Search

From CS260 Fall 2011
Jump to: navigation, search

Bjoern's Slides

Media:cs260-22-search.pdf

Extra Materials

Discussant's Materials

File:Jason-toy-discussant-materials.zip

Reading Responses

Valkyrie Savage - 10/31/2011 19:24:31

Main point:

Search is a hard problem. You can get the interface wrong, you can get the context wrong, you can get the result set wrong, you can get the ranking wrong...

Reactions:

In general, I think of search as a pretty interesting topic. I’ve spent something like a year over a couple different jobs working on different kinds of search interfaces. My fianceé does search for Facebook. My roommate does search for Yelp. We’re a very search-y household. Obviously, it’s one of the big problems on the web.

The UI paper was cool. I remember working at Google when they performed their infamous “42 shades of blue” study. Heck, they gave away t-shirts that showed the heatmaps of gaze-tracking on their logo. Interface is important! I liked the discussion of reducing the user’s memory load, since I just got done reading the CodeBubble paper which also talks about doing that. I wonder how computers are working to reduce users’ memory loads in general? I have a sneaking suspicion that fact memorization is headed the way of the Dodo, in that we’re unlikely to need anything more than Google+Wikipedia to get answers to things, so why should we bother testing students’ memorization? Similarly for photos: I can’t remember lots of stuff I’ve done, but “a picture is worth a thousand words”, as they say.

The discussion in this paper on ranking and other things that it’s ok to make opaque to the user was a bit frustrating. Mostly what it said was “don’t bother showing them what’s happening, as long as what’s happening is right.” This is a problem for frustrated users who are having difficulty finding information relevant to their query. Perhaps it should be invisible to the quickly clicking-through user, but it might be important for power-users to diagnose what’s going wrong.

So, Vark. I have used Vark some, and no longer do. Their goal is noble, but they slipped on execution. I found that their social graph was somehow off, and that their topic gleaning from questions was poor. I was frequently asked questions by people whom it was never explained to me that I knew, the questions were often not in understandable English, and the questions were often tagged incorrectly.

That said, the village paradigm is an important one. As the web gets bigger and bigger and less and less personal, it’s important for people to have an outlet to still interact with other people. It’s a big, lonely Internet out there, and it’s nice to have a “face-to-face” conversation about something you’re curious about or need information about. Related to that, I found it interesting that “chattiness” was one of the indexed features for a person. I would not have thought of that, but it’s a great idea!


Yun Jin - 11/19/2011 18:32:30

This chapter of the user interface has introduced the ideas and practices surrounding user interface design in general, and search interface design in particular. It has explained some of the difficulties with search interface design and provided a set of design guidelines tailored specifically to search user interfaces. These guidelines includes offer efficient and informative feedback, balance user control with automated actions, reduce short-term memory load, provide shortcuts, reduce errors, recognize the importance of small details, and recognize the importance of aesthetics. This chapter has also summarized some of the most successful design ideas that are commonly in use in search interfaces today. This summary is based on generalizing over the results of years of research, experimentation, and tests in the marketplace. The coming years should reveal additional new, exciting ideas that will become reliable standards for search user interfaces. The second paper presents Aardvark, a social search engine. With Aardvark, users ask a question, either by instant message, email, web input, text message, or voice. Aardvark then routes the question to the person in the user’s extended social network most likely to be able to answer that question. As compared to a traditional web search engine, where the challenge lies in finding the right document to satisfy a user’s information need, the challenge in a social search engine like Aardvark lies in finding the right person to satisfy a user’s information need. Further, while trust in a traditional search engine is based on authority, in a social search engine like Aardvark, trust is based on intimacy. The paper also describes how these considerations inform the architecture, algorithms, and user interface of Aardvark, and how they are reflected in the behavior of Aardvark users.


Hanzhong (Ayden) Ye - 11/20/2011 15:23:11

The first article for today’s topic of searching user interface is a very good introduction of several general principles and rules widely applied in searching user interface design. Besides the most important rule of keeping the interface simple, there are many other important principles that would make significant difference on interface usability. The article discusses the shift of searching UI design and the reason for such transition. Many useful tips are then given to guide a better design of searching UI. Some tips among these guidelines that I consider to be important are: offer efficient and informative feedback, balance user control with automated actions, provide shortcuts, etc. Many successful ideas in searching UI have been developed based on these important design principles.

The second paper is much more interesting to me because it introduces a tangible application based upon several credential and reasonable assumptions in social network. Their work named as Aardvark enables users to ask questions, either by instant message, email or other form, and get quick answers from their own social network. Compared to a traditional web search engine, this new type of social search engine focuses on the process of finding the right person to answer the questions. The design and implementation are described in details, and many reasonable mathematics models for modeling the social network and expertise are built. I consider the idea of searching answers from friends to be a smart one because friends are essentially the most important resources towards which one can seek answers for subjective questions and call for suggestions. The implementation and examples are very exciting to me because it showcases a very novel way to post inquiries and seek for quick answers and feedback. I consider this paper to be a very good example which leverage the enormous power of social network.


Hong Wu - 11/20/2011 16:17:52

Main idea:

The two papers discussed the principle to design the user interface for search purpose.

Detail implementation:

“Search User Interface” described the principle of designing UI for search. As search is the mean to complete another goal rather than the goal itself, the search should not disrupt the processing of finishing the primary goal. Based on this principle, the UI for search should be precise, informative and easy to use. It will be hard to design a perfect UI as the principles may conflict with each other.

Nowadays, the search engine like Google is popular but also has a lot to be improved. First, the key word selection is crucial to the result of search. Second, there are a lot duplicated search results because some websites refer or copy the content on other websites. Third, it is not convenient to use control key words “and”, “or” and “not”, which can be replaced by some symbols, such as “+” and “-“. Fourth, at the top of Google’s result page, it shows the number of the returned item and the search time. This is unnecessary information for user as a user may only browse 50 search results at most. It is an advertisement of Google.

“The Anatomy of a Large-Scale Social Search Engine” proposed a model which searched the person who may answer the question instead of search the answer of the question. My may concern of the model is that how long user needs to wait to get the answer. In another word, the model is not only based on whether it can find the right person but also on how the person involved in the community. A lot of forums focus on the small community such as programmer, which based on same interest or target instead of intimacy in this paper.


Peggy Chi - 11/20/2011 16:38:41

Search is such a common activity in our daily life. However, the search experience via computer interfaces is still somehow frustrating, compared to more direct ways such as asking a real person in daily life or posting to Q&A sites like Quora or Yahoo!Answer. Hearst gave an overview of the search user interfaces, and identify several guidelines including providing efficient feedback, reducing errors, etc. Horowitz and Kamvar designed a search engine that spreads out a user query to users' own social network by finding the "right person" to ask.

It is interesting when Hearst indicated that the search interfaces nowadays are almost identical to those in 1997 (in section 1.1). Although the reasons listed make sense, recent projects such as Wolfram Alpha, social Q&A sites, and conversational agent like Siri have gradually changed the way search engine should be. The paper talked lots about the design issues of search interfaces. However, rare of the new interaction styles were explored. For example, other than text inputs in keywords or natural language processing, searching by image (Google Image search) and audio (SoundHound) are also possible now. Iterative search such as narrowing down the search scope interactively is also not mentioned. Would the design guidelines still valid? Or are they too general?

The Aardvark paper demonstrated a good example of using social network. It is now already common to post questions on Facebook or other networking sites, but the authors built a system as an additional layer on top of existing services. The idea and the paper are very convincing, but I wonder if the system could successfully survive in a long term. I personally do not want to be bothered all the times, and the social pressure might be an issue. It's also worth noticing that the company was acquired by Google, though the service is closed: "Aardvark was a start-up we acquired in 2010. An experiment in a new kind of social search, it helped people answer each other’s questions. While Aardvark will be closing, we’ll continue to work on tools that enable people to connect and discover richer knowledge about the world." (http://googleblog.blogspot.com/2011/09/fall-spring-clean.html)


Alex Chung - 11/20/2011 18:15:58

The Design of Search User Interface by Marti Hearst summaries the best practices of search engine user interface design and provides insights for future generation on the important topics to consider. Its journalist style of relating research studies with real world examples is similar to Brad Myer’s paper on UI Toolkits and Alexander Quinn’s paper on Human Computation. The author provides nothing new in terms of technical advancement but a convenient one-stop shop for beginners who wanted to survey the study of search user interface.

I found the paper produced by Brad Myer more satisfying because he includes both good practices as well as samples that did not catch the public’s attention. It was a more comprehensive and education than this paper on Search UI Design. Having said that, I really enjoy this reading because it has lightened my memory load on design philosophy and sharpened up my vocabulary when I continue my UI design career in the industry.

However, I have trouble understanding the dilemma between designing for personal use versus publicly used software. While some might find the approach of Google search engine being too broad and impersonal, it is unassuming and easily adaptable by anyone. It’s inefficient to design one language for everybody when everyone’s is so different.

In contrast, the second paper about using social search engine to match questioners and answerers provides new direction of using the social network as a medium to connect people. The effort from this research contributes more to the human computation research of HCI.

The probability model in section 3.1 multiplies the probability of social connectedness and profile similarity. Then the author must assume that two variables are independent from one another. Yet this logic has ben proven incorrect in many social science studies that connectedness and profile similarity are highly correlated.

While I agree a public forum deters freedom of expression and participation, these factors are irrelevant because the goal of online Q&A is finding the best possible and unbiased suggestions. On the other hand, I really like the “I know who to ask” feature because the recommendation encourages people to answer the question. However, the response rate would drop off as the question is being passed further away from the questioner’s origin. Did the author measure the response rate in relation to the out-degree? I didn’t see it.

“The desire to have a fellow human being understand what you are looking for and respond in a personalized manner in real time is one of the main reasons why social search is an appealing mechanism for information retrieval.” What about a computer that understands and responds like a human with a personal touch? For example, iPhone users ask Siri questions using natural language and the app responds in playful manner like a human being. And who knows you better than yourself?


Laura Devendorf - 11/20/2011 20:06:01

Hearst's Chapter provided a detailed list of guidelines to consider when developing search user interfaces. Aardvark is a social search engine that searches across friends and acquaintances in place of documents.

With the ubiquity of Google, I had almost forgotten that search interfaces need be designed. It is interesting to consider how the Google design choices have permeated other forms of search simple because the users are so accustomed to it. One interesting point that Hearst makes is that users don't understand how the pages are ranked however, if Google were to reveal this it could have opened the market to competition. It would also be pretty hard for most people to understand due to it's complexity. I also thought it was interesting, considering Google today, that Artificial Intelligence and Machine learning were not discussed in relation to searching.

While the structure, implementation and evaluation seem novel and sound, I'm interested in thinking of how the social life of Aardvark. In Facebook, there are always those people that update their status about nearly everything - would the same happen in Aardvark - would you always be getting requests to answer questions? The Trivial Question Answerer could be helpful in this case but I would still be interested in seeing how it functioned in a day to day setting and how it would effect relationships between people. If you have one friend who is a doctor, are they now going to have to field all of your questions about whether or not you are sick? If I put I'm good with computers, I am inevitably going to get questions about how to trouble shoot someone's printer. Sure, I could "pass" but then I feel like a jerk. Put more plainly, what intensive are there for the answerers? Many of my issues with this concept stem from my own personal preferences. If I don't know someone well enough to know what they are good or not good at, in most cases, I would be uncomfortable asking questions about it. Also, having questions has always been an excuse for me to call an old friend and have a catch up conversation. I like having that excuse and I wouldn't see myself using Aardvark.


Viraj Kulkarni - 11/20/2011 21:36:07

Design of search user interfaces' is all about how much design choices influence the usability of the search application. The authors also lay down guidelines that can be followed when designing search interfaces. Reading this chapter helped me realize the amount of thought and effort that goes into creating a seemingly simple user interface. The article mentions but does not talk at length about the statement that many people who use search engines do not understand boolean syntax. Does this mean that natural language search interfaces would do a better job? From my intuition, I feel that some sort of syntax would always give better and more relevant results than asking natural language questions. If that is true, would training users to formulate search queries be a better option than designing a search interface that works with natural language (the user adopts to search syntax than the UI adopting to natural language)?

'The Anatomy of a Large-Scale, Social Search Engine' describes Aardvark, a social search engine. This is a search engine that directs the user's queries to other users of the system who are 'experts' in that field. The authors describe the working of the search engine in this paper. I do agree with the fact that, at times, you want answers from real people who you know rather than an automated search bot. But, considering the vast array of information on the web, I wonder how often would I need to use a 'search engine' like Aardvark. Don't forums and general non-search social networking suffice? On one side of Aarvdark is the space of anonymous answers from people I don't know that I can get through forums and question websites (like ask.com). On the other side, for more personal recommendations or answers, I can contact someone I know and ask them. Aardvark sits between these two ends and I feel that there is very little space there. I would either ask my question on a web forum or personally contact someone I trust.


Sally Ahn - 11/20/2011 23:27:44

The first chapter of Marti Hearst's Search User Interfaces discusses the factors that contribute to the relatively simple and unchanging layout of search interface and provides an overview of its history and design guidelines. Horowitz and Kamvar's paper presents Aardvark, a social search engine that seeks to find the right person who may provide an answer to the user's query, rather than retrieving existing documents directly from the web.

Hearst makes an interesting observation that search interface has remained fairly constant over the past decade. Although this may seem strange given the vast difference of the web today compared to the web in the 90's, the reasons she cite for the permanence of the simple interface make sense; users engage in search as an intermediate step of gathering some information for a larger task, and so the interface must be generalizable to a huge variety of users and their backgrounds. However, even within the bounds of the seemingly simple keywords-to-results-list interface, there are a number of design elements that can aid or detract from the user's search experience, as Hearst reveals throughout this chapter. Perhaps it is due to these bounds that subtle design choices (e.g. length of the entry form) and aesthetics seem to make such a huge difference in user experience.

While these factors address the visual aspects of the interface, some other challenges (e.g. query transformation, the vocabulary problem) address the need to abstract linguistic complexities from the user, which seems to be a much more challenging problem. With the widespread usage of mobile devices, I think it would be worthwhile to investigate how these design guidelines might transfer to a mobile platform. It's possible that rather than simply adapting these guidelines to the smaller screen space and location-tracking capability of mobile devices, an entirely new search interface paradigm may be in order (e.g. voice commands).

The motivation behind Horowitze and Kamvar's Aardvark wasn't entirely convincing to me. They compare the "village paradigm" of their social search engine to a library paradigm, which I think misrepresents the real alternative search engine (especially since this paper was written as late as 2010), which can query the world wide web, not "the knowledge base…created by a small number of content publishers." I am skeptical about several assumptions that underly Aardvark's statistcal model. In particular, I have doubts that the quality of the answer being defined as a function of social connectedness and profile similarity, especially since the factors for the former (e.g. IM shortcuts and use of "Thanks!") seem arbitrarily chosen (if they weren't I wish the paper described the reasoning behind these choices).

Morever, much of the queries passed through Google often return responses from real users through similar platforms like Yahoo! Answers. Although the authors compare the answering time with Yahoo! Answers, claiming that Aardvark produces answers faster (by a number of minutes), I don't really think a scenario would rise often when a user would benefit from receiving an answer several minutes faster; I think it might be easier for the user to post a question then check back for answers later in the day. As the authors point out, many questions asked in a social platform are of a subjective nature, and in such cases, the user is often interested in a collection of opinions, from which he might draw his own conclusions, and existing social platforms seem sufficient for this. I did find the authors' footnote on "social cost" interesting; Aardvark does provide the advantage of acting as ahe "intermediary" who bears the "social cost" of asking (possibily annoying) questions.


Ali Sinan Koksal - 11/21/2011 0:52:24

Chapter 1 of "The Design of Search User Interfaces" surveys a number of design guidelines for search interfaces that ought to be adopted for improving these. The Aardvark paper presented a social network search engine, redirecting relatively long queries to members of a user's social network, exploiting intimacy as a source of trust.

Most of the guidelines discussed in the book chapter are, I believe, taken for granted right now, and look obviously beneficial. For instance, showing the query in context inside the content snippets and highlighting it is a useful feature offered by Google, allowing users to learn about webpage content without having to visit them. Auto-correcting spelling mistakes and thereby avoiding empty search results is also clearly valuable. Faceted search refinement is quite useful in websites where content is thoroughly categorized, like in shopping sites. Google's simplistic design choice is what distinguished it from its competitors (along, of course, with the revolutionary PageRank algorithm for ranking results).

Responsive search systems and fast feedback to queries allow for fast refinement of search queries, which bears similarity to the guidelines for designing responsive direct manipulation interfaces. As for having multiple ways of formulating the same queries, this challenge may be addressed more efficiently as natural language processing techniques progress.

The library paradigm vs. village paradigm in discussing Aardvark is an interesting one. Aardvark was closed in September after being acquired by Google, and looked like a well-thought approach for addressing the challenge of getting fast answers to queries that a simple Google search could probably not answer. Of course, there is the risk that users may become reluctant to keep investing their time answering queries for free.


Shiry Ginosar - 11/21/2011 0:54:36

These two papers discuss various innovative search interfaces that differ from the usual key word search paradigm. In her book, Marti Hearst presents guidelines for the design of such interfaces while focusing on faceted search. Aardvark presents a social search engine that allows one to search for answers from one's friends rather than from online data publishers.

I think that in all cases of innovative search interfaces the interesting open question is that of evaluation. Normally, information retrieval systems are evaluated and compared to each other using the TREC metrics of precision and recall. These metrics do not easily translate to the case of an interactive system such as Hearst's faceted search. Even in the case of Aardvark which is arguably not interactive in the same way, the use of these metrics would be unclear. The paper itself presents a form of evaluation, but the design of the experiments as well as the analysis of results is very unclear to me. The comparison between Aardvark and Google almost seemed to be in Google's favor, despite the fact that the authors argued otherwise...

Galen Panger - 11/21/2011 3:38:11

It’s interesting to see Aardvark published in an academic journal when it was a commercial product; in a way, it’s nice that the authors/creators laid out the functionality for the world to see and to use. However, the paper also points to the problem of academic conferences accepting papers about a product rather than a broader set of concepts and challenges. Yes, it’s possible to present Aardvark that way, but the first three words are “We present Aardvark,” rather than “here’s the motivation/challenge,” or whatnot.

Aardvark as a tool, I think, is interesting but particularly burdensome to users. First of all, it requires that we know and identify what we are good at broadly—I said once, for example, that I play piano and know about piano. But a question about a contemporary artist threw me. My expertise about playing piano did not translate into a broader awareness of contemporary pianists. Aardvark also requires users to type individual responses that, unlike forums or discussion sites, don’t scale. So my work is for the benefit of someone I know, but for them only. Furthermore, Aardvark does not seem like a tool that would survive the curiosity stage for most users. After all, most users of Wikipedia and other online knowledge repositories are consumers rather than contributors.

Hearst’s guidelines seem somewhat obvious in light of my/our heavy usage of Google (“of course suggestions are helpful!”), but it is nice to hear about academic research in the area of search interfaces. It’s also nice (though obvious) to hear discussion of some of the design trade-offs, for example, in ranking transparency versus relevance. It’s easier to understand why something is the first result, if the ranking algorithm is simple and clear, but if the algorithm is simple it might not be effective. Left unstated are elements that affect the user experience but are out of their direct experience—for example, more transparent ranking algorithms are easier to game (and spam).


Suryaveer Singh Lodha - 11/21/2011 6:17:23

The Design of Search User Interfaces:

The author provides guidelines for design of search user interfaces. The five major indicators of the 'usability' of the interface, as pointed out by the author are: Learnability, Efficiency, Memorability, Errors and Satisfaction. The author also discusses in details the 8 design guidelines for search user interfaces. One thing I found particularly interesting was:

The mportance and implications of a rapid response: The fact that user search strategies might be impacted by slower response from the search engine was insightful. The fact that a perceivable lag may interrupt user's thought process was something I had never thought before, but now makes sense! I can certainly see, how having a quick response for a query would help me in working with the "flow", and not feel interrupted. But at the same time, while searching for a comparative flight cost, the delay somehow feels acceptable to users right now. Wou;dn't it be wonderful for such complex queries to have a real-time turn around time as well!

One thing I did not find in the paper, but has bothered me about a particular search interface recently is Amazon's Mechanical Turk interface. I think that in such search interfaces, users might be typing the smae query very repetitvely. For example, lets say I'm a worker on Mturk, and I like the tasks posted by a group - "Berkeley Rotobears", and among all the tasks they post, I like the task which is about image (keyword) and pays more than 10 cents. Should I type all of this whenever I'm logging on to MTurk webpage? It would be great if there was a way to store my search as a macro which I could choose while searching, or better I could also have a default macro search. I think having features such as those in special use search interfaces could be very useful. I used something similar in industry, where I could track all tickets (work assignments) assigned to me by default on my home page in office. Certainly these examples are not relevant to the larger mass who uses search interfaces such as google/yahoo/msn, but still such features certainly do play a role in developing a good search UI, based on information as to how it would be used!


The Anatomy of a large scale social search engine:

The authors present a novel way to search using a user's social network. A user can post a question, and then the system tries to "search" for the best person in user's social network to answer the question. Here the trust on the result of search relies on the level of trust between user and the person who answers the query, which is very different from traditional search engines. In essence, Aardvark is a social search service that connects users live with friends or friends-of-friends who are able to answer their questions. Users submit questions via the Aardvark website, email or instant messenger and Aardvark will identify and facilitate a live chat or email conversation with one or more topic experts in the asker's extended social network. Aardvark can be used for asking subjective questions/recommendations/advice for which human judgment or recommendation is desired. It can also be used extensively for technical support questions. Users can also review question and answer history and other settings on the Aardvark website. I would have wanted to use it, but Aardvark was bought by Google and the project is now discontinued as of Sept, 2011 (I'm really interested in knowing the reason!). I haven't used it, but I felt this might have been useful for people with good /strong active social networks. In my network, I have seen many firends posting such questions as their staus message on Facebook/ Google+/ Gtalk. The hope is that if someone knows the answer/wants to help out and is active on social forms will probably answer the question anyways. Aardvark would have been useful if it provided control to asker of selecting groups in which the question may be posted, when the asker is looking for an answer, but doesn't want the whole world to know about it! Example, as a student,what if you are looking for a sample solution to a problem set, but fear posting such things online, because may be your professor/GSI etc is on your social network as well!


Allie - 11/21/2011 8:32:40

In "The Anatony of a Large-Scale Social Search Engine", Horowitz and Kamvar eplore the concept of a social search engine, in which queries are highly contextualized and subjective. In Aardvark, users ask a question, either by instant message, email, web input, text mesage, or voice. Aardvark then routes the question to the user's extended social network most able to answer the question. The search is for the right "person", rather than "document" to satisfy the searcher's information need. Rather than "trust", the answer Aardvark generates is determined by "intimacy" to the user.

Its main components are: 1) crawler and indexer 2) query analyzer 3) ranking function 4) UI, constructed using a social graph and an inverted index (based on the forward index). Since the quality of answers Aardvark generates is dependent on past user-generated data, and attempts to model the user as a content-generator, with probabilities indicating the likelihood she wil respond about given topics. Users are mainly ranked via expertise p(ui|q), connectedness p(ui|uj), and availability.

Aardvark is apparently most useful when used like online tech support, via a chat-like interface.

In contrast, Chapter 1 of Marti Hearst's "Search User Interfaces" book emphasizes the importance UI design in relation to document search. She introduces 8 benchmarks for evaluating good UI 1) Offer informative feedback 2) Support user control 3) Reduce short-term memory load 4) Provide shortcuts for skilled users 5) Reduce errors; offer simple error handling 6) Strive for consistency 7) Permit easy reversal of actions 8) Design for closure. As the Aardvark paper suggested, document search relies on trust, determined by how well the system and UI is designed from the getgo. Depending on the quality of one's social network, I would only seek answers from my extended social network to a degree.


Apoorva Sachdev - 11/21/2011 8:52:36

Today’s readings were about search engines. ‘The Anatomy of a Large-Scale Social Search Engine’ by Damon Horowitz and Sepandar D.Kamvar focused on a social search engine where one’s extended friend circle answers the questions rather than a traditional search engine. The other article was ‘The design of search user interfaces’ which concentrated on the features of search engines and how they can be improved in various ways.

The social search engine ‘Aardvark’, was an interesting article. My first impression was that it seemed similar to Facebook’s Q&A interface and a more personal version of yahoo answers or other such question forums. I think the main catch of this implementation would be getting a large enough friend base to use the engine first, so that people can get reliable answers quickly. In some sense, it deals with problems similar to crowdsourcing, where every extended friend is like a worker and their availability, expertise and skillset have to be determined. There is also an issue of privacy since I am not sure if the system lets you define the boundary of your friend circle, so you don’t have control over which friend-of-friend the system will send the query to. Also, the analysis that the authors provided was pretty limited since they only tested for questions that were mainly targeted for Aardvark. The difference in % of satisfactory answers between the two search engines wasn’t that great although the answers provided by friends were ranked higher. I am a little split about the future of this engine, as more websites like yelp grow, that have credible readings almost about everything and as personal phone assistants like Siri become better would one wait for social-search engine response? On the other hand, as world becomes more and more social, this seems like a natural direction to proceed in.

The other reading concentrated on the design of search engines and UI design can be improved based on the guidelines mentioned in an influential paper by Shneiderman et al. A lot of the things that the article mentions have already been implemented in popular search engines like Google. For instance, the remembering past searches feature reduces short-term memory overload. The instant search option reduces the response time of the search engine. I also the liked the idea of sorting search results depending on your preference (Date, relevance etc.) instead of just how the search engine presents them. I know a lot of other websites that primarily do internal searches allow the user to restructure the results list but I don’t think Google does this. It might be an interesting option to explore. I like the way bestbuy.com and several other websites categorize their items so one can exactly see how many items will remain once a certain option is chosen/filter is applied. I think it makes the selection process much easier and less taxing on the user. Overall, the points mentioned in this article were very useful.


Manas Mittal - 11/21/2011 8:54:02

For the Hearst et al. work, I wonder how the Google Instant (re)design was introduced. While the redesign is useful for the power user in saving a few keystrokes (the efficiency argument), it might be much more useful for novice users who are able to learn the functionality of the search engine as they type their query (the learnability argument). For example, the are likely to discover the 'anding' of terms as they see the search results change when they type. Additionally, there is some impact on the satisfiability criterion - since results are so fast, users might not mind typing the wrong query (as indicated in sec 1.5.6 of the paper).

The Aardvark paper presents a social search engine. One thing to remember is that even thought Aardvark had this cool technology - nobody used Aardvark. Why was that? Do we not trust our friends? Do people not respond to Aardvark requests? At the time this paper came out, the authors called themselves Google 3.0 (and Google brought them, perhaps just as a precaution since it was 'only' 50M). Quora is somewhat similar in functionality. The Quora system is more explicit and follows a different model that Aardvark (A pull vs push model). It would be interesting to discuss this differences.


Yin-Chia Yeh - 11/21/2011 8:59:36

Today’s two papers are about search engine. The design of search interface one introduces important factors needed to be consider in designing a search interface. The Aardvark paper, on the other hand, presents a new type of search engine that aims to direct users’ query to someone who might know the answer in users’ social network. I am a little bit surprised to find out that there are so many factors to consider within a seemingly simple and common search interface. Especially those details I never recognized until reading this chapter. For example, the keyword suggestion of search result shows in both top and bottom of result page. Even this kind of small detail can improve our user experience a lot. I tried Google and Bing to see if they are both using this trick. The answer is yes. Moreover, I find that Bing also provides a search bar in the bottom of search result. I never notice this kind of nuance before. The most interesting part of Aardvak paper is the query is answered by human. Therefore it allows (or requires) the query in human language and it can deal with very subjective questions. I think it is a great advantage comparing with traditional search engine because it addresses a blind spot of traditional search engine, providing customizing search result for different users. One design I really like in Aardvark is that if you choose to pass a question nobody will know about that. It really provides a nice buffer between people that is not doable in one on one communication, ex., face to face talk or email. What surprises me is that how fast people answering others’ questions. I would like to learn more about how the size of social network affects the quality of answers or speed of getting answers.


Rohan Nagesh - 11/21/2011 9:02:08

This week's readings revolved around the theme of search interfaces and novel search paradigms. While information retrieval has been a studied problem for quite some time now, I found the first paper "Ch. 1 Design of Search Interfaces" to be a very useful summary guide of best practices in designing a search tool. The second paper presented Aardvark, a social search tool that provided hypertargeted answers to queries that wouldn't be answered well on traditional engines such as Google.

Google appears to be a good example of an engine that incorporates many of the design practices stated in the first paper. For instance, if we take keeping the interface simple, Google is classically known for its all white, simple interface with just a search box on its home page. Second, in terms of loading results faster, Google Instant search takes that to the next level allowing the user to dynamically modify the query based on results. Also, the ads are blended in nicely to not overburden or confuse the user. Lastly, the PageRank algorithm is great at assigning priority to popular documents, which in turn will greatly increase the relevance and reduce the task completion time of searching.

The second paper, which discussed Aardvark, a startup acquired by Google but then shut down in September 2011 is a novel approach to search. Rather than finding the document that best captures a searcher's query, Aardvark searches your social graph (Facebook, LinkedIn, etc.) to find the best person to answer your question. The UI is real simple to use--pretty much just like an IM chat session. This kind of model gets accurate results for questions such as "What's a good babysitter for my two kids in Berkeley?" Reputation and recommendation-based queries I think are best suited for Aardvark's model.

I did find it interesting that Google pulled the plug on Aardvark, and my hunch is that the latency in the Aardvark system just made the system not feasible for most users. Additionally, the subset of queries that can be answered better by Aardvark over Google is very small, and perhaps Google thought they could simply integrate Aardvark search into their own search once their network (say Google+) grew.