FP-Krishna

From CS294-10 Visualization Sp11

(Difference between revisions)
Jump to: navigation, search
(Michael Porath - Apr 07, 2011 04:16:32 pm: new section)
(Michael Hsueh - Apr 07, 2011 11:52:27 pm: new section)
Line 65: Line 65:
Again, reiterating my comment in class, I would focus on just one aspect. That is, just the querying, or just the results. In terms of the visualization challenge I personally would chose the results, because querying would introduce lots of UI challenges that would take away from the focus on the visualization.
Again, reiterating my comment in class, I would focus on just one aspect. That is, just the querying, or just the results. In terms of the visualization challenge I personally would chose the results, because querying would introduce lots of UI challenges that would take away from the focus on the visualization.
As an inspiration you might want to look at the linkedin visualization from earlier this year: http://blog.linkedin.com/2011/01/24/linkedin-inmaps/
As an inspiration you might want to look at the linkedin visualization from earlier this year: http://blog.linkedin.com/2011/01/24/linkedin-inmaps/
 +
 +
== Michael Hsueh - Apr 07, 2011 11:52:27 pm ==
 +
 +
I am also curious about the interface of the tool, which might depend specifically on what domain it is targeted for as well as how general the tool is. The medical pattern query tool has a relatively well defined domain in terms of the types of elements that can be queried. For the purpose of visualizing the results, a limited knowledge domain can be exploited to improving the fidelity of the results... in your case with Presidents and physicists, for example, you might easily be able to color code the various types of relationships that might exist between nodes...

Revision as of 07:05, 8 April 2011

Contents

Group Members

Krishna Janakiraman

Description - Exploring time series through visual queries

Existing time series exploration techniques enable exploration via zooming the axes, drawing query contours, via techniques such as generalized selection, by bounding boxes and by building augmented suffix trees. However, these techniques do not seem to show the existence of recurring patterns at varying resolutions within the time series.

For example, a given 'U' or a downward slope shaped query pattern could be present in the time series in a stretched form and can be composed of smaller similarly shaped patterns. As another example, suppose a user wants to identify a sequence of events in a time series, the matching pattern could be a contiguous sequence of events or sequence of events interspersed with other unrelated events. At a higher resolution, certain classes of the latter pattern could be relevant for the user.

The main research questions I would like to investigate are: how can patterns {shapes, sequences and conditions} be expressed and identified across varying resolutions in a time series, how such patterns can be both relaxed and constrained to allow greater flexibility and finally how the results can be summarized and visualized.

Description - Visual queries and generalized selection for exploring knowledge representations

Traditional approaches towards querying and exploring knowledge representations (or) ontology instances have been through building question answering systems or through using sophisticated query languages such as SPARQL. While the latter is almost impossible to learn for non-technical users, the former is incredibly hard to implement. In addition, interfaces built using either of these approaches typically give 'point answers' and do not depict the rich network between subjects, objects and predicates while showing the results. For my final project, I will be developing a visual query language to explore a knowledge representation. Users can build sophisticated queries by drawing 'query graphs'. The result is displayed as a graph too, and the layout is determined by the query. Users can use generalized selection and brushing to prune the results and update the queries.

Initial Problem Presentation

I will be proceeding with Idea II

Link to slides: File:Krishna-initial-preso.pdf

Jvoytek - Apr 06, 2011 03:20:10 pm

This is a very interesting idea. How will the user interface work exactly though? How will you add new filtering criteria, what criteria will determine which type of selector (slider, text box, etc) will be associated with each filter?

Dan - Apr 06, 2011 03:18:57 pm

Interesting! Writing queries for finding representations of relationships between objects is definitely difficult... very good problem! It was hard to understand some of the ways of using datasets, for example, the idea using a notes data set, was that for music? Using graphs as queries seems like an interesting idea, but how usable would it really be for non-technical users? What is the target (non-technical) user? If it’s a web developer, then it makes more sense.

Siamak Faridani - Apr 06, 2011 03:31:16 pm

I really like this idea. I TA database systems here and I relate to this problem. The idea seems to be a great contribution. I was not sure how the UI is going to be constructed. Also how are you planning to implement things like join or functions like sum, avg and those things?

David Wong - Apr 06, 2011 03:24:18 pm

What knowledge repository do you plan on viewing initially? Will your approach be general to apply to any repository or a specific knowledge domain? Also, in regards to using a graphical structure to represent queries, you should consider which properties of graphs to leverage to make your system unique in comparison to what's already done in the literature (eg changing queries into nodes themselves). Ultimately, the question is how graphical queries differ and are better than using a traditional drop down menu approach.

Matthew Can - Apr 06, 2011 03:02:52 pm

I like the idea of doing visual queries and generalized selection on graph structures. Like David said, it's important to think of what interaction techniques a graph can afford. An initial study with some paper prototyping might be of use here, but I'm not sure if you'd have the time for that.

Julian Limon - Apr 06, 2011 05:25:15 pm

This is a very interesting project. I believe there could be a fair amount of overlap in your literature research for creating optimal graphs with our project. I'll be looking forward to hearing more about the solutions you chose and why you chose them.

Like I mentioned in class, I believe you could use visual cues to give people information that is beyond what they can get in a simple query generator. For example, using size or color for the objects in the graph while the user is constructing the query may help them understand the size of the retrieved data set.

I also liked the idea of using visual queries to find similar patterns. I think you can also add cardinality to the dotted line metaphor to allow the user to refine the query.

Michael Cohen - Apr 06, 2011 11:40:37 pm

As I looked at your examples, I became interested in how/if you'll represent the concept of distance. For instance, if a user just does a general query on presidents somehow related to physicists, they are likely to get a huge result graph since many presidents will be related to many physicists given enough "jumps". Specifying "through congressmen" would help, but may require more specificity than is practical in some cases. It seems to me that your UI could benefit from a way to describe a variable number of generic links (e.g. a symbol that means "1 to 3 degrees of separation") but this may be hard to communicate clearly within the graph paradigm.

Saung Li - Apr 07, 2011 12:37:37 am

I really like this approach to analyzing knowledge repositories. You may want to think about what would happen when the resulting graphs are huge, as this may be difficult to read and trace out paths between nodes. One cool thing though is you could display more information about a node when you hover over it.

Sally Ahn - Apr 07, 2011 03:53:26 am

This is an interesting idea. I wonder how generalizable such visual queries would be...in which datasets would it work well and where would it not? Thinking about this might help to narrow the scope of your project and to tailor the visualization for the appropriate datasets.

Michael Porath - Apr 07, 2011 04:16:32 pm

Again, reiterating my comment in class, I would focus on just one aspect. That is, just the querying, or just the results. In terms of the visualization challenge I personally would chose the results, because querying would introduce lots of UI challenges that would take away from the focus on the visualization. As an inspiration you might want to look at the linkedin visualization from earlier this year: http://blog.linkedin.com/2011/01/24/linkedin-inmaps/

Michael Hsueh - Apr 07, 2011 11:52:27 pm

I am also curious about the interface of the tool, which might depend specifically on what domain it is targeted for as well as how general the tool is. The medical pattern query tool has a relatively well defined domain in terms of the types of elements that can be queried. For the purpose of visualizing the results, a limited knowledge domain can be exploited to improving the fidelity of the results... in your case with Presidents and physicists, for example, you might easily be able to color code the various types of relationships that might exist between nodes...

Personal tools