Search

From CS260Wiki
Jump to: navigation, search

Lecture Slides

File:Cs260-slides-23-search-small.pdf

Extra Materials

Discussant's Slides and Materials

Reading Responses

David Wong - 11/20/2010 21:33:02

1) The first chapter of The Design of Search User Interfaces discussed how search user interfaces follow a set of 8 design principles. It goes into detail about related work to each design principle and how these principles have been realized in actual search interfaces. The 7th chapter of Exploring and Finding Information describes information search with an analogy to foraging. There is an in depth discussion of how it applies to a Scatter/Gather interface and how it might apply to the user behavior on the Internet.

2) The first chapter of The Design of Search User Interfaces summarizes well the recent innovations in search interfaces. While I don't think I've learned anything new from reading the chapter, it gives more breadth and context to how these innovations in search iterfaces have developed and the extent of research that has gone into these design principles. As a testament to the success and accuracy of these design principles, most of these innovations seem natural and obvious, such as search completion and spelling correction. Altogether, the chapter offers a good summary and paints a clear picture of what it's like to design search user interfaces.

The Exploring and Finding Information paper takes an information centered approach on analyzing how a user will interact with a website. While I think that this is an interesting and valuable approach to analyzing user behavior, the article took a long time to lay down the theoretical foundation of its argument and spent a lot of time describing Scatter/Gather, an interface that seems ancient and not very useful. Its main point was to analyze how a user makes decisions based off of information scent, a set of constraints, and cost/benefit analysis. This seems obvious, but I liked how the article described using simulations on how information is processed to predict user behavior. What was especially interesting was the article's description of future work and how one might analyze and model how a user responds to images, rich content, and media on a website. Overall, I think that analyzing user interfaces and user interaction using a content based approach as opposed to the actual application UI is helpful.

3) The first chapter of The Design of Search User Interfaces demonstrates the need of the design principles well. It goes into detail on how these design principles have evolved in search user interfaces and how they have been realized through recent innovations. While it is impossible to create an all-encompassing list of design guidelines, the paper takes the 8 stated by Schneiderman, and clearly illustrates how these principles have been utilized in everyday search engines.

The argument provided in the Exploring and Finding Information article was well established. The article went first discussed the fundamental theory, other related work, applied it concretely to their own system, and then discussed how it might be applied to the web. I think the article did a good job of illustrating how might their approach be useful in analyzing user interfaces.


Kurtis Heimerl - 11/21/2010 0:24:40

Exploring and Finding Information

This paper describes the state of the art (circa 2002) in information retrieval. The basic premise is that they use a "foraging" model to explain user behavior,and provide tools that allow for users to reason intelligently about the 'optimal foraging path" to find their data.

First off, lolz pre-google search. Secondly, I have so many issues with this work. Evolutionary metaphors just upset me at my core, presumably because of my stalwart belief in intelligent design. Actually, now that I've made that joke, it almost seems honest. It's like those people arguing agriculture and medicine are bad things because they subvert evolution. They mention this briefly in the paper, but foraging behavior, even if we actually understood it completely, was constructed in an environment with many different constraints from web search. For instance, I'm not as likely to be eaten by a bear when searching for a website.

Aside from this fact, I also disagree with the fundamental assumption that we should be doing things in the most "natural" way. That's the naturalistic fallacy, and it is well known. Evolution is a constant endeavor, and short temp optimizations toward ancient hunter-gatherer societies just seems like folly.

The chapter itself was fine, examples were fine, and google's success really makes the whole thing seem silly. Fundamentally, I don't think it proved it wrong, really. It just made the search tree cap off at four nodes. Not much scent needing to be followed when you have either planted the seeds yourself, or own an industrial strength blueberry finder.

Their future work was a little ray of HCI trying to address the problems of other fields. Has HCI really stepped into these questions of "big data"? I don't think so, and I do think there's value in mixing those two up. Maybe there's some papers I don't know about...


Bryan Trinh - 11/21/2010 13:54:14

The Design of Search User Interfaces

This excerpt from the book Search User Interfaces focuses on the creating user interfaces for search. After a brief description of the draw backs and history of search interfaces, they propose a number of guidelines to designing effective search interfaces: offer informative feedback, support user control, reduce short-term memory load, provide shortcuts for skilled users, reduce errors; offer simple error handling, strive for consistency, permit easy reversal of actions, design for closure.

In general summaries of design guidelines for a certain domain of design tend to be useful. They serve as an easy to reference check list that will point the designer in the right direction from the onset of design. It also serves as a very good reference for further reading into the original research that first published the ideas the author is summarizing.

Just recently Google has been changing up their search UI quite a bit--changes that reflect some of these guidelines. Google has made it so that the searcher no longer needs to press enter/return to execute the search and instead it just happens automatically with every keystroke. This provides very quick feedback for the user to prune through the data before refining her search. To take it even further, just last week Google has provided a way for searches to look at small thumbnails of the web pages before opening them. An interesting shortcut that they have that I just recently noticed is the use of an arrow to point to the page. This arrow can be further controlled by moving the keyboard arrow keys to either scroll down the search results or drill into them.

The authors could have provided a set of guidelines in choosing guidelines. They stated early on that these guidelines are not expected to be used at the same time for one user interface. The next logical step to me was, which of these guidelines should be used together and with what weight.


Charlie Hsu - 11/21/2010 15:12:04

Design of Search User Interfaces

This reading discussed search user interface design, focusing on some design guidelines specific to search user interface design. Search user interfaces should be designed to offer efficient and informative feedback, balance user control with automation, reduce short-term memory load, allow for expert use via shortcuts, attempt to reduce the number of user errors, take into account small details in search query, and be aesthetically pleasing. Each of these guidelines is explored in the context of a search user interface, and certain characteristics of search users and the Web search experience are used to justify the guidelines.

A common theme in all of HCI research is examining the tradeoff between allowing for direct manipulation and creating more automated agents. In the context of attempting to support rapid, relevant response, the paper mentions that long retrieval times should not unduly penalize the user. Perhaps in this context, it is best to err on the side of greater automation: if a query takes a long time, the system could opt to return more information based not only on relevancy but also predicted future iterative searches. This allows people to be rewarded for waiting; airline flight searches and public transportation searches often show a large timetable of flights/trains in close proximity to the original search query, making reasonable estimates on the user's train of thought.

It also seems like there are particularly dangerous situations caused by the nature of search that search user interface design needs to deal with. Though it is important to address the vocabulary problem, dealing with word choice amongst different statements where the meaning is the same, it is important that morphological analysis is not done too aggressively, as seen by the example in the paper ("typing" converted to "type", which could mean a type categorization or the action of type). Reducing the amount of clutter on the screen and keeping the interface simple is countered by attempting to reduce short-term memory load by putting MORE information on the screen. Search user interfaces ultimately deal heavily with artificial intelligence, attempting to make sense out of unreliable human input, and the design tradeoffs for how much to analyze and how to prompt users for more information are important to consider.


Brandon Liu - 11/21/2010 16:41:42

"The Design of Search User Interfaces"

Two points I found interesting from the article:

1. The idea that search interfaces could 'penalize' users, decreasing their willingness to use the interface further. This came up when search results took too long to display, so that irrelevant search results due to the wrong query were wasted time. This point was a good justification for immediate search feedback like Google Instant.

2. The operation to be optimized is to get users off the search engine. While other UIs use engagement as a metric, the case with search engines is that less time spent on the UI is better.

Something that could have been improved in the article is more discussion of the statement "Very few members of the lay public understand Boolean syntax and even fewer are willing to learn command languages." This seems to imply that natural language search interfaces (e.g. Wolfram Alpha) are better since inexperienced users don't have to learn the interface. The issues here seem to be similar with those of end-user programming. In my opinion, an open possibility is that users of a UI can be trained into learning a query language if it has sufficiently high payoff (a better experience).

One thing outside of the scope of the article but relevant to the topic is the fact that creating good search result pages is an adversarial process. The two enemies of good search results are:

1. Search engine optimization by the actual content providers. The ideas in the paper really only work under the assumption that the pages being indexed are designed independently of the search results. When shopping or linkfarm sites are introduced that try to 'hook' users with their lead-ins, designing search result pages is much harder.

2. Trademark law. Search engines that sell ad slots need to be extremely careful to make sure that searchers can distinguish between fair use of a trademark in advertising and trademark infringement. This has huge implications for how Google represents paid advertising at the top left of its pages.


Dan Lynch - 11/21/2010 16:57:40

Search User Interfaces

This article starts off by comparing Google's search interface to a search engine from 1997 claiming that they were nearly identical. The idea is that the paradigm is to enter keywords in a form and then the results are displayed in a vertical list. They also brought up many of the misconceptions that many people have with the underlying implementations of search interfaces, particular in regards to boolean operations. Originally, searching for information was for a small educated group, but after the web bubble blew up, searching for information has become something for the masses. How do we deal with this? Well, the authors of this article proposed that we focus on 5 elements of usability: Learnability, Efficiency, Memorability, Errors, and Satisfaction.

Overall I think this paper is extremely relevant because finding information these days is a daily task for people of all backgrounds. Finding methods to optimize all search queries, provide non-empty results, and provide error-free results is a must! However, I think many searches (besides Google) are lacking. Many database searches are advanced and work well, however, I don't think they take into consideration a lot of what was discussed in this paper. I think that although they support boolean operations, they take searches verbatim and also often return empty result sets. I guess these search engines are still stuck in the 90s.


Thejo Kote - 11/21/2010 17:15:42

The design of search user interfaces:

In this book chapter, Hearst provides guidelines for the design of interfaces in general and search user interfaces in particular. She provides a historical overview of the design of search interfaces and ideas for good design by surveying relevant research in the field. She focuses on the holistic intent of making a search user interface "usable", which incorporates learnability, efficiency, memorability and satisfaction.

With a focus on search, the guidelines are practical in terms of what what they should offer users. Hearst's suggestions cover the major areas of feedback, cognitive aspects of design, the importance of details and aesthetics. She provides a number of useful examples from contemporary systems to illustrate her point. It was interesting to think of the evolution of web search over the last decade in the context of the design recommendations she makes. Even ongoing changes like the introduction of Google Instant Search incorporate a number of the best practices suggested by Hearst and I now better understand the rationale for them.


Linsey Hansen - 11/21/2010 17:25:42

====The Design of Search User Interfaces====

In this reading, the author describes search interfaces and design practices and strategies that they should follow.

One thing I found interesting was the first section on simplicity. While it definitely makes sense after reading it that having an overly complicated search interface would suck, I have never really thought about it. The fact that people generally do like putting all of their focus on some task while searching, and do not want to be distracted by obnoxious questions makes sense, and explains why interfaces with lots of search fields are generally not appreciated. Even though there are a lot of sites that offer both basic and advanced searches, there are still many that only offer some weird combination, and these are generally sites that are not used often, or that may quickly lose a user base.

While I am not sure what “immediately” might mean when the author says to show search results “immediately,” (ie as soon as the search is submitted vs. not showing some extra junk before showing the results) I do find it neat to relate that to Google’s current instant search, where it shows results as the user types the words. The way Google can now also show a page preview immediately upon hovering over the result link is also an awesome example of getting the search results to the user as soon as possible, because it can often give the user a better idea of the content and layout of the site without them needing to load it, see it isn’t what they want, then go back to the search results (which can get old).

One thing that I do not fully agree with is how Google was attributed with having good “aesthetics.” If anything, I would say that Google is just simple, which is awesome, but it could in theory look a lot better. What really probably matters is the fact that because Google is simple, it isn’t cluttered, which MSN and Yahoo definitely are, so this is more a matter of clutter vs. no clutter. Honestly, even if there was another search that worked like Google but looked more aesthetic, I would probably use Google anyways out of attachment for its silly name. But yeah, I feel like what the author is describing is just simplicity, cleanliness and the ease of finding things, while aesthetics (which I guess could refer to looks) means more the prettiness factor, kind of like what bing has.


Richard Shin - 11/21/2010 17:43:24

The design of search user interfaces

This book chapter discusses the special considerations and history of, as well as guiding principles for, the design of search user interfaces. The author overviews the historical forces that steered the evolution of search user interfaces, considerations that search user interfaces need to address, and specific guidelines for people who need to design search user interfaces. Drawing upon a large body of previous research, the author argues for simple, uncomplicated interfaces accessible to a large demographic of users.

Unlike most of the other papers that we have read in this class, this one does not seem to present new research results, or even be aimed at other researchers; it serves more as a how-to guide for search engine creators, summarizing previous developments and synthesizing other research results in order to provide concrete suggestions. In that sense, the value of this article in itself as a resource for other researchers seems light, but surely, the lessons that it contains are valuable to those building any practical search user interface. Overall, I found the arguments sound and well-supported, as well as intuitive in their applicability. I especially appreciated the subsections regarding small details and aesthetics in design. While the other guidelines tend to be exposed as specific features by existing search engines, the importance of small details and aesthetics would be much harder to learn from examining search engines alone, exemplified by the lack of pictures in that section.

As an overview of how to design search user interfaces that draws from a wider body of existing research, this article seems to have little to find fault. Nevertheless, while the author weaves a narrative from the previous research papers, I thought that this article didn't seem to amount to much more than a sequence of summaries for them. Also, while the author supports all of her arguments with previous research results, her presentation leaves little room for discussing previous approaches that have perhaps now been discredited; although very polished, she reveals little about how thoughts about search user interfaces changed over time. I thought the author could have perhaps presented a stronger voice of her own while discussing the previous research results and their relation to each other.


Krishna - 11/21/2010 17:49:40

Search User Interfaces - Marti Hearst

The author provides a detailed overview of search user interfaces: their history, challenges and design considerations. The chapter starts off with discussing the primary challenges in designing search UIs. They have to be simple and yet generic to be applicable to a wide user base and to further complicate things, most search users have primitive knowledge of how search works and do not understand the concept of incremental search - an implicit assumption made by most search engines. The author gives a nice overview of the search UI history, an important take away is that historic search systems operated on top of controlled information and well designed vocabularies and assumed users with explicit training and expertise, all these criteria are no longer valid and in fact, as the author mentions, it is the exact alternate situation that current search technologies have to deal with: uncontrolled information base and users with primitive skill sets.

The author's suggested design considerations for search UIs can be summarized as follows. Search UIs should offer feedback to the users - example feedback can be dynamic query term suggestions, document summaries with term highlighting. Search engines should balance between hiding certain actions and offering explicit user control over others - for instance while it is useful for search engines to correct spelling mistakes and expand abbreviated terms, it is also necessary for the user interface to make these changes transparent to the user and allow the user to make changes if necessary. An important criteria, that is a recurring theme in this course, the author mentions is that search UIs should help reduce memory overload and reduce the cognitive efforts required by the user - this can be done in the form of saving frequent searches or using some kind of a categorization system and show users how the underlying information is organized. Other design considerations include providing intelligent shortcuts - as in the case of direct links to within site searches, handling user errors by correcting trivial errors or by suggesting underlying facets, interpreting user intent - as in the case of providing information on time and weather when appropriate search query is given. Finally, the author stresses on the importance of aesthetics and gives the example of how Google's search UI team focuses on small things such as dividing the search result pages and respective ads using lines instead of boxes and how they found such, what might many consider as trivial, things to be effective.

The author gives a scholarly overview of search UI history and design considerations. Although, most of the design suggestions sound simple, implementing them is hard and in most cases search engines deal with enormous amount of uncertainty. Evaluating design decisions, such as query term suggestion and ranking methodology, is even more hard given the wide user and information base. It seems that the author has focused this chapter towards text search engines, it would have been interesting if the author had discussed design considerations for multimedia search engines and how some of the suggested design criteria might vary - for example, what would document summaries mean in the case of video search ? and some introductory discussion on the social aspects of search and the subsequent privacy issues would have been interesting.

Information Foraging Theory - Pirolli

The authors discuss information foraging theory - they consider humans as informavores and compare the act of humans finding information in a complex information setup to that of animals finding and competing for food in a complex ecology. They compare information clues such as hyperlinks and summaries to food scent used by animals when they look out for food. They cast the whole information retrieval problem as an optimization problem, they draw ideas and concepts from mathematical models that describe food foraging behaviors in animals and apply it towards building a computer simulated user that finds documents. Their ideas are based on the fact that there are costs associated with searching for information, choosing and handling information and effective users find some optimum which reduces these costs and maximizes the benefits.

They describe building a simulated user that finds information using their scatter/gather system. The scatter/gather system is essentially a hierarchical clustering system, initially the user is presented with a set of clusters based on his query. Each cluster is represented as a bag of words commonly occurring the documents present within the cluster. A typical user uses the system by selecting clusters and by recursively clustering documents in the selected clusters until he finds the document or the information he is looking for. The authors associate costs with each possible action in using the system - selecting a cluster, looking at document titles within the cluster, etc. They come up with a set of production rules and use these associated costs to come up with an optimal model that selects documents based on an information need. The model selects clusters to view documents based on how much the words associated with the cluster is correlated with the words associated with information need - based on prior knowledge.Given the setup, each cluster was given some cost based on how relevant it is to the information need and as mentioned before, selecting each cluster and reading the titles in it has some associated cost. The authors then show that there is an optimal number of clusters that users should select and this would correspond to the maximum possible gain for that set of documents and the information need. They perform similar analysis and estimate optimal criteria for their other production rules.

The paper is interesting as it describes alternate strategies towards building information retrieval systems. However, it is not clear what the exact "take aways" are. I interpret their results as one that suggests designing search systems such that relevant and precise information scents are provided to the users - I guess this relates to the emphasis Marti Hearst made on feedback. As they themselves mention earlier in the paper, information is different from food and most the cognitive models that explain food foraging theories should be revised and adapted.





Anand Kulkarni - 11/21/2010 18:33:59

THE DESIGN OF SEARCH USER INTERFACES

The author summarizes some design guidelines for search engines based on what has succeeded in the past several years.

The main contribution of this article is a survey and analysis of successes in search engine design over the past several years. Web search remains an important and financially valuable problem and an area in which it is often difficult to separate core innovations in technology (ie, the challenge of finding relevant content) from issues in design that ultimately speed the process of content discovery. While several of the issues discussed are obvious upon reading, it's useful to have them compiled in one place; many early web designers seemed to miss these features. I like the idea of decreasing memory loads for users; in particular, the idea of integrating search and navigation seems to be an essential aspect of search designs. I also appreciate the emphasis on aesthetics, although as the author observes styles are rather homogenous between search engines today.


The document provides a survey of results as its primary means of evaluation; it does not provide original results. A survey is appropriate here. I don't agree with some of the suggestions and would have liked to see more comprehensive results; for example, the idea of providing graphical search indicators for quality (such as stars) seems to conflict with the idea of maintaining aesthetics, and there is a quantitative tradeoff here that's being ignored. I wish also that the author had mentioned possible solutions to the vocabulary problem rather than just emphasizing its difficulty; this is a central issue in the design of search engines and it would be helpful to see some possible solutions.


Siamak Faridani - 11/21/2010 18:35:42

The Design of Search User Interfaces, Marti Hearst, 2010. Chapter 1 from Search User Interfaces

The article looks at elements and best practices in designing interfaces for search. The article points out challenges in designing UIs for search and by looking at some of the existing examples and patten points out recommendations for designing future UIs. Author starts by highlighting the fact that the interfaces has been almost identical over the course of many years. For example she looks at the search interfaces from InfoSeek and Google ten years apart and shows that they both use a text input box and show the results in a long linear list. I personally enjoyed the fact that she looks at search as an intermediate step toward a final goal. As a result, she concludes, search should be as simple as possible not the distract the user from her final goal.

Additionally the author cites many interesting user studies that show that many users still struggle with using current search engines as a result the author suggests that any effort to make the UI more sophisticated will fail due to inability of users in adopting to the new interface. Although I found this statement rather confusing “Novice searchers must learn to expect that a query will not yield immediately usable results, and that they must scan search results lists, navigate through Web sites and read through Web pages to try to find the information they seek.” my question is why? Is it not contradictory to the fact that we do not want to make users think more and we want the search UI to be transparent?

I am wondering though, if we assume that the whole point of search is about helping user work toward another goal, and find other resources. Is it not going to be contradicting with the fact that search engines need to monetize on search results? Every click on an ad distracts the user from her final goal and takes her focus out of her goal. Then how can a search engine be as transparent as possible and at the same time make money?

Hears continues by adding that web completely changed the way we search for information. Before search was only available through libraries or by professionals who had access to high quality materials that were available only to a few but we are now able to search trough millions of (sometimes low quality) documents, photos, videos and metadata. Searching by keywords is now completely meaningless. Keywords that are not related to the text are now completely being ignored by intelligent algorithms. And we are finding ways to index other forms of data in addition to text. Audio, video, and other forms of information are now intelligently being indexed by our systems, thus we need more efficient interfaces to query these indices. People now want to ask questions in natural languages and get results that are mash ups of text, graphs and other forms of media. Modern UIs for search should be able to satisfy this type of demands from users.

Author claims that the most important aspect of a search UI is usability (although she does not provide any alternative, I mean why search quality is not considered an important element here) Nielson defines usability as a combination of learnability, efficiency, memorability, errors and satisfaction. Hearst highlights the importance of user centric design methodologies in which a UI is deployed and is tested by collecting predefined metrics information from users. In the user centric models designers assume that they do not know what users really need and they need to determine that by testing, collecting information and iterating on the design.

Shneiderman (the author of one of the earlier papers on creativity support tools) provides a number of guidelines for designing UIs for search. While it was interesting to see all these items together I found the list to be a compilation of intuitive elements that every designer already knows. Hears goes on to highlight different ways to give feedback and enhance the interaction with the user, for example enabling users to perform sort on results based on different criteria. show suggestions, highlight search terms in the results, and finally support rapid response. I found the last one interesting since it was recently that Google added this element to their UI by providing Google instant.

Hears also summarizes the intelligent behaviours of a search UI, for example she points out that a search UI should be able to intelligently modify user’s query if we are confident that the query should be different . This intelligence and automation provides more flexibility and allows users to focus on what they want instead of how they should formulate their query.

Reducing short term memory load is another element that consists of providing suggestions in the search box, history logs, and providing tools to navigate and filter results. The author provides a very short list of elements for reducing errors.

This chapter is a wonderful intro to designing UI for search but I didn’t find new or interesting methods in the chapter. It is more like a survey of existing models and does not provide any insight into UIs of the future.

Author also assumes that we are only focused on textual information, she does not give any help in how we can highlight search terms in a video or audio? How can we help users to navigate through multimedia info? Additionally she does not make any comments about personalization and customization in search. What about searching personal information? I am wondering if at any point our search engines can answer queries like “where should I take my friend , Mark Alen, for his birthday and what activities interest him?”



Luke Segars - 11/21/2010 18:36:19

Exploring and Finding Information

This paper centered around the idea of "information-foraging theory." This theory relates the search for information to the 'survival of the fittest' idea from evolutionary biology, stating that individuals who have better survival qualities (ability to find information) will also have a higher chance of reproduction (chance of success at finding the information and completing their search task).

The idea of information-foraging theory seems to be somewhat accurate intuitively, although I'm not sure that this idea drives human life as much as the term 'informavore' suggests it does. We certainly have a need and dependence on certain information, such as the presence of immediate risk or the location of food and water. Many other types of information that have become readily accessible with the recent digital movement are not necessarily obviously important (or even desirable) for all people. Nevertheless, when we are in need of more specific information (email, weather, encyclopedia information, how-to's), I think that the information-foraging theory applies well. The paper mentions that we often want to understand *why* something works the way it does. While I think this is true for many people (engineers in particular), the vast majority of people settle for knowing that something does work, as opposed to wanting to dig deeper to understand why it does. The abstraction of complicated engineering designs is what makes technological progress possible.

Despite the intriguing presentation of the information-foraging theory, the Scatter/Gather model doesn't seem to represent my approach to information finding. It is an interesting idea but I'm not convinced that it's a common approach for many people, meaning that the simulation wouldn't be particularly useful. One particular type of activity that the scatter/gather model seems applicable to is creative idea generation (brainstorming, designing something from scratch, etc). The paper didn't actually give me a solid understanding of what the simulations could be useful for, which may stem from my lack of enthusiasm for the scatter/gather model as an information-gathering technique. The author discusses information scents which, while important, seem to offer more of a keyword trigger than any sort of 'scattering' effect.


Drew Fisher - 11/21/2010 18:46:35

The Design of Search User Interfaces

Having grown up in the age of the Web, I have a hard time rationalizing the recommendations given - they just seem so mindbogglingly intuitive in hindsight that I can't fathom an interface that would try doing anything else. As a result, to me, this book takes the apologist's approach to user interfaces. Surely, however, in their development, these rules were not quite so clear.

I wish that section 1.5.2 had discussed the likely reason for the benefit of showing search terms in document preview snippets: if the user is searching for a particular page she's seen before, she can positively identify it more easily, since she likely remembers the context in which the query terms appeared in the document. If she's searching for a new page about a topic, she can see how that page relates to her train of thought, which makes analyzing a result's value invoke less mental load. Either way, it's a win.

It seems as one underlying idea presented by all of this chapter's design principles is "don't interrupt the user's flow" (quick response, lower mental load, etc.). Another would be "do what the user meant, not what he said" (reduce errors, spelling correction, shortcuts, context inference).

I think search is an interesting field in that most of the user specified terms are either awfully underspecified or excessively overspecified. Most people I know outside of technology-related fields seem to not have developed a "feel" for how much information to put into search query terms. I wonder if this user learning is considered (or worth considering) in the design at a level deeper than "offer features to expert users."


Shaon Barman - 11/21/2010 18:50:23

The Design of Search User Interfaces

This chapter discusses the evolution of search, from a highly specialized languages used to query a structured set of entries to web search interfaces which use natural language queries to find web pages.

The authors point out that searching is a difficult tasks, which is complicated by the fact that users will have different levels of expertise. Some users will want to have full control of their query while others will not have a basic understanding of how the interface works. Difficulties arise in accommodating both types of people and still maintaining accurate results. The authors provide a set of 8 design guidelines that should be in a good user interface. Accommodating all 8 is difficult, since some guidelines conflict with each other.

As with most criteria lists, they are incomplete and seem a little ad hoc. But this list seems to highlight the major issues with search interfaces. One part of the paper I like was the use of real-world examples to highlight the points. Showing specific cases of bad design decisions are quite informative (such as the changes required to make the spelling suggestions interface useful). One aspect that I would like to see is how advertisement fits in the user interface design. Because search is usually a free service, advertising plays an important part in making the product successful. Advertising also greatly affects the user interface and how a search interface embeds ads can have a drastic affect in both how users feel about the system, and how well the ads work.


kenzan boo - 11/21/2010 18:51:10

The Design of Search User Interfaces, Marti Hearst


The Article is on designing a user interface for search. It points out how the UI for search has not changed in over 10 years from old search UIs to what google currently does. The main guidelines for making a good UI are outlined as: Offer informative feedback. Support user control. Reduce short-term memory load. Provide shortcuts for skilled users. Reduce errors; offer simple error handling. Strive for consistency. Permit easy reversal of actions. Design for closure. One of the main things is to offer constant feedback to the user like highlighting the search terms.

There is a lot of content here that points out the various key points in search engines. The one I found particularly true was balancing user control and automated actions. There is a lot information out there and for most users, it is very difficult to exactly say what you want. Its more useful for the search engine to determine what is best for you. Rank ordering is the key point that does that. Also reducing short term memory is very important. Its good to have a history of what you searched and what sites you clicked. Something like bookmarking searches would be great for reducing memory load.


Matthew Can - 11/21/2010 18:52:22

The Design of Search User Interfaces

In this book chapter, Hearst provides an overview of the history and process of search user interfaces, followed by detailed descriptions of several design guidelines for search interfaces.

The chapter divides the history of search interfaces in two periods: before and after the Web. This is because the Web radically changed information retrieval. For example, before the web search was costly and only performed over specialized, high quality content, whereas now, search is free and is done over a wide range of quality and content. Given that search today is so different, it presents new challenges for search interface design, so it is worth thinking about this historical shift.

Perhaps the most important design principle in this chapter was that search interfaces should provide immediate, informative feedback to the user. It goes without saying that the best search interfaces show the search results immediately after the initial query. In fact, Google takes this a step further now and displays results as the user enters the query. Also, many sites provide feedback in the form of incremental query suggestions as the user inputs the query (Bing, Google, Facebook, Yelp). It is interesting to think of the tradeoffs involved in creating effective document surrogates. For example, an interface that highlights query words helps users understand why a document is relevant and how their query fits in context. But, highlighting too many words is overkill. Likewise, though it is useful to show query words in the search result snippet, showing too many can result in incoherent sentence fragments. It’s worth thinking about how interfaces can overcome these limitations. One suggestion is that the interface can reveal additional context around the query when the user mouses over the document surrogate. Google currently tries something like this. They show a full page preview of the site, but it doesn’t work well because it is impossible to read the text in the preview, and they don’t highlight additional context.

The chapter also discusses shortcuts as a design principle in search interfaces. While there is no doubt that shortcuts can improve the search experience, the chapter should have noted that they are critically important to mobile search interfaces. Mobile search has a higher cost per query than desktop search because it takes more effort to input the query and takes longer to receive the results (this will change in the future with higher mobile bandwidth). So, it is immensely important for the search interface to minimize the number of actions the user must take.


Matthew Chan - 11/21/2010 18:54:13

The Design of Search User Interfaces

In this paper, the authors highlight the user interface design for search and how it has remained constant for the last two decades. The paper also explores the history of search from librarians and clerks searching via titles to the common user who can type keywords today. One of the unique aspects of the paper is the frequently cited references to other research in the field, such as how novice users tend to type out questions like "What is the population of Madagascar" and how the upper left corner is the hot/sweet spot.

When it comes to the process of designing search interfaces, the 5 main components are

  • Learnability
  • Efficiency
  • Memorability
  • Errors
  • Satisfaction

Still, these are difficult to design highly usable interfaces since some might conflict with each other. Next, there are 8 design criteria for a successful design as well (whcih I own't list). One of the things i liked reading most was about error handling and quick recoveries for the users, such as offering other search key words while they are still typing, suggesting other related search, spelling correction, and reducing short term memory (as shown from NYT search). For the short term memory load, Google and Safari address this in their browsers by displaying the recently accessed webpages or frequently accessed webpages.

The results of the evaluation is enlightening to me since I never really read about search UI. Everything I read sounds like a description of Google and Bing. It also reminded me of a "categorical-visual" search engine called SearchMe.com (which is no longer around) and it displayed results like a coverflow and the image of the website with keywords highlighted. However, there were very few surrogate information. Now, i notice that GOogle is displaying a visual counterpart to all the search result.

This paper relates very much to technologies today because Google and Bing are now trying to display more recent results (ie. twitter posts), utilizing the "instant" features, and are quick for error handling. Reading about search also reminded me of Berkeley's library search engine and some of the perks (good or bad) that i've experienced; this paper also made me wonder about Facebook's search for friends and how they address. However, this paper does not relate to my work.


Thomas Schluchter - 11/21/2010 18:56:33

=== Search User Interfaces ===

Hearst's book chapter outlines some design principles for Search UIs that integrate with general usability guidelines but expand significantly on details specific to the task of retrieving information from an unknown collection.

What I found most interesting about the chapter was how specific the design recommendations become when one focuses on one task or domain. In my experience, usability heuristics and other guidelines that aid designers in their decision-making process are vague and very high-level. By focusing on a particular task, the recommendations become much more actionable and measurable. It would be interesting to see whether for other domains or tasks similarly well developed recommendations exist, and whether the existence of such recommendations maps to the relative importance of the task or the domain.

Search, with its contrast between complex computational problem and simple human-interaction cycle (query, present, evaluate) illustrates why the study of user interfaces is such an interesting field. The interface is to hide the computational complexity from the user without creating an unintelligible, black-box style feedback. The problem that makes this hard is that users are, from the viewpoint of the system, fundamentally incompetent searchers. According to the garbage-in-garbage-out principle, search systems must work around these deficiencies and infer in all kinds of ways what users mean. (The large-scale data collection of web search engines has done more to actually understand these deficiencies than years of AI research.) Techniques for nudging the user in the right direction when a query was faulty without disrupting the user experience is therefore one of the most important features of search UIs.

The topic of search UIs is highly relevant as search has become the default answer to the question how to deal with the massive amounts of information that current systems generate. As the evolution of Google's search UI shows, there is a trend towards ever shorter feedback cycles between query formulation and results presentation. Google Instant is the latest step, showing the results as you type. Since many of the search interfaces we use access network-based collections, it is important to assist the user as much as possible in query formulation to reduce unnecessary network traffic.


Airi Lampinen - 11/21/2010 18:57:04

The first chapter from the Search User Interfaces book is a broad introduction to the ideas and practices of user interface design in general, and to search interface design in particular. The text discusses basic usability concepts and guidelines and shows how they have been and could be applied to search user interfaces.

The provided set of design guidelines that concerns specifically search user interfaces. These guidelines contains items such as balancing user control with automated actions, providing shortcuts, reducing memory load and offering appropriate feedback. The discussion on automation and user control was especially interesting, as this trade-off is central also in thinking about social network services and the like. While allowing computers and algorithms to bear some of the workload, it is important to allow users enough control over what happens as well as to make sure that the automated solutions don't make mistakes that are hard to accept or difficult to recover from.

Overall, I found the text a useful recap of usability basics and a nice summary to the small but important features that make search user interfaces as good as they are today - something I (surely along millions of other computer users) am usually taking for granted without giving any further thought to how many design decisions have been made to allow for the effortless user experience.


Arpad Kovacs - 11/21/2010 18:58:40

The Search User Interfaces chapter analyzes the requirements of a usable search engine interface, and explores how these requirements evolved during the emergence of the web and thus converged into the canonical results-in-vertical list form that is so widespread today. Before the Web, search was specialized and restricted to highly educated users, who assembled queries using boolean operators via a command-line interface to search over document metadata, often on a pay-per-search basis. In contrast, today's Internet-using population is interested in performing full-text search of content using graphical interfaces, and often lack the training necessary to understand boolean operators and syntax, which makes the usability of search interfaces more important than ever before.

The most important contribution of this paper were the design guidelines for search interfaces listed in section 1.4, and elaborated on in sections 1.5-1.9. To provide for satisfaction and error-recovery, it is essential to offer the user efficient and informative feedback. For example, the interface should show search results immediately and highlight query terms in the context of their containing documents, in order to letting users know whether they are on track. Dynamic sorting and query term suggestions are also helpful in helping users adjust their search parameters on the fly or specify alternative wordings, and thus improve usability through efficiency. Small details, such as increasing the length of the query terms input box and aesthetics can also subconsciously affect the user's efficiency while using the interface and ultimate satisfaction. Shortcuts improve learnability, efficiency, and memorability by attempting to predict user intent, and leading more directly to relevant pages. Finally, features such as web history, integrated navigation and categorization systems, and query refinement options offload short-term memory and thus reduce errors and improve satisfaction.

The chapter also highlights important tradeoffs between automation and transparency/control provided to the user. Since most modern users do not understand the inner workings of search engines, automation is generally preferred; however the user should still have a manual option eg for rank ordering by chronology. Likewise, many search engines autocorrect spelling mistakes, as well as modify or ignore query terms to increase the number of hits; however this may distort the meaning of the query. It is also difficult to call attention to relevant data while minimizing distractions/interruptions to the user's thought flow.

I find that most modern search engine interfaces already implement the ideas mentioned in this chapter, and thus provide for an extremely efficient search experience, while catering to a wide variety of users. In particular, I am a huge fan of zero-click information such as Google Calculator and Duck Duck Go's embedded Wikipedia summaries, which make the search engine itself a source of relevant knowledge, rather than merely a layer of indirection leading to a website that may or may not contain relevant information. Suggested keywords/related queries are also an incredibly useful feature for refining queries, and also as a social metric to see what other people are searching for. However, I feel that the emphasis on instant everything has reached the point of diminishing returns, and actually changing the search result listings on every key stroke is distracting and disorienting. Instead, I would prefer a greater emphasis on configurability for advanced users, eg customizing the size of the snippets shown and the sensitivity thresholds for including/discarding given keywords. Today it is very easy to find _a_ reasonably relevant result for a query, but finding the optimum link for a very specific search criteria (especially if it is a less popular page) or improving your search results still remains a challenge.

It amazing how much information the canonical results-listed-in-order-of-relevance interface can show while maintaining a relatively clean appearance. The fact that a user can even gauge whether a page is relevant or not based on just a title, 2 lines of snippets, and a url are quite major accomplishments. Finally, I am continuously impressed by how robust query parsers have become, and can handle inputs that even some humans would not understand. In summary, I find modern search engines to be one of the most refined and successful user interfaces in existence, as the millions of Internet users running billions of queries every year can attest.


Aditi Muralidharan - 11/21/2010 19:42:31

In "Search User Interfaces" Ch. 1 Hearst gives an introduction to the main challenges and design principles in the space of search user interfaces. As far as principles go, she gives the following:

   * Offer feedback,
   * Allow the user to have control
   * Reduce short-term memory load (history, backtracking)
   * Make it easy for people to scan results quickly

And stresses that small details and interface aesthetics can have an impact on how people percieve their search experience.

It was incredible to me that people have trouble understanding keyword search, that the amount of white space and font size can have an impact on how people type their queries and understand their results, and that the words in which different people search for the same things are so numerous.

I design search user interfaces, so this book has been extremely instructive for me. I think this chapter a great job summarizing the main ideas around search user interfaces, but more than that, the list of references to previous work was very thorough and detailed. A very useful starting point for a research project. I can't really think of anything bad to say about it.


Pablo Paredes - 11/21/2010 19:56:03

Summary for Pirolli, P. - Exploring and Finding Information

The paper presents an interesting analysis of information searching based on the theories of human cognition/evolution and foraging. The notion that living creatures strike a balance between the cost of transportation to get food and the expected benefit of this food intake is used to associate the description of a human being as an "informavore", an individual hungry for information about the world. The key anchor point of the analysis if the definition of adaptionist approaches, which define the need to solve problems and the paths followed to do so, and how our cognitive and perceptual systems adapt to these tasks.

I believe the notions of information scent and infromation foraging derived from the analysis by the author are great paradigms of analysis for information searching. Information scent, defined as the clues to judge distal information sources and the perceived effort needed to get there propose that the real design problem is not so much how to collect information, but rather how to increase the amount of relevant information encountered (i.e. following a scent path). Studies describe people's preference on designs that support this information foraging paradigm.

Although I believe the scatter-gather model is only one model of search, and currently not the most used, the analysis derived that shows that in the information search task, diminishing results are obtained rather quickly, clearly shows the notion that the tradeoff of perfectionism versus good-enough information is highly important. The results show that clustering, rather than reading the documents is the best choice for the initial few (2 to 3) relevant information documents, however past this mark, incorporating documents with less relevancy provides diminishing results (compared to the rate of gain). Also, another result shows that re-scattering initially presents higher rewards, however quickly (after about 400 seconds) of search, it is better to begin displaying documents. These two results define, in my judgement, information scent. People that is hungry, may have cravings for the ideal food, however, they will follow the path of the strongest scent that approximates to the type of food that is being searched and will settle with this less-than-optimal food, for the sake of balancing the cost of reaching the ideal food, that could demand exaggerated expense of energy to evaluate all the options available (all the scents) to pick the right one.

It is interesting (yet expected based on the information scent theory) to note that interface and clustering algorithms represent opposing gains when time is introduced as a constraint. For soft tasks (unlimited time), clustering presents higher rewards, however, for hard tasks (limited time), interface actually represents a higher benefit. My interpretation is that in limited time tasks, we will be much effective following an information scent by using tools that quickly track (visualize) this scent path, rather than expending time in analyzing the options to pick a scent path. On the opposite, with more time, we can gather better information by analyzing the presented information better. It is interesting to see that in the case of website navigation this notion is expressed by the relative few number of clicks that a web-page receives... apparently, a great number of the people "navigating" to web pages, is doing it mainly following a scent, rather than reaching a destination. The author argues that this "greedy" use of perceivably unlimited results could be the explanation of Internet congestion peaks, and the "winner-takes-all" characteristic of website navigation.

I believe the notion of information scent and the behaviors observed under soft and hard tasks are valuable to describe the paradigm of information search in reach-content media. It would be very interesting to observe if the same paradigm is also observed in mobile mediums, where there is already an embedded perception of scarcity of resources and higher cost. We may need to propose other search paradigms, where the convergence time to a result is shortened by improving clustering, maybe using context awareness (location, time, activity), to help users make better decisions upfront, and therefore improve the relevancy of the information retrieved.