Qing Luan
USTC
Hefei, China
Steven M. Drucker
Microsoft Live Labs
Redmond, WA
Johannes Kopf
University of Konstanz
Konstanz, Germany
Ying-Qing Xu
Microsoft Research
Asia Beijing, China
Michael F. Cohen
Microsoft Research
Redmond, WA
SUMMARY:
Panning and zooming interfaces for pictures with billions of pixels(gigapixel images) have recently appeared on the internet. The huge size of the images causes some issues that are discussed in this paper. There are three main types of annotations a text label, an audio loop, or a narrative audio. The main contribution of this paper is a distance function between the individual annotations and views of the image that guide the rendering of both audio and text annotations.
DISCUSSION:
Although this paper is somewhat short, they introduce some important concepts. Annotating gigapixel images is going to be important as computers become faster and more powerful. It is important to annotate the images in a way that is quick and doesn't obstruct the user's view of the image.
Sunday, March 21, 2010
Saturday, March 20, 2010
UIST2008: Kinematic Templates for Content-Relative Cursor Manipulations
Richard Fung, Edward Lank, & Michael Terry
University of Waterloo, Computer Science
Waterloo, Ontario, Canada
Celine Latulipe
University of North Carolina at Charlotte
Charlotte, NC, USA
SUMMARY:
The authors of this paper explore kinematic templates. Kinematic templates are "an end-user tool for defining content-specific motor space manipulations in the context of editing 2D visual compositions." The picture below will give you a better idea of what they are. The shape on the left is the user input. The shapes with the background grids are kinematic templates. As you can see the software automatically adjusts the pend stroke depending on the template.
Previous work in a similar area has been done on snap and go cursor movements. Snap and go techniques cause the line or shape to snap or connect to another shape in the general area. Kinematic templates differ because they change anything written anywhere on its surface. Kinematic templates can be defined anywhere on the drawing surface and they can be of any size. Below is a result from their study where a user drew a sun.
In this drawing two templates were used. The first was a compass template to help the users draw near perfect circles. The second is a sand paper template which modifies the control-display gain of the cursor. Control-display gain is the when the cursor moves at a slower or quicker speed than what the user actually inputs. In this example the control-display gain is lowered to help the user draw the sun within a certain region.
The authors distinguish between two types of templates:
Passive Templates - Alter the control display gain
Active Templates - Alter the cursor movement independent of changes in the x and y coordinates
An example of an active template would be the magnetic template. This template actively pulls the cursor towards a defined point. It is important to note that ALL templates affect the cursor only when the mouse button is pushed down.
In conclusion templates provide a way of manipulating cursor placement without strictly prescribing output. They lie somewhere between free hand drawing and snap and go techniques.
DISCUSSION:
This is a very interesting tool that could provide useful for programs like paint, photoshop, and paint shop pro. It would also be interesting to try and apply this technique to our second project. It might even provide more useful when using a drawing pad than a mouse.
University of Waterloo, Computer Science
Waterloo, Ontario, Canada
Celine Latulipe
University of North Carolina at Charlotte
Charlotte, NC, USA
SUMMARY:
The authors of this paper explore kinematic templates. Kinematic templates are "an end-user tool for defining content-specific motor space manipulations in the context of editing 2D visual compositions." The picture below will give you a better idea of what they are. The shape on the left is the user input. The shapes with the background grids are kinematic templates. As you can see the software automatically adjusts the pend stroke depending on the template.
In this drawing two templates were used. The first was a compass template to help the users draw near perfect circles. The second is a sand paper template which modifies the control-display gain of the cursor. Control-display gain is the when the cursor moves at a slower or quicker speed than what the user actually inputs. In this example the control-display gain is lowered to help the user draw the sun within a certain region.
The authors distinguish between two types of templates:
Passive Templates - Alter the control display gain
Active Templates - Alter the cursor movement independent of changes in the x and y coordinates
An example of an active template would be the magnetic template. This template actively pulls the cursor towards a defined point. It is important to note that ALL templates affect the cursor only when the mouse button is pushed down.
In conclusion templates provide a way of manipulating cursor placement without strictly prescribing output. They lie somewhere between free hand drawing and snap and go techniques.
DISCUSSION:
This is a very interesting tool that could provide useful for programs like paint, photoshop, and paint shop pro. It would also be interesting to try and apply this technique to our second project. It might even provide more useful when using a drawing pad than a mouse.
IUI2009: Understanding the intent behind mobile information needs
Karen Church
Telefonica Research
Barcelona, Spain
Barry Smith
CLARITY
Belfield, Dublin, Ireland
SUMMARY:
The authors who wrote this paper performed several user studies that were meant to explore the goals behind mobile information needs. Devices like the iPhone have greatly increased the number of mobile users who access the world wide web. In particular web search needs differ for mobile users and PC users. The studies found that mobile user needs differ depending on time and location. Non-informational needs account for 42% of user needs while geographical needs accounted for 31%. Personal information was the intent only 10% of the time.
DISCUSSION:
This paper contains valuable information for any software engineer that works with mobile devices. Knowing what information people are looking for on their mobile phone can help programmers design their applications appropriately. This would also be useful for a company like Google or Microsoft that try to provide optimal search functionality for their users.
Telefonica Research
Barcelona, Spain
Barry Smith
CLARITY
Belfield, Dublin, Ireland
SUMMARY:
The authors who wrote this paper performed several user studies that were meant to explore the goals behind mobile information needs. Devices like the iPhone have greatly increased the number of mobile users who access the world wide web. In particular web search needs differ for mobile users and PC users. The studies found that mobile user needs differ depending on time and location. Non-informational needs account for 42% of user needs while geographical needs accounted for 31%. Personal information was the intent only 10% of the time.
DISCUSSION:
This paper contains valuable information for any software engineer that works with mobile devices. Knowing what information people are looking for on their mobile phone can help programmers design their applications appropriately. This would also be useful for a company like Google or Microsoft that try to provide optimal search functionality for their users.
IUI2009: Parakeet A Continuous Speech Recognition System for Mobile Touch-Screen Devices
Keith Vertanen
Per Ola Kristensson
Cavendish Laboratory, University of Cambridge
SUMMARY:
This paper presents a continuous speech recognition system for mobile touch screen devices. Participants had a word entry rate of 18 words per minute while seated and 13 words per minute while walking around. In the past speech recognition methods have suffered from poor accuracy. This paper uses Parakeet which is a touch screen system for recognizing continuous speech. Parakeet was the first system to test voice recognition while walking around.
DISCUSSION:
In my mind this system fails to prove it is accurate enough for people to actually use. It mentions that when the intended sentence was in the confusion network it was able to find and select the correct sentence 96% of the time. Future work would have to be done to improve performance with sentences that are not in the confusion network.
Per Ola Kristensson
Cavendish Laboratory, University of Cambridge
SUMMARY:
This paper presents a continuous speech recognition system for mobile touch screen devices. Participants had a word entry rate of 18 words per minute while seated and 13 words per minute while walking around. In the past speech recognition methods have suffered from poor accuracy. This paper uses Parakeet which is a touch screen system for recognizing continuous speech. Parakeet was the first system to test voice recognition while walking around.
DISCUSSION:
In my mind this system fails to prove it is accurate enough for people to actually use. It mentions that when the intended sentence was in the confusion network it was able to find and select the correct sentence 96% of the time. Future work would have to be done to improve performance with sentences that are not in the confusion network.
IUI2009:Learning To Recognize Value Tags
Shilad Sen, Jesse Vig, John Riedl
Grouplens Research, University of Minnesota
SUMMARY:
This paper discusses automatic tagging systems.
Wikipedia : In online computer systems terminology, a tag is a non-hierarchical keyword or term assigned to a piece of information (such as an internet bookmark, digital image, or computer file)
The authors looked at 21 different tagging algorithms and chose the best three to deploy on a movie website. The rest of the paper talks about the different methods.
DISCUSSION:
This is a very useful paper for someone looking to learn some tagging methods for their application/website. Future work involves testing their methods on more than just a movie application.
Grouplens Research, University of Minnesota
SUMMARY:
This paper discusses automatic tagging systems.
Wikipedia : In online computer systems terminology, a tag is a non-hierarchical keyword or term assigned to a piece of information (such as an internet bookmark, digital image, or computer file)
The authors looked at 21 different tagging algorithms and chose the best three to deploy on a movie website. The rest of the paper talks about the different methods.
DISCUSSION:
This is a very useful paper for someone looking to learn some tagging methods for their application/website. Future work involves testing their methods on more than just a movie application.
Sunday, March 14, 2010
CHI 2009b: A Sustainable Identity The Creativity of an Everyday Designer
Ron Wakkary and Karen Tanenbaum
School of Interactive Arts and Technology
SUMMARY
This paper explores sustainability in interaction design by reframing concepts of user identity and use in a domestic setting. The paper starts off by mentioning Western cultures are typically very wasteful. This means designers need to design for sustainability. The study in this paper focuses on interaction design principles in the domestic environment. They basically said that many designs in the home can be reusable. This will increase a products life cycle and mitigate harm to the environment.
DISCUSSION:
I thought I was going to be reading a paper about good design principles. The authors of this paper seemed to be environmentalists. All the paper did was talk about designing for sustainability. This means design with materials that will last people longer so they don't have to throw them away as quickly.
School of Interactive Arts and Technology
SUMMARY
This paper explores sustainability in interaction design by reframing concepts of user identity and use in a domestic setting. The paper starts off by mentioning Western cultures are typically very wasteful. This means designers need to design for sustainability. The study in this paper focuses on interaction design principles in the domestic environment. They basically said that many designs in the home can be reusable. This will increase a products life cycle and mitigate harm to the environment.
DISCUSSION:
I thought I was going to be reading a paper about good design principles. The authors of this paper seemed to be environmentalists. All the paper did was talk about designing for sustainability. This means design with materials that will last people longer so they don't have to throw them away as quickly.
Saturday, March 13, 2010
CHI 2009b: NewsCube Delivering Multiple Aspects of News to Mitigate Media Bias
Souneil Park, Seungwoo Kang, Sangyoung Chung, Junehwa Song
KAIST Computer Science, South Korea
SUMMARY:
This paper discusses a web service called NewsCube. NewsCube aims to mitigate the effects of media bias. They are the first know experiment to tackle the problem computationally. They mention they can reduce bias, but underlying causes make it impossible to eliminate bias. Look at the picture below to see how news gets from reality to print.
Some groups such as fair.org, NewsTrust.net, and skewz.com have tried to serve as watchdog groups on media bias. Problem is these news is usually immediate and can't wait for user reviews. NewsCube focuses on letting the user develop their own view on media bias.
NewsCube uses news structure-based extraction which involves classifying bias of a news article based on the aspect the author is writing the story from. For example, a news article could state Obama passes health care. While an article on the same topic from a different aspect might say McCain criticizes Obama's health care plan. NewsCube aims to distinguish the two views and present both of them to the user.
The authors claimed that NewsCube was effective. They stated they were planning on incorporating it as a service with a major internet news company in Korea.
DISCUSSION:
It is hard to evaluate NewsCube without actually using it but according to the study it works. The main problem is that there is no perfect algorithm for evaluating bias on a news article with an undefined subject. However, this product still interests me since I believe at critical times mainstream news can often be very biased.
KAIST Computer Science, South Korea
SUMMARY:
This paper discusses a web service called NewsCube. NewsCube aims to mitigate the effects of media bias. They are the first know experiment to tackle the problem computationally. They mention they can reduce bias, but underlying causes make it impossible to eliminate bias. Look at the picture below to see how news gets from reality to print.
Some groups such as fair.org, NewsTrust.net, and skewz.com have tried to serve as watchdog groups on media bias. Problem is these news is usually immediate and can't wait for user reviews. NewsCube focuses on letting the user develop their own view on media bias.
NewsCube uses news structure-based extraction which involves classifying bias of a news article based on the aspect the author is writing the story from. For example, a news article could state Obama passes health care. While an article on the same topic from a different aspect might say McCain criticizes Obama's health care plan. NewsCube aims to distinguish the two views and present both of them to the user.
The authors claimed that NewsCube was effective. They stated they were planning on incorporating it as a service with a major internet news company in Korea.
DISCUSSION:
It is hard to evaluate NewsCube without actually using it but according to the study it works. The main problem is that there is no perfect algorithm for evaluating bias on a news article with an undefined subject. However, this product still interests me since I believe at critical times mainstream news can often be very biased.
CHI 2009b: How Well do Visual Verbs Work in Daily Communication for Young and Old Adults?
Xiaojuan Ma, Princeton University Computer Science
Perry R. Cook, Princeton University Computer Science/Music
SUMMARY:
This paper studies how verbs are visually conveyed in daily communication for young and old adults. There are four visual modes compared in this study:
DISCUSSION:
This paper caught my eye at first but wasn't as interesting as I thought. I was hoping they would go into further detail about how to design images that communicate cross culture. I felt rather than focus on age difference, this study should have focused on people from different cultures.
Perry R. Cook, Princeton University Computer Science/Music
SUMMARY:
This paper studies how verbs are visually conveyed in daily communication for young and old adults. There are four visual modes compared in this study:
- Single static image
- A panel of four static images
- An animation
- A video clip
- Verb visualization - represent verb by using photos, signs, and other graphic designs
DISCUSSION:
This paper caught my eye at first but wasn't as interesting as I thought. I was hoping they would go into further detail about how to design images that communicate cross culture. I felt rather than focus on age difference, this study should have focused on people from different cultures.
Tuesday, March 9, 2010
CHI 2009b: The Performance of Touch Screen Soft Buttons
Seungyon "Claire" Lee
Georgia Institute of Technology
Atlanta, Georgia
Shumin Zhai
IBM Almaden Research Center
San Jose, California
SUMMARY:
This paper discusses the performance of soft buttons on different types of touchscreens. A soft button refers to any button that is generated by the hardware. A hard button is an actual hardware button either on the screen or on an input device. The advantage of soft buttons is they can appear and disappear according to the user's context. The disadvantage of soft buttons is that it is difficult to give user feedback as to whether or not it is being pressed.
A previous study at the University of Maryland showed that when the cursor was placed 1/2 inch above the finger tip, less errors were made. More recent studies have shown complex techniques that amplify the fingertips movement and placement which is a tradeoff for detection speed.
One issue the authors plan to investigate is the accuracy of a soft button click with a stylus versus the accuracy of a fingertip. Another issue is button size. A previous study at grocery stores suggested 22mm as the minimum suggested size for a soft button.
The authors performed tests on five products:
Experiment 1: Operating Mode and Feedback
Experiment 2: Contact Force vs. Force Activation
In general there was no difference between the two types of touchscreens. I believe they should have used a phone that has a hard flip out keyboard. They also should have tested sending SMS messages in one of their experiments.
Georgia Institute of Technology
Atlanta, Georgia
Shumin Zhai
IBM Almaden Research Center
San Jose, California
SUMMARY:
This paper discusses the performance of soft buttons on different types of touchscreens. A soft button refers to any button that is generated by the hardware. A hard button is an actual hardware button either on the screen or on an input device. The advantage of soft buttons is they can appear and disappear according to the user's context. The disadvantage of soft buttons is that it is difficult to give user feedback as to whether or not it is being pressed.
A previous study at the University of Maryland showed that when the cursor was placed 1/2 inch above the finger tip, less errors were made. More recent studies have shown complex techniques that amplify the fingertips movement and placement which is a tradeoff for detection speed.
One issue the authors plan to investigate is the accuracy of a soft button click with a stylus versus the accuracy of a fingertip. Another issue is button size. A previous study at grocery stores suggested 22mm as the minimum suggested size for a soft button.
The authors performed tests on five products:
- Apple Iphone (capacitive sensor, finger)
- HTC Touch phone (resistive sensor, finger or stylus)
- Canon DK10i (calculator with hard buttons)
- HP iPAQ 6315 smartphone (resistive sensor, finger or stylus, big like calculator)
- Nokia 6269 (mass market hard buttons)
Experiment 1: Operating Mode and Feedback
- Enter a multiplication operation, required 8 digits and 2 operator (x,=)
- Tested Canon calculator vs. HP smartphone
- Tested 4 different categories of feedback
- No feedback
- Audio feedback
- Tactile feedback
- Audio + tactile feedback
- Speed was similar in both cases
- Error rate was high with no feedback for the touch screen
- Audio + tactile feedback isn't necessary, just choose one
Experiment 2: Contact Force vs. Force Activation
- Iphone, HTC Touch, Nokia 6260
- Task was to dial 7 digit phone numbers
- In this experiment very few errors were made, speed was similar
- There wasn't a significant difference in any of the phones at this task
- Tested soft screen button sizes on the iPhone(contact) and HTC Touch(force)
- User had to enter a number, performance was measured by clicks per second
- Two phones had similar results
- Showed wide buttons(7.2mm) had better performance(clicks per second) than narrow buttons(4.9mm)
In general there was no difference between the two types of touchscreens. I believe they should have used a phone that has a hard flip out keyboard. They also should have tested sending SMS messages in one of their experiments.
Thursday, March 4, 2010
IUI2009: A Low-Order Markov Model integrating Long-Distance Histories for Collaborative Recommender Systems
SUMMARY:
This paper discusses a method to extract the most pertinent resources for a given user. This is a challenge for applications in areas such as e-commerce, resource acces, web navigation, and online communities.
One previous method, known as collaborative filtering, determines what resources to recommend based on similar users. Resources that have been highly rated in the past by similar users are recommended to the user. The traditional problems with this method are most calculations are done online and the context of the user is not considered. This causes a lack of speed and accuracy.
One way of improving upon online collaborative filtering is to use data mining techniques. This requires having a database of user histories to analyze. The data is made up of a set of users(U) and a set of resources(R).
Two techniques explained in the paper are association rules and sequential patterns. An association rule is when one resource U1 usually leads to another resource U2 being pertinent for the user. Thresholds are need to determine if there is an association between resources. Association rules can contain more than just 2 resources. Sequential patterns are similar to association rules but here the order matters. This results in sequential patterns being more constrained than association rules.
Another approach used in this paper is know as Markov Models. Markov models limit the window of history that is used for data mining on the user. A kth order Markov Model looks at k+1 history of resources. In this paper they use a second order markov model which looks at a history size of 3.
The final technique used in this paper is skipping. Skipping simply means that you skip over certain resources to save either time or space. The authors talk about several time when skipping might be applicable.
The authors of this paper combined several of these techniques to create a Skipping Based Recommender. In their study they investigate some of the tradeoffs between the methods to see what works best.
The most important finding in there study was that a window size of three for their Markov model worked fastest but was least accurate. They decide that the best solution is to calculate resources from a Markov model of three, then if there is more time, calculate pertinent resources based on higher order Markov models.
Discussion:
This article was confusing at first but became very interesting after I got a better understanding of what the authors were creating. The Skipping Based Recommender system would be very useful for a program like iTunes when they want to display songs that a user is most likely could by. This would result in a user being more likely to buy the song resulting in increased profits. There are many programs and websites who could benefit from this sort of sytem.
This paper discusses a method to extract the most pertinent resources for a given user. This is a challenge for applications in areas such as e-commerce, resource acces, web navigation, and online communities.
One previous method, known as collaborative filtering, determines what resources to recommend based on similar users. Resources that have been highly rated in the past by similar users are recommended to the user. The traditional problems with this method are most calculations are done online and the context of the user is not considered. This causes a lack of speed and accuracy.
Sergei Brin - Co-founder of google, famous for his data mining techniques
One way of improving upon online collaborative filtering is to use data mining techniques. This requires having a database of user histories to analyze. The data is made up of a set of users(U) and a set of resources(R).
Two techniques explained in the paper are association rules and sequential patterns. An association rule is when one resource U1 usually leads to another resource U2 being pertinent for the user. Thresholds are need to determine if there is an association between resources. Association rules can contain more than just 2 resources. Sequential patterns are similar to association rules but here the order matters. This results in sequential patterns being more constrained than association rules.
Another approach used in this paper is know as Markov Models. Markov models limit the window of history that is used for data mining on the user. A kth order Markov Model looks at k+1 history of resources. In this paper they use a second order markov model which looks at a history size of 3.
The final technique used in this paper is skipping. Skipping simply means that you skip over certain resources to save either time or space. The authors talk about several time when skipping might be applicable.
The authors of this paper combined several of these techniques to create a Skipping Based Recommender. In their study they investigate some of the tradeoffs between the methods to see what works best.
The most important finding in there study was that a window size of three for their Markov model worked fastest but was least accurate. They decide that the best solution is to calculate resources from a Markov model of three, then if there is more time, calculate pertinent resources based on higher order Markov models.
Discussion:
This article was confusing at first but became very interesting after I got a better understanding of what the authors were creating. The Skipping Based Recommender system would be very useful for a program like iTunes when they want to display songs that a user is most likely could by. This would result in a user being more likely to buy the song resulting in increased profits. There are many programs and websites who could benefit from this sort of sytem.
BOOK: Emotional Design
SUMMARY:
The first chapter talks about why attractive designs work better. This is because machines can cause people to experience emotions. This results in people perceiving that beautiful things work better. Norman claims there are three levels of thinking: visceral(automatic), behavioral(every day), and reflective(contemplative). In this chapter he also mentions that it is trick to design things that accommodate both creative thinking and focus.
Three levels mapped to product characteristics
Visceral ==> Appearance
Behavioral ==> Pleasure & effectiveness of use
Reflective ==> Self-image, personal satisfaction, memories
In the second chapter Norman says to create a product that is truly long lasting it must evoke memories. This will help a person become emotionally attached to an object. Another way to induce positive emotions is for a product to get the user to feel positive about themselves. Products also contain a personality. This is another important aspect when designing for a positive emotional response.
The third chapter talks about the visceral, behavioral, and reflective design in products already out there. It reinforces the three levels mapped to product characteristics. In the fourth chapter he talks about designing for fun and pleasure. He also mentions designing for seduction. Another way to design for emotion is to use sounds. Music and sounds can bring out all kinds of emotions. He then talks about how people video games and movies seduce people.
The fifth chapter mentions how people often get mad at inanimate objects. This is a perfect example of people having emotions from design.
DISCUSSION:
I thought this book started off interesting then slowly got off topic as the book went on. I thought the distinction between three levels of thinking was clever. It allows software engineers to think on how people will react to their programs on three different levels. Towards the end of the book he started talking about robots. It was interesting at times but i felt like it contained little applicable knowledge. The last two chapters were more science fiction than fact.
The first chapter talks about why attractive designs work better. This is because machines can cause people to experience emotions. This results in people perceiving that beautiful things work better. Norman claims there are three levels of thinking: visceral(automatic), behavioral(every day), and reflective(contemplative). In this chapter he also mentions that it is trick to design things that accommodate both creative thinking and focus.
Three levels mapped to product characteristics
Visceral ==> Appearance
Behavioral ==> Pleasure & effectiveness of use
Reflective ==> Self-image, personal satisfaction, memories
In the second chapter Norman says to create a product that is truly long lasting it must evoke memories. This will help a person become emotionally attached to an object. Another way to induce positive emotions is for a product to get the user to feel positive about themselves. Products also contain a personality. This is another important aspect when designing for a positive emotional response.
The third chapter talks about the visceral, behavioral, and reflective design in products already out there. It reinforces the three levels mapped to product characteristics. In the fourth chapter he talks about designing for fun and pleasure. He also mentions designing for seduction. Another way to design for emotion is to use sounds. Music and sounds can bring out all kinds of emotions. He then talks about how people video games and movies seduce people.
The fifth chapter mentions how people often get mad at inanimate objects. This is a perfect example of people having emotions from design.
DISCUSSION:
I thought this book started off interesting then slowly got off topic as the book went on. I thought the distinction between three levels of thinking was clever. It allows software engineers to think on how people will react to their programs on three different levels. Towards the end of the book he started talking about robots. It was interesting at times but i felt like it contained little applicable knowledge. The last two chapters were more science fiction than fact.
Subscribe to:
Posts (Atom)