Co-presented by Will Evans (@semanticwill) and Brynn Evans (@brynn) at the Enterprise Search Summit West 2009.
Social search has the potential to improve search practices beyond what is possible with traditional informational retrieval algorithms. Two different models of social search should be incorporated into enterprise and conventional search systems today. Collective Search involves aggregating social metadata, trends, and previous tags, bookmarks, or information shared by social networks. Collaborative Search, or question-answering, occurs when two or more participants actively engage in an information seeking task. Interactions include everything from replying to a one-time question to dually negotiating the query formation and relevancy of specific results to arrive at a shared consensus of best fit. This talk will frame the relevant models of social search in the context of Brynn’s research, and discuss the potential benefits for both users as well as organizations. We will extend these trends and findings to concrete design considerations that we encourage system designers to consider in order to leverage social search capabilities within the enterprise.
1. PHOTO BY GAIJINSEB
Designing for Sociality in
Enterprise Search Brynn Evans & Will Evans
Hello!
We’ll be talking today about Designing for Sociality in Enterprise Search.
2. Who are we?
Will Evans Brynn Evans
http://semanticfoundry.com/ http://brynnevans.com
Before we get started, we wanted to briefly introduce ourselves. (And no, we are not related!)
Will self-identifies as an interaction designer. He began thinking about social search when he
designed the user experience for the website Kayak, an online travel assistant.
Brynn is a design researcher who has done several studies of social search over the last 2
years. Social search popped out to her as a major trend (and interest) when she was doing a
study of enterprise search in early 2007. After getting reports from 150 people about their
search experiences, she saw that a huge proportion were involved in social interactions with
their colleagues at some point in their search process. The subsequent focus of her work has
been how can we leverage those social interactions during online search: when, why, and
how.
Today’s talk is a synthesis of findings from Brynn’s work, with an eye on how to actually
design for sociality in enterprise search.
3. LEGEND I n u e n ce r ( B l o g g e r ) S i te
w/detachable
s h a re d co nte nt
Brand Umbrella
Community
Core Social Features
Brand Module
B ra n d
Ap e r t u re
Branded Theme Module
Lifestyle/Editorial
B ra n d B ra n d
Content I nte re s t i nte re s t
(article, idea, comment, tip)
Community Member
Wa l l e d
In uencer G a rd e n
Community Manager
L i fe s t y l e / B ra n d
Ed i to r i a l
Audience [Non-member] I nte re s t
S h a re d Eco s ys te m
B ra n d
I n u e n ce r ( B l o g g e r ) S i te
w/detachable
s h a re d co nte nt
i nte re s t
Wi l l Eva n s
Us e r E x p e r i e n ce Arc h i te c t
S e m a nt i c Fo u n d r y, L LC
Today we’ll also share our different perspective on design and search.
We are first and foremost interested in how people engage with each other in mediated
spaces—and the complex behaviors that arise from social interactions.
This ‘Conceptual Model of The Social Web’ was drawn by Will to illustrate how we think about
sociality in mediated spaces. The point of the diagram is to draw connections between people
and the context, situations, and social objects (content, searches, etc.) that inform their
interactions. These interactions leave behind what we’re calling “social residue.”
Social residue is important to consider in design, and especially for “social interaction
design”. Social search carries with it all the concerns of designing for sociality in mediated
spaces, and involves consideration of this perspective of social interaction design. We’ll talk
more about this later in the talk.
4. Search is an activity
PHOTO BY DAVID WILD
We will begin by framing our talk with our perspective on how to think about search.
The first thing to recognize is that search is an activity—something that extends over a
period of time, not just something that happens in a quick moment and fades from memory.
Many searches take minutes to hours to days to (sufficiently) complete.
And even with shorter searches, people still experience different phases of search. Early
search queries often involve scoping out the problem space (wide, open-ended questions),
while later queries are intended to narrow the focus.
This perspective alone should make us rethink the traditional search engine approach, which
is about a much more isolated and momentary experiences.
5. An extended search process
Early Middle Late
One way to visualize this is with a map. This is an illustration of an extended search process
through time. You can imagine that you, as a searcher, advance through phases (like early,
middle, and late) which are characterized by a different set of information needs,
requirements, knowledge, state, context, and trajectory.
Keep this in mind through the rest of the talk.
6. An extended search process
Early Middle Late
One way to visualize this is with a map. This is an illustration of an extended search process
through time. You can imagine that you, as a searcher, advance through phases (like early,
middle, and late) which are characterized by a different set of information needs,
requirements, knowledge, state, context, and trajectory.
Keep this in mind through the rest of the talk.
7. Search is also already social!
PHOTO BY BREWBROOKS
The second thing we know is that search is already (very) social! Recall that the number of
implicit and explicit social interactions that Brynn saw in her original study of enterprise
search speaks to this.
However, this is not new: librarians and even researchers inside organizations have been
telling us this for years. They’ve documented how co-workers routinely turn to each other to
ask for help, seek advice and suggestions, and get feedback on early search results.
Sometimes people search together (what might be called collaborative search) when they
have shared goals; but many otherwise solo searches also involve social resources in some
way or another.
For details on when, where, and why these interactions occur, see Brynn’s papers on the
“Model of Social Search”:
http://brynnevans.com/papers/evans-chi-social-search-final.pdf
http://brynnevans.com/papers/evans-chi-inSubmission.pdf
The problem is that today’s social search solutions are kludgy. You can email or IM your
colleague. You can post a ‘lazytweet’ on Twitter. Or you can knock on the cubicle next door.
These all break you from your concentration on the task at hand, and often involve drastic
state changes (as well as interface switching). Not ideal.
8. Social interactions improve search
PHOTO BY INGORRR
Finally, there’s evidence that social interactions can improve search outcomes. Brynn has
seen this in her latest research (not yet published) as have others (Peggy Anne Salz found
that asking people for help during mobile phone search, via ChaCha, greatly improved search
outcomes).
This makes sense intuitively, but is also supported by knowing that:
1) 2 out of 3 searches involve finding things your friends/colleagues have found previously.
2) our networks have lots of tacit knowledge that we can tap
Ref: http://brynnevans.com/papers/evans-kairam-pirolli-inSubmission.pdf
http://brynnevans.com/papers/Cognitive-Consequences-of-Social-Search-WIP.pdf
9. 1) 2 out of 3 searches involve finding
things your friends and colleagues
have previously found
2) Our networks have tacit
knowledge
Social interactions improve search
PHOTO BY INGORRR
Finally, there’s evidence that social interactions can improve search outcomes. Brynn has
seen this in her latest research (not yet published) as have others (Peggy Anne Salz found
that asking people for help during mobile phone search, via ChaCha, greatly improved search
outcomes).
This makes sense intuitively, but is also supported by knowing that:
1) 2 out of 3 searches involve finding things your friends/colleagues have found previously.
2) our networks have lots of tacit knowledge that we can tap
Ref: http://brynnevans.com/papers/evans-kairam-pirolli-inSubmission.pdf
http://brynnevans.com/papers/Cognitive-Consequences-of-Social-Search-WIP.pdf
10. Mere degrees of separation
PHOTO BY DAVID RODRIGUES
Now, one of the reasons it’s possible to take on the social search challenge today is that
networking technology is finally catching up to what people hope to experience in online
search.
We all know about the “6 degrees of separation” studies — it means that any two people in
the world are only about 6 hops away from each other (median = 4.5), almost regardless of
where they live, what they do for a living, etc.
Researchers have been replicating these experiments for some time, and recently in online
social networking sites (like Delicious, Flickr, LinkedIn). Remarkably, people are only about
2-3 nodes away from each other in online networks.
This means that our networks are rather small and inter-connected, and that these tools are
making it easier to connect with each other (all good news for social search).
Ref: http://videolectures.net/kdd08_leskovec_mesn/
11. But, before we go on, we want to emphasize that we’re not advocating for merely adding
social networks like Facebook, Twitter, or LinkedIn to your enterprise search tool. Linking up
network accounts won’t get at the kind of sociality that we’re talking about.
The reason for this is that it’s not just the presence of social connections that matter — it’s
how those connections get put to use, how they’re used, and why they’re useful. That’s the
sociological and anthropological side of studying social interaction design that’s relevant to
understand here.
12. Social interaction design
(SxD)
Social interaction design is the conceptualization, research,
strategy, and production of dynamic ecosystems that support
conversation. -Will Evans (2008)
As we’ve mentioned, what we’re advocating for is a shift in perspective from mere HCI and
network theory, to social interaction design (SxD).
This is not just web design, either. SxD is about designing for complex dynamic ecosystems,
where the content is the conversation and the user interface is the social interface.
Another way to think about this is that we are now tasked with designing interfaces where a
user is connecting to a community, a large group of people with whom s/he expects to
interact. All parties are connecting via the same interface, same tool, same functionality. Our
attention must be put to the behaviors and expectations that arise as a result of this shared,
mediated space.
Pure HCI (user-computer interaction) no longer informs us completely about what to do here.
13. Why you should care:
1. This is a business problem
2. The traditional IR perspective doesn’t
capture what people are doing
3. The technology is there, but it’s the
sociology that matters.
Alright, but why should you care?
1) This is inherently a business problem. Business need to care about social search because it
very much can enhance productivity among workers, and in the current state, several days
per month are wasted on searching and looking up information.
2) Traditional information retrieval (IR) doesn’t account for the social behaviors involved in
seeking information. IR doesn’t capture what people are doing. We have to introduce new
perspectives from sociology to understand this.
3) Social networks are now serious mass of connectivity — recording not only our
relationships, but also our activities, interests, preferences, friends, links, etc. The technology
has gotten to the point so that it’s possible to integrate social networking with search. But
remember again, that although the current state of technology will make this possible, we
have to pay even greater attention to people’s expected and emergent behaviors around
sociality in search.
14. I. Social search strategies
II. Modalities of social search
III. Design considerations
This is what we’ll talk about today.
First, we’ll review a few social search strategies that Brynn has identified in her work.
Then, we’ll talk about three modalities of social search — or three ‘flavors’ of social
interactions that are importantly different but will serve different roles in search.
Finally, we introduce design considerations and early mockups of what a social search system
might look like.
15. I. Social search strategies
First, social search strategies, which we’ll introduce from the perspective of a persona.
16. The new hire,
Molly ...and her team
PHOTOS BY SUZUKIMARUTI AND ROO REYNOLDS
Meet Molly.
Molly is the new hire at your firm. She’s been brought in by the CTO to manage a team of
engineers who are tasked with developing the next generation enterprise social software
services. Molly has many years of experience in enterprise software development, but is still
learning about the enterprise 2.0 space.
In particular, she needs to know the state of the art in enterprise social software as well as
the attempts her company has made in the past to offer their own unique services in this
space.
17. Search Strategy: Consulting the network
PHOTO BY BENSPARK
She takes a two pronged approach to solving this search problem. (These two strategies are
based an analysis of search failures; as of yet unpublished: Brynn Evans & Ed Chi).
First, in the early phases of her learning process, she decides to consult her network. Asking
colleagues for help directly lets her quickly get up to speed with what they know, so she
doesn’t waste time duplicating their efforts.
To achieve this, she calls a meeting with members of her team and asks them to brief her on
their specific work to date. She also leads interviews with a few key stakeholders who reside
outside of her immediate group.
After doing this, she has a roadmap and a framework for thinking about enterprise 2.0. The
briefings act as guideposts of important trends and directions in the field, which point her in
the right directions going forward.
18. Search strategy: Forge ahead, ask later
PHOTO BY US NATIONAL ARCHIVES
Molly’s next strategy is to forge ahead on her own, only asking for additional help at a later
date.
This is an important strategy for Molly for several reasons. First, Molly is the director of her
team and it’s her responsibility to master the problem space herself. Second, Molly has
already burdened her team with questions up front; now she must embark on her own. And
third, she's finally equipped with sufficient background knowledge and know-how to decide
which tools to use to dig deeper and unearth more information about enterprise 2.0.
Forging ahead on one’s own is a common strategy for all the reasons Molly gives. However,
equally important is checking back in with colleagues to get feedback on what’s been
discovered. This strategy is actually quite common among everyday searchers as well, and
Molly is already planning the next meeting where she will present her findings to her team
and request their feedback.
19. I. Social search strategies
II. Modalities of social search
Keeping in mind that asking for help appears separately in different search strategies
throughout a search process...what other ways can people help us search better?
In this next section, we’ll talk about three distinct ‘flavors’ of social search, and what their
advantages and drawbacks are.
20. Three Modalities of Social Search
collective collaborative
social search social search
friend-filtered
social search
These three modalities of social search are generally called:
Collective Social Search;
Friend-Filtered Social Search;
and Collaborative Social Search.
Ref: http://www.readwriteweb.com/archives/3_flavors_of_social_search_what_to_expect.php
21. collective
social search
PHOTO BY http://whowantstobe.co.uk/bench/faq-en.php
"Collective social search" is similar in concept to the wisdom of crowds, in that search is
augmented by trends shared on a network (a la Twitter Trends) or results ranked against the
real-time buzz of a group.
Pooled, aggregated data from the collective may point us to new avenues that expand our
discovery process.
22. tre nd 1
topic 2
them
e3
Use case: Exploratory search; new domains
Problem: The crowd is anonymous
ICONS BY ICONAHOLIC.COM & ICONARCHIVE.COM
This generally looks like pulling trends, current topics, or themes from the activities of many
individuals and using this to inform the search experience (results, UI, etc.)
When is this useful?: This is most useful when exploring a new search space — through an
exploratory or open-ended search problem. It could also help for studying new domains
where you may not know the vocabulary of that domain, or you don’t know how to drill down
to an answer on your own. (How do I learn more about this discoloration in my fingernail?)
What is the drawback? The problem with information sourced from a crowd is that the
crowd is anonymous to you. These aren’t people you know, which means you may not trust
their information, it may be irrelevant to your current search, or it may be real-time trends
which are only useful if you have real-time search needs. (What are the Red Sox up to?)
23. There are some implementations of collective social search in mainstream products today.
Yahoo suggest provides suggested query strings based on millions of people’s search
behaviors.
Trends on Twitter shows popular keywords being tweeted right now (tapping the real-time
aspects of collective social search).
And OneRiot takes this one step further to highlight popular URLs and topics (not just
keywords), but looking at activity across multiple social sharing websites.
24. There are some implementations of collective social search in mainstream products today.
Yahoo suggest provides suggested query strings based on millions of people’s search
behaviors.
Trends on Twitter shows popular keywords being tweeted right now (tapping the real-time
aspects of collective social search).
And OneRiot takes this one step further to highlight popular URLs and topics (not just
keywords), but looking at activity across multiple social sharing websites.
25. There are some implementations of collective social search in mainstream products today.
Yahoo suggest provides suggested query strings based on millions of people’s search
behaviors.
Trends on Twitter shows popular keywords being tweeted right now (tapping the real-time
aspects of collective social search).
And OneRiot takes this one step further to highlight popular URLs and topics (not just
keywords), but looking at activity across multiple social sharing websites.
26. friend-filtered
social search
PHOTO BY CLAUDIA LIM
Friend-filtered social search is approximately what Google is doing with its social search
experiment: providing social data that your peers, friends of friends, and wider "social circle"
have shared. This data could appear alongside traditional search results (as with Google) or
be exclusive results from within your peer network (as with TuneIn.com).
Ref: http://www.google.com/support/websearch/bin/answer.py?answer=165228
27. d
nd tren
frie
friend post
frien
d lin
k
Use case: Information from trusted people
Problem: Personal networks are narrow
ICONS BY ICONAHOLIC.COM & ICONARCHIVE.COM
Friend-filtered implementations could look like filtering trends from only your personal peer
network; or it could be sharing information and links that your friends have specifically
shared.
When is this useful?: This is most useful when the searcher has a specific search goal in
mind (i.e. searcher is drilling down), or when the searcher is actively looking for
recommendations or tips for something like travel, restaurants, bars, etc.
What is the drawback? The problem with friend-filtered social search is that personal
networks tend to be narrow, meaning that your network may know nothing about your
current search query (depending on the query and nature of the network). Personal networks
can’t be relied upon for general-purpose searching.
28. The prominent example of this at the moment is Google’s social search experiment.
Here, imagine that Molly is doing a general search for ‘enterprise social search’ and related
topics. If she clicks on Google’s Social Search option, she’ll see links and blog posts that
colleagues in her network have shared.
This is great since it provides three links to posts she hasn’t yet seen; and at the same time, it
only provides her with three links.
Note: there are other services like TuneIn.com, statussearch.net, and FriendFeed (now
Facebook search) who return search results based only on data shared by your personal
network.
29. collaborative
social search
PHOTO BY BREWBROOKS
"Collaborative search" is when two or more users work together to find the answer to a
problem. This could look like IM-based question-answering (a la Aardvark), Yahoo! Answers
or LinkedIn Answers (which is relatively passive and asynchronous) or over-the-shoulder
two-person search.
30. conversation
Use case: Addressing the ‘vocabulary problem’
Problem: Fear of embarrassment (FOE)
ICONS BY ICONAHOLIC.COM & ICONARCHIVE.COM
This is commonly referred to a question-answering, as it’s obviously different from the other
modalities of social search. Here, two users can engage in a conversation with each other to
dissect a problem space and help provide search tips to each other. A major advantage of
question-answering is that people speak to each other using natural language, which is
incredibly useful for open-ended queries (e.g. "What is 'design thinking'?") or queries about
unfamiliar domains (e.g. law, health, business, depending on your background).
When is this useful?: Thus, the best use case for collaborative social search is for addressing
the “vocabulary problem” — when the terminology of a particular domain is unknown to the
searcher, making it hard to think up the right terms to search with. It’s also useful when
searchers get stuck on a search after several attempts, since they can benefit from talking to
experts for help or bouncing ideas off of colleagues to put them back on the right path.
What is the drawback? There are several drawbacks with this approach. First, it takes
precious time away from your colleague to engage in a question-answering session with you.
Second, people often feel that they’re burdening their colleagues by asking a question,
instead of searching for it on their own (or incurring a obligation or debt). And third, there’s
the fear of embarrassment factor (or FOE). People avoid asking a colleague for help when that
lack of knowledge reflects badly on themselves.
And yet, talking to a real human can greatly assist in information seeking — but it’s things
like fear of embarrassment and incurring a social debt that a social interaction design
approach wants to consider before designing a social search system.
31. Examples of this in the wild are, of course, face-to-face questioning, instant messaging, and
posting a lazyweb tweet.
There’s also Aardvark, which tries to connect questionners with knowledgeable people from
their extended trust network to answer a question. It works remarkably well at finding target
answerers, and reduces a lot of the social issues associated with in-person question-
answering.
32. I. Social search strategies
II. Modalities of social search
III. Design considerations
Given what we’ve talked about so far, what are the design considerations for social search
systems?
33. Early Middle Late
Search
Phase
Recall this diagram of the extended process of search. We’ve illustrated how social
interactions come into play during many phases of search: both early, middle, and late.
34. Early Middle Late
Search
Phase
Recall this diagram of the extended process of search. We’ve illustrated how social
interactions come into play during many phases of search: both early, middle, and late.
35. collective
social collabora-
search tive
social
search
friend-
Early filtered
Middle Late
Search social
Phase search
If we plot the modalities of social search on top of this extended search process, we see that
the strongest value add (for incorporating social resources in different phases) is
approximately this.
There’s a strong case for providing passive social support early in search. Users don’t want
to interrupt their colleagues quite yet, and they want to try to search using the sophisticated
tools available to them at the outset. Social support during the first few query attempts
should be implicit, passive, and likely in the form of collective social search.
Later in search, providing explicit social support is preferable. People may be refining
their query at this point and/or get stuck with a particular query formulation. Asking
colleagues for help (using natural language queries) can help with the final phases of a
search.
Friend-filtered social search is relevant throughout. This means showing trends from your
network early or passively showing people from the network who are knowledgeable. It also
means interspersing data and information shared by your network into the search results
listings.
36. We’ll take you through a few mockups for what this might look like. (Keep in mind that these
are wireframes, meant to convey a concept, not the final visual implementation).
Alright, imagine that our character Molly begins her search by tying in the query “Enterprise
Social Search” ....
37. ...of course, she can see a set of recommended searches while she’s typing. These can come
from her personal network (as opposed to from a wider network).
And there is some indication of how many related search terms exist in the system. Here, you
see it says:
Search suggestions for Enterprise Social Search (8)
38. 1
2
3
Next, she sees a set of (1) search results, plus (2) related concepts and (3) related people in
her network. These items in the side panel (2, 3) are meant to be present, but not imposing—
merely passive social support in this early phase of search. Green/gray indicator lights (by
the people) refer to the availability of these people. Molly can click on and interact with
‘Brynn Evans’ because she’s listed as green and currently available.
We expect, however, that the side panel support provides guidance but does not detract from
the presence of the main search result items.
39. 1
2
3
4
5
There are few points to make about what results get returned to Molly.
(1) Keywords in bold indicate a match to the search terms she initially used (this is standard).
(2a) Keywords in blue, however, indicate a related term from related searches in Molly’s
network. These are social signals pointing out related concepts or terms that experts may be
using to describe this domain.
(2b) The idea with the number next to the social terms (8): is to indicate how frequently this
term occurs in related searches — or something along those lines.
(3) This is similar to the “Social Depth” concept shown below the result. We imagine that as
peers/colleagues search for stuff, those search terms and the items found during the search
get added to a “Stack” or “Collection” that Molly has access to while she’s searching. If Molly’s
current search matches a items in a related Stack, the “Social Depth” indicator will be high.
She now knows that there is a lot of related search material from her network if she continues
down this line of querying.
(4) Molly also has the ability to mark, share, and tag good results she comes across. This will
add items to her Stack/Collection, which her colleagues can access at a future date.
(5) Finally, Molly can consult with her network directly (from within the search interface).
40. This is what Molly sees if she clicks on one of her network contacts: an option to see their
Lists (Stacks/Collections), to ask a question via instant messenger, or to subscribe to their
content. Subscribing would prioritize their content over other contacts’ content when she
performs searches in the future.
42. This brings up an instant messaging interface directly in the search results page.
Molly can now pose any question she wants to Brynn. They can negotiate shared meaning of
Molly’s search problem, using natural language. And Molly can get direct tips/advice/
suggestions from this knowledgeable contact, Brynn.
43. When Brynn suggests a new keyword combination that sounds promising, Molly can click-
and-drag the line from the IM box and add it to the search box.
44. A final idea we wanted to introduce was what happens when Molly persists in her search but
loses steam or fails to find what she’s looking for.
We can detect this behavior to some extent. “Thrashing” is what happens when you add or
subtract one or two words from your search query, but you don’t deviate TOO much from the
initial formulation. Once searchers get ‘anchored’ with one formulation — or one way to go
about finding an answer — they often end up thrashing like this.
If we detected this behavior, knowing that this is a later-search stage, we can now offer
explicit social support to the searcher. This might be in the form of a pop-up saying “You
seem to be having some trouble. Would you like to ask your network for help?”
Asking your network could be like pinging one person individually (like the IM screens we saw
on the previous slide) or posing a question to your entire social network (like a lazytweet).
Minimally, a pop-up like this might help “break” Molly from her initial formulation, to realize
that there are new ways of thinking about her problem. Brynn’s research has also shown that
when searchers go through the process of asking a large public network for help, they spend
considerable time thinking about how best to ask the question. And in the process reveal to
themselves new ways to formulate their own query.
We think it’d be great to offer this type of support to searchers within their search
45. collective
social collabora-
search tive
social
search
friend-
Early filtered
Middle Late
Search social
Phase search
In conclusion, this diagram illustrates a number of the concepts we’ve discussed today:
People employ a variety of strategies when they engage in information seeking. Social
interactions occur in many stages of search, and both implicit and explicit social search can
help people search better.
Our hope is to find ways to design for these strategies and embed them in the proper context
within enterprises.
46. Summary
I. Social search strategies
II. Modalities of social search
III. Design prototypes
End. Questions?