author gives several new views of visualizing geographic data, like tag clouds, tag maps, and data dials. At the end, the article includes the spatial processing functionality part, which is quite valuable, since the processed data could reflect patterns more easily.

For the second article, the author generally enumerates various ways of visualizing large data records. But comparing to the first article, it doesn`t discuss a deeper relationship and differences between these methods mentioned and the conclusion is also a little superficial, since it only addresses the point that aggregate visualization show the greatest promise, which is obvious.

So, I prefer the first article, since it`s more close to my information goal. I am doing this research for exploring effective approach for information visualization, for which the first paper gives valuable suggestions. Mashup approach is heuristic since it makes use of various web interfaces and integrates them into application. The development iteration could be less and the platform could be more flexible. Also, the approach introduced in the first article is accessible and the detail explanations make it more persuasive and sounder.