=Paper=
{{Paper
|id=Vol-443/paper-13
|storemode=property
|title=Visual Interfaces for Improved Mobile Search
|pdfUrl=https://ceur-ws.org/Vol-443/paper13.pdf
|volume=Vol-443
}}
==Visual Interfaces for Improved Mobile Search==
https://ceur-ws.org/Vol-443/paper13.pdf
Visual Interfaces for Improved Mobile Search
∗ †
Karen Church Barry Smyth Nuria Oliver
Telefonica Research CLARITY Telefonica Research
Via Augusta 177, 08021 University College Dublin Via Augusta 177, 08021
Barcelona, Spain Belfield, Dublin 4, Ireland Barcelona, Spain
karen@tid.es barry.smyth@ucd.ie nuriao@tid.es
ABSTRACT However, the Mobile Internet represents a challenging infor-
The Mobile Web promises a new age of anytime, anywhere mation access environment, particularly from a search stand-
information access to billions of users across the globe. How- point. Limited screen-space and restricted text-input and in-
ever, the Mobile Internet represents a challenging informa- teraction capabilities exacerbate the shortcomings of modern
tion access environment, particularly from a search stand- Web search. To date most mobile search interfaces are sim-
point. In this paper we present two visual interfaces for im- ple adaptations of standard Web interfaces, where users are
proved mobile search. First, we present SearchBrowser, a presented with a ranked list of results. For mobile search to
map-based interface that offers richer end-user interactions succeed we need to think beyond simply query-based inter-
by taking into account important mobile contexts including faces and towards interfaces that can offer richer interactions
location and time. Second, we consider the social context by taking into account important mobile contexts that have
of mobile search and present SocialSearchBrowser; a proof- an impact on mobile users needs.
of-concept interface that incorporates social networking ca-
pabilities to improve the search and information discovery In this paper we focus on the mobile search interface and
experience of mobile subscribers. we offer on a more radical rethink of mobile search. It has
always been our contention that mobile search differs signif-
Author Keywords icantly from Web search, not just because of the devices but
Mobile Search, Search Interfaces, Social Search, Social Net- also because people’s information needs differ when mobile.
working, Mobile Web, Context, Preferences, Location, Time Previously we examined the information access patterns of
real mobile subscribers using log analysis techniques [6].
More recently, we investigated mobile information needs in-
ACM Classification Keywords
situ, examining the unique contextual factors that impact on
H.3.3 [Information Storage and Retrieval]: Information Search user needs [5]. Our findings indicate that when users are
and Retrieval, H.5.2 [Information Interfaces and Presenta- mobile there is a clear location and temporal dependency
tion]: User Interfaces in their information needs. Furthermore, we found that the
needs that arise when mobile cannot always be answered by
INTRODUCTION existing search engines, because existing search engines do
There are over 3.5 billion mobile subscribers worldwide1 not take key mobile contexts into account.
and with continued advances in devices, services and billing
models, the number of subscribers venturing online via their Based on the findings of these previous studies, we devised
mobile handsets is increasing. Thus the mobile space looks two new visual interfaces for mobile search, both designed to
set to usher in a new age of anytime information access. emphasise the importance of location, time and preferences
∗The early work presented in this paper was carried out while as key elements of search context. Unlike traditional search
Karen Church was a PhD student in University College Dublin.
interfaces, which require user input before providing infor-
This material is based on works supported by the Science Foun- mation to end-users, our interfaces give mobile users inter-
dation Ireland under Grant No. 03/IN.3/I361 and Grant No. esting information from the beginning. Our approach is de-
07/CE/I1147. The later work, i.e. the SocialSearchBrowser pro- signed to change the mobile search paradigm. The interfaces
totype is being carried out at present in Telefonica Research. present historical query, comment and result-selection data
†CLARITY: Centre for Sensor Web Technologies.
for users to navigate through on an interactive map-based in-
1
http://www.un.org/apps/news/story.asp?NewsID=28251 terface. The rich user interface enables users to interact with
&Cr=Telecommunication&Cr1 the past activities of other users, execute searches, view past
result-selections and filter queries based on context informa-
tion. In short by presenting users with information about
what others are searching for we believe we can offer an im-
proved search experience.
This paper is organized as follows. In the following section
Workshop on Visual Interfaces to the Social we present some related work. Next, we describe Search-
and the Semantic Web (VISSW2009), IUI’09,
Feb 8 2009, Sanibel Island, Florida, USA.
Browser, a map-based interface that offers richer end-user
Copyright is held by the author/owner(s).
1
interactions by taking into account important mobile con- More recently researchers are investigating the social side to
texts including location and time and we describe the results Web search. For example, Collaborative Web Search (CWS)
from a recent user study. Based on the outcomes of this eval- involves utilising the search histories (i.e. queries and result-
uation and the findings from our diary study [5], we turn to selections) of communities of like-minded individuals. In
the Social Web and explore the social context of search. In [7], Freyne et al. looks at integrating CWS with social brows-
the final section of this paper we propose a proof-of-concept ing, i.e. leveraging past browsing behaviour of users to guide
interface called SocialSearchBrowser that incorporates so- others to relevant web content, to produce an integrated so-
cial networking capabilities to improve the search and infor- cial information access service. Preliminary results from a
mation discovery experience of mobile subscribers. live user trial indicated that the use of social cues helps users
to access relevant information in an easy and efficient man-
RELATED WORK ner.
The focus of this paper is on novel mobile search interfaces
that utilize key mobile contexts. There has been a range Another approach is to exploit Web 2.0 technologies, specif-
of previous research that investigates improved search inter- ically Web annotations, to improve Web search. The basic
faces in the general Web space. Our current work combines premise is that by allowing users to annotate search results
work on exploratory search, mobile search and social search. and to share these annotations with others, the search ex-
As such we have identified three areas of related research: perience can be improved. In [3], Boa et al. propose two
novel algorithms, SocialSimRank (SSR) and SocialPageRank
Exploratory Search (SPR) to explore the role of social annotations on similar-
Traditional approaches to Web search typically involve a user ity ranking and static ranking respectively. Results from an
submitting a query via a search box and viewing a list of re- evaluation using a del.icio.us dataset shows that both SSR
sults. More recently, a new class of search has emerged, and SPR could benefit Web search significantly.
called exploratory search [14], which supports the explo-
Another related area of interest is social search. Social search
ration and discovery of information through both querying
in this context involves exploiting different forms of human
and browsing strategies. There have been a number of ex-
judgements, ratings and interactions to improve the overall
ploratory search systems developed to date. For example,
Hearst presents Tile-Bars [9], a technique which uses the search experience. For example, Microsoft’s U Rank2 , is a
structure of text to provide a visualization aid to end-users. prototype search engine that allows people to edit, annotate
TileBars help users to visualize the document length, query and organise search results. U Rank enables users to collab-
term frequency and query term distribution, thus assisting in orate with one another through sharing and recommendation
relevance assessments of documents. Yee et al. [15] presents of search results in easily accessible lists.
an alternative interface for exploring large collections of im-
Most relevant to our current work is utilizing social net-
ages using hierarchical faceted metadata and dynamically
works to enhance search results and online interactions. In
generated query previews. While recent work by Alonso et
[8] Golbeck and Wasser introduce an application called So-
al. [1] describes an interface that utilises timeline data to en-
cialBrowsing which works by analyzing web page content
able effective presentation and navigation of search results.
and highlighting words or phrases which have some con-
textual social information. In [12], Mislove et al. present
Mobile Search PeerSpective, an experimental prototype which exploits both
Another area of research related to this paper concerns in- the hyperlinks of the Web as well as the social links within
novative approaches to mobile search interfaces. FaThumb communities of users to inform a new search result rank-
[11] is a user interface designed for navigating through large ing approach. An evaluation of the PeerSpective search en-
data sets on mobile devices providing a more efficient means gine showed that it performs well in terms of disambigua-
of mobile search. FaThumb uses faceted metadata naviga- tion, ranking and serendipity of search results.
tion and selection as well as incremental text entry to narrow
the results. A user evaluation demonstrated how the facet
Our Proposal & Contributions
based navigation is faster for less specific queries.
Our current work is similar in nature to the QnA approach.
Questions not Answers (QnA) [10] also provides an inter- The QnA system essentially tags queries with a location.
esting alternative to the traditional search interface. Rather These queries are displayed on a map-based interface en-
than examining how to provide high-quality search results, abling users to visualise the search space. The QnA pro-
the QnA approach is to provide access to previous queries totype does not, however, provide any means for a user to
posted from the user’s current location. This novel user in- filter queries, other than by location. Given that the vol-
terface displays queries made by other people in a given lo- ume of queries at specific locations is likely to be quite high
cation using a map-based interface, providing users with an and there is no means to filter queries, the QnA prototype
enriched sense of place. By clicking on the queries users raises a new interface/presentation challenge. Furthermore,
can execute the displayed search. In a live user evaluation our prototypes focus on the social side to mobile search al-
[2], users found the interface to be useful and they enjoyed lowing users to interact with the result-selections and com-
the increased level of interaction the interface enabled. ments of other users. In the SocialSearchBrowsr application,
we investigate this social context further by utilizing social
2
Social Search http://research.microsoft.com/projects/urank/
2
networks for improved information access. We think this 1. Query 2. Query with result-
is a core area to address given the unique characteristics of selections
the mobile space. Thus the core contributions of this pa- 3. Query with comments 4. Query with comments
& result-selections
per are as follows: (1) we present SearchBrowser, a context-
aware mobile search interface that enables situated discovery
of information, (2) we describe a recent user evaluation of
SearchBrowser and demonstrate some initial positive results
and (3) we propose SocialSearchBrowser, an extension of
SearchBrowser, which explores the social context of search
by incorporating social networking to improve the informa-
tion access experience of the end-user.
THE SEARCHBROWSER INTERFACE
The basic premise behind the SearchBrowser interface is that
by allowing users to see what other users have been search-
ing for and interacting with, we can improve the search ex-
perience. The interface utilises contextual information such
as location and time to provide a unique experience. The in-
terface provides mobile users with information more proac-
tively, thus encouraging discovery of content. The proto-
type3 consists of a text box that allows users to issue new
SearchBrowser interface on startup
queries and a small map centered at the user’s current phys-
ical location. The map shows queries submitted by other
users in that location and two sliders at the bottom of inter- Figure 1. The SearchBrowser interface: shows queries, comments and
result selections made by other users in a given location. Note the leg-
face are used to filter the queries displayed on the map. end, which describes the interface icons, is shown for illustrative pur-
poses only and is not shown to users of the system on startup.
The Map Interface
When the user first initialises the application, he/she is shown
a map centered at their current location (Figure 1). The map to one of the listed URLs or to re-execute the query4 .
shows all recent queries entered by other users in that lo-
cation. We refer to these queries as the prime set. The To help users distinguish between popular queries, the icon
map is updated periodically so that newly entered queries sizes of the queries change based on their popularity. We
are displayed. Queries submitted by other users, but without use a simple measure of popularity based on the number of
any result selections, are identified by a small magnifying times the query has been submitted and the amount of result-
glass with an associated label (See Figure 1 icon (1)), while selections and comments associated with the query. Smaller
queries that have resulted in the selection of at least one Web icons indicate a low level of interactivity, while larger icons
search result are identified by the small globe/Web icon with indicate a high level of interactivity.
an associated label (See Figure 1 icon (2)). The label dis-
plays the actual query text. If a query or result-selection Context Sliders
has a comment associated with it, the associated icon is aug- At the bottom of the interface there are two sliders. One
mented with a small user image. Comments can come in slider represents time while the other slider represents query
the form of answers, suggestions, tags, descriptions, general similarity. Users can manipulate the sliders to adjust the
comments/remarks, etc. Queries with comments are shown set of prime queries and to filter these queries. For example,
in Figure 1 icons (3) and (4). users can adjust the time slider to go back in time and display
queries submitted during different time periods. Thus rather
Search Histories than simply displaying queries submitted now (i.e. in the last
Clicking on the query icons/labels opens an information win- couple of hours), users can view queries submitted over the
dow/bubble (Figure 2), showing the query, the time the query entire day, yesterday, the last few days, last week, etc.
was last executed and a link to execute the query in ques-
tion. If the query lead to a result-selection the information The same principle applies to the query similarity slider.
window also displays the most recent result-selections. Fur- However, instead of time, the query similarity slider filters
thermore, if the query has any comments associated with it, by query term overlap. When a user accesses the application,
a link to view these comments is also shown (Figure 3 illus- the system automatically calculates the similarity between
trates the comments facility). Users can choose to go directly the user’s queries and all other queries in the dataset. Mov-
3 ing the query similar slider, changes the similarity threshold
The work presented in this paper builds upon earlier work pre-
sented in [4]. We have extended our previous work by: (1) enhanc- and thus filters queries from the prime set. In the following
ing the interface component of the SearchBrowser application, (2) section we describe our evaluation of SearchBrowser.
by adding a comments feature in which users can add comments to
4
queries and (3) by carrying out an initial user evaluation designed Note that if a user chooses to re-execute a query they will receive
to test the basic utility of the interface. a set of results from the standard Google search engine.
3
query for each6 . This resulted in 444 generated queries which
were then used as a basis for the prime dataset7 . For each
query, we extracted the associated WikiMapia entry, gener-
ated a random date and latitude/longitude within the cho-
sen time period and given location boundary, i.e. central
Dublin)8 . Any URLs associated with the WikiMapia entry
were used as the result-selection(s) and if the entry had tags
associated with it, we used the corresponding title/name as
the comment. The outcome was a set of time-stamped, geo-
coded, query, comment and result-selection data.
Participants & Methodology
20 participants took part in the study, 18 male and 2 female.
The participants comprised a mix of computer science staff
and post-graduate students from UCD, ranging in age be-
tween 25 and 40 (average: 30, standard deviation: 4.23).
85% of users had some previous Mobile Internet experience,
Figure 2. Clicking on a query opens an information window which
shows the query, when the query was last executed, result-selections
but most of these users (approximately 60%) accessed the
associated with the query and a link to view comments. Mobile Internet on an infrequent basis.
The participants carried out the experiment online using a
standard Web browser. The Web browser emulated the Search-
Browser interface by using similar screen real-estate to an
iPhone (320 x 480 pixels). Participants were asked to (1)
to familiarise themselves with the interface for the first few
minutes of the experiment and (2) to formulate and submit
five queries of their own using the interface. We informed
participants that the queries were open-ended, however, we
did ask participants to bear in mind that the interface is de-
Figure 3. Choosing to view query comments shows a list of comments signed for mobile devices and as such would be used while
with information about when the comment was added.
on-the-go. When generating their queries, we asked partic-
ipants to try to think of things they might need/like to find
EVALUATING SEARCHBROWSER out if mobile and in the location presented on the map. Be-
We had two main aims in carrying out an evaluation of the fore they were exposed to the interface the participants were
SearchBrowser interface. First, we wanted to assess the ef- presented with a description of the various features of the
fectiveness of the interface, focusing on key features of the interface. At the end of the evaluation, users were presented
interface and their usefulness. Second, we wanted take the with a post-task questionnaire designed to measure their sub-
first steps to investigate the potential of the new interface to jective reactions to the interface9 .
encourage discovery of new interesting content.
Usage Results
Dataset
6
To demonstrate the range of functionalities supported in the Each participant was presented with the same list of WikiMapia
SearchBrowser application, we needed a source of queries, entries and participants were instructed to generate queries for as
many entries as possible out of the list of > 200 entries.
comments and result-selections as the basis of our dataset. 7
We are aware that artificially generating queries in this manner
To generate the seed queries, we manually extracted > 200 is a limitation of this study. However, without deploying Search-
entries from the online WikiMapia service, focusing on en- Browser, it is impossible to obtain a realistic source of queries. For
tries with a latitude/longitude in the central Dublin, Ireland. users to evaluate the application we needed to be able to show
WikiMapia5 is a Web 2.0 application designed to encourage them a sample of queries, comments and result-selections and
WikiMapia offers a rich set of geo-coded data as the basis.
users to describe the world. WikiMapia allows users to mark 8
A note on random location values: in a realistic setting it is un-
areas on a Google map and describe those areas using titles, likely that a user will always submit a query related to an exact
descriptions, tags, categories, images and links to external physical location. For example, if the user wants to know where
URLs. Given that each entry in WikiMapia includes rich they can find a coffee shop on a particular street, the query could be
descriptive information, along with an original creation date generated anywhere along that street or within close vicinity to that
and a physical latitude/longitude value, it provided a good street. Therefore, we opted to generate a random latitude/longitude
within close vicinity to the actual latitude/longitude value.
basis to generate seed user queries for our evaluation. 9
This study allowed us to gather interesting feedback about the in-
terface and it’s overall usability. We are aware that the evaluation is
To generate realistic queries we then asked 3 different users limited in that is does not take place in a mobile setting, however,
to view the list of WikiMapia entries and to formulate a we feel that the evaluation in it’s current form still yields some
interesting results and represents an important first step in our on-
5
http://www.wikimapia.org going work in the mobile search space.
4
In this section we focus on the quantitative results by ex- Measure #Q # Qn #C # Cqn # CM
ploring the user interactions with the map-based interface as Total 297 126 76 57 23
well as general usage statistics. Mean (per-user) 14.9 6.3 3.8 2.9 1.2
Min 5 5 0 0 0
Interactions with the Map Interface Max 84 13 9 6 14
Using click-thru and mouse-over data we were able to anal- SD 17.4 1.9 2.6 2 3.1
yse what features of the map and user interface the partici- # Users 20 20 16 16 7
pants interacted with. Although the level and type of interac- % Users 100 100 80 80 35
tion with the SearchBrowser interface is likely to be different
in a real mobile setting, we were still interested in examin- Table 1. Basic usage statistics, where Q: queries, Qn : newly submitted
participant queries, C: click-thrus, Cqn : click-thrus generated from
ing interactions with the map so that we could gather some newly submitted participant queries, CM : comments.
insights into which features of the current user interface par-
ticipants were drawn too. Overall we found a high degree
of interactivity from end users. All users interacted with the IBM Computer Usability Satisfaction Questionnaires12 . We
map based interface using both zoom and drag functions to also included some more detailed usability and user-acceptance
navigate. All users clicked on either a query or result se- questions focusing on key aspects of the SearchBrowser in-
lection marker and opened an information window bubble. terface. Participants rated their agreement with various state-
We found that 95% of users clicked on the query markers ments on a 7-point anchored likert scale13 , with a rating
while 75% of users clicked on the result-selection markers. of 1 indicating “strongly disagree”, a rating of 7 indicating
We also found a high degree of interactivity with the vari- “strongly agree”, while a rating of 4 indicates “neutral”. The
ous markers/query icons, with mouseover events tracked for survey questions fell into three categories: (1) overall satis-
the vast majority of users. Thus, users did interact with the faction, (2) application features and (3) user interface (UI).
queries and past result-selections of other users. A full list of questions can be found in the appendix.
We found that most users selected search results within the Overall Satisfaction
SearchBrowser application. However, only 10% of users Q M1 SD M2 M3 Frequency Count
chose to click on a URL in the result-selection bubbles, thus 1 2 3 4 5 6 7
indicating a low level of interactivity with the past result- 1 4.50 1.47 5 5 1 0 5 2 7 4 1
selections of other users. Our later analyses indicate that 2 5.85 1.35 6 7 0 1 0 2 3 6 8
poor search results may have been the main cause for such 3 5.55 1.28 6 5 0 1 0 2 6 6 5
a low level of interaction. We also found that 70% of users 4 5.85 1.14 6 7 0 0 0 3 5 4 8
chose to view the comments of other users, but less than 50% 5 3.70 1.66 4 4 1 5 3 6 1 3 1
chose to generate comments of their own10 . 6 4.00 2.00 4 4 2 3 4 4 2 1 4
7 5.65 1.27 6 6 0 1 0 2 4 8 5
Search Usage 8 5.30 1.42 5 5 0 1 0 5 6 2 6
The results so far demonstrate that from an interactivity stand-
point, all users engaged with the SearchBrowser interface. Table 2. Results for the user satisfaction section of the survey. Q is the
Table 1 presents some basic usage statistics. The participants question number, M1 is the mean, M2 is the median, M3 is the mode
generated almost 300 queries, 126 of which were newly gen- and SD is standard deviation.
erated queries (i.e. submitted via the search box and not
through interactions with queries presented on the map inter- The list of satisfaction questions can be found in Table 7 in
face). Interestingly we find that 45% of all newly submitted the Appendix. Overall, the participants’ subjective assess-
queries by participants lead to at least one result-selection. ment of satisfaction with the application was positive, with
This represents a significant increase on the success rates an average response of 5.05. Participants found the applica-
found in [6] in which only 11% of mobile queries lead to the tion easy to use (q2=5.85) and easy to learn (q4=5.85). They
selection of at least one search result. found performing tasks to be straightforward (q7=5.65) and
in general felt that they could imagine using the application
Questionnaire Results while mobile (q8=5.3). However, users were unbiased in
In this section we examine the participants’ subjective reac- their rating of statement 6 regarding expected functions and
tions to SearchBrowser. At the end of the evaluation, users capabilities, and we found the general satisfaction rating as-
were presented with a post-evaluation survey. The survey signed by users was more neutral (q1=4.5). We attribute this
was designed using a combination of questions from well- to one key issue: users found it somewhat difficult to find
established usability questionnaires such as QUIS11 and the the information they needed/wanted (q5=3.7). The goal of
10
this evaluation was not to assess the search result quality, but
In most social websites, the majority of users don’t actively partic- rather the interfaces effectiveness and in this evaluation we
ipate in the generation of new content. In an analysis from Yahoo!
Groups, 1% of users actively create new content, 10% of users ac- were limited by the underlying search engine. We used the
tively contribute to such content (e.g. replying to a blog post) while Google search API for the search component of the applica-
100% of users benefit by reading/viewing the content generated by tion. We attempted to localise the queries by appending the
the others. See http://www.elatable.com/blog/?p=5.
11 12
Questionnaire for User Interface Satisfaction: http://drjim.0catch.com/usabqtr.pdf
13
http://hcibib.org/perlman/question.cgi?form=QUIS http://en.wikipedia.org/wiki/Likert scale
5
terms Dublin and Ireland to participant queries before issu- quality of the search results presented to users. It is likely
ing them to Google. However, one of the main comments by that the ratings for such features would increase if the search
participants was that the search results were not as localised results returned improved.
as they expected/wanted.
The comments feature resulted in a relatively neutral rating
Application Features
on average (q14=3.95, q15=4.10). In fact we find that par-
ticipants were quite divided in their opinion on the useful-
Q M1 SD M2 M3 Frequency Count
ness of the comments feature. For example, when asked if
1 2 3 4 5 6 7
the comments feature helped them to learn more about the
1 4.25 1.48 4.0 4 0 3 2 8 3 2 2 query, we find that 10 users agreed, 3 users were unbiased
2 4.40 1.54 4.5 5 0 2 5 3 5 3 2 and a further 7 users disagreed (q15). Interestingly we found
3 4.30 1.66 5.0 5 1 3 2 3 6 4 1 that users were more in agreement that they added comments
4 3.75 1.77 3.5 6 2 4 4 2 3 5 0 to their own queries (q19=3.45), rather than adding com-
5 3.55 1.70 3.5 5 3 3 4 3 4 3 0 ments to other people’s queries (q18=1.55). After examining
6 4.20 1.94 5.0 6 2 4 1 2 4 6 1 the remarks of participants about the comments feature, we
7 5.25 1.59 5.5 7 0 2 1 2 5 5 5 found that some users were not clear on what constitutes a
8 5.85 0.99 6.0 6 0 0 0 2 5 7 6 comment. This is something we look at improving in the
9 5.40 1.67 6.0 7 0 2 1 2 4 4 7 SocialSearchBrowser application.
10 5.60 1.19 5.5 5 0 0 1 2 7 4 6
11 3.80 1.67 4.0 5 2 4 2 3 6 3 0 User ratings for the two slider features were generally posi-
12 4.45 1.39 5.0 5 1 0 4 4 7 3 1 tive. We found that 12 users (60%) assigned a positive rat-
13 4.40 1.19 4.5 5 0 0 6 4 7 2 1 ing when asked if the time slider is useful, while 13 of the
14 3.95 1.90 4.0 4 3 2 2 6 2 3 2 users (65%) liked being able to filter queries based on time.
15 4.10 1.74 4.5 5 2 2 3 3 7 1 2 Users found the time slider more intuitive and as such inter-
16 3.55 1.85 4.0 2 3 5 1 4 4 2 1 acted with the time slider more frequently. Reaction to the
17 3.60 1.98 4.5 5 5 3 0 2 8 1 1 query similarity slider was less positive overall. For exam-
18 1.55 0.94 1.0 1 13 5 0 2 0 0 0 ple, users were quite neutral when asked if the query simi-
19 3.45 2.14 4.0 1 7 1 1 3 4 3 1 larity slider was useful (q23=4.15), however they were quite
20 4.85 1.95 5.0 7 0 3 4 1 4 1 7 positive when asked if they liked being able to filter queries
21 5.85 1.23 6.0 7 0 0 1 2 4 5 8 based on query similarity (q25=5.05). Interestingly, we find
22 5.10 2.22 6.0 7 1 4 1 1 0 5 8 that when we examine the frequency count for each of the
23 4.15 1.93 4.5 2 1 5 2 2 5 2 3 7 ratings assigned to the slider questions, the most popular
24 4.90 1.92 5.0 7 2 0 3 2 4 4 5 rating is strongly agree (score of 7), indicating that the users
25 5.05 1.70 5.0 7 0 2 1 5 5 0 7 who did like the slider features found them very useful.
Table 3. Results for the features section of the survey. Overall the SearchBrowser features were well-received by
participants, with the queries and time slider features rated
The list of feature questions can be found in Table 5 in the most positively out of the five feature sets. The results in-
Appendix. We found the majority of users were almost un- dicate that with some straightforward improvements, the re-
biased in their responses to the first set of questions regard- maining features (result-selections, comments and query sim-
ing the query feature. For example, we found that in gen- ilarity slider) could become more effective.
eral participants didn’t find that they interacted with queries
frequently (q5=3.55) and they were unsure as to whether
other people’s queries helped them form their own queries User Interface
(q6=4.2). However, users did rate statements 7, 8 and 9 pos- The list of user interface questions can be found in Table
itively, indicating that the queries provided an understand- 6 in the Appendix. Most of the participants were satisfied
ing of the type of information that is relevant to the location. with the interface (q1=5.05), found the interface pleasant
Users liked the ability to browse other user queries. Further- (q2=5.9), intuitive (q17=5.9) and liked interacting with the
more, they thought it was an interesting way to discover new interface (q3=5.75). Users also found the interface easy to
information (q8=5.85) and it helped them learn about other interact with (q14=6.2). Furthermore, users were able to eas-
people in the area (q9=5.4). One of the main aims of the ily explore the various features of the map (q16=5.9) indicat-
evaluation was to assess whether users liked the exploratory ing that perhaps such an interface would work well as a in-
interface provided by SearchBrowser and these initial results formation discovery tool in the mobile space. Users noticed
indicate that this may be the case. the queries on the map (q4=6.45), enjoyed the icons used to
represent queries (q5=5.45) and were somewhat positive as
Although participants found the ability to view result-selections to the intuitiveness of the query icons (q6=4.85).
useful (q11=5.60), they found they did not interact frequently
with the result-selections of others (q11=3.8) and were neu- When examining the two sliders, we find that users rated
tral in their opinion as to whether the result-selections of the time slider more highly, indicating that they noticed the
other users provided them with additional information about time slider (q7=6.45), they found it intuitive (q8=6.3) and
the query (q12=4.45). We attribute this finding to the poor they liked the time slider (q9=6.3). The ratings assigned to
6
similar statements for the query similarity slider, although that is location and time specific and is influenced by social
positive, leaned more towards an unbiased rating. As men- context [5]. Although existing search giants attempt to pro-
tioned in previous sections, we included the time and query vide some solutions — for example, Google’s mobile search
similarity sliders in the SearchBrowser application so that facility utilizes a users default location in order to contextual
users could quickly and easily filter the set of queries dis- search results14 — these solutions don’t go far enough.
played on the map. However, even with such features, we
find at times that the interface became cluttered with infor- One of the unique features of the SearchBrowser interface
mation (q15=4.4) thus making it more difficult to read the is that it provides a comments facility which allows users to
information presented (q13=4.95). add comments, tags, answers and suggestions to the queries
submitted by other users. The key idea behind this facility
Overall we found the response to the user interface by partic- is that it allows users to provide helpful information to assist
ipants was very positive, with the majority of users assigning other users with their information needs, thus embracing the
top marks to the vast majority of statements, thus indicating social side to mobile search. Although the comments fea-
that the current SearchBrowser interface design is both us- ture represented a simply first step at utilizing people-power
able and aesthetically pleasing. to enhance the search experience of mobile users, we be-
lieve that there are a number of opportunities in this research
Q M1 SD M2 M3 Frequency Count space. In particular, we think that there is great potential
1 2 3 4 5 6 7 in utilising a users social network as a source of valuable
1 5.05 1.76 5.5 6 1 1 3 0 5 6 4 query answers, comments, etc. Furthermore, incorporating
2 5.90 1.41 6.5 7 0 1 0 2 4 3 10 a users social network into the mobile interface would al-
3 5.75 1.16 6.0 6 0 0 1 2 4 7 6 low some novel and interesting filtering methods based on
4 6.45 1.00 7.0 7 0 0 1 0 1 5 13 ‘friend’ queries. Thus we have developed a prototype called
5 5.45 2.09 6.5 7 2 0 3 0 2 3 10 SocialSearchBrowser which allows users to execute queries
6 4.85 2.41 6.0 7 2 4 1 1 1 2 9 in various physical locations but also enables friends of the
7 6.45 1.19 7.0 7 0 1 0 0 1 4 14 current user to answer these queries in real-time. In the
8 6.30 1.22 7.0 7 0 1 0 0 2 5 12 following section we describe SocialSearchBrowser in more
9 5.40 2.04 6.5 7 1 1 3 2 0 3 10 detail.
10 4.65 2.52 6.0 7 5 1 0 2 1 4 7
11 4.65 2.11 5.0 7 3 1 1 3 4 3 5
SOCIALSEARCHBROWSER
12 4.60 2.04 5.0 7 3 0 2 4 4 2 5
Human beings, by their nature are social creatures. We live
13 4.95 1.99 6.0 6 1 3 1 2 2 6 5
by communicating with others, building relationships and
14 6.20 0.89 6.0 7 0 0 0 1 3 7 9
forming friendships. In fact, many people view the mobile
15 4.40 2.14 4.5 7 3 2 1 4 2 4 4
phone as a social communications device, that is, a device
16 5.90 0.97 6.0 6 0 0 0 2 4 8 6
which can be used to stay in contact with friends and family
17 5.90 0.97 6.0 6 0 0 0 2 4 8 6
[13]. Online social networking sites such as Facebook and
Table 4. Results for the user interface section of the survey.
MySpace have experienced a huge increase in usage in re-
cent times, with more and more users seeking novel ways of
As well as asking users to rate their perceptions of the Search- interacting with their friends and family15 . And in the near
Browser application on the 7-point likert scale, we also asked future it is likely that mobile phones will be used as the first
users some more general freeform questions. 90% of users port of call in accessing these online social networks.
said they would use the SearchBrowser application if the ser-
vice was easily/readily available. When asked under what The SocialSearchBrowser is made up of two components.
circumstances would they use such an application, partici- The first component is a map-based interface that works in a
pants submitted a range of responses including, if there were similar way to the previously discussed SearchBrowser ap-
in an unknown physical place (e.g. a new city), to find in- plication. The second component is a Facebook application.
formation about local services/products, to keep up-to-date The interface consists of a text box that allows users to issue
with current events and finally to find directional/travel-related queries, a small map centered at the user’s current physical
information. Interestingly, users also commented on the so- location which displays all queries executed in that location
cial aspect of the application, indicating that the social side and three sliders at the bottom of the interface for filtering
to the SearchBrowser application could be very useful for the set of queries displayed (See Figure 4). We have intro-
query recommendations. duced a new social slider which allows users to show queries
submitted by everyone or to display only queries submitted
Overall the results of the evaluation were positive. The Search- by friends. Manipulating the social slider changes the level
Browser study represented an important first step in evaluat- of friendship threshold and as such updates the queries dis-
ing this type of interface and it provided us with some valu- played on the map. The premise behind this slider is that
able feedback regarding the interface components and the users are likely to be interested in the queries and interac-
supported interactions. However, the evaluation results also 14
http://www.google.com/m
encouraged us to re-think some elements of the prototype. 15
The latest statistics from Facebook highlight that there is cur-
Furthermore, results from a recent diary study of mobile in- rently 150 million active users worldwide (Jan 2009). See:
formation needs indicate that mobile users seek fresh content http://www.facebook.com/press/info.php?statistics
7
tions their friends have participated in.
The Facebook application comprises of an information page
showing all queries submitted through the SocialSearchBrowser
interface. The information page lists the query submitted,
the name of the user who submitted the query, the location
of the user and a timestamp indicating when the query was
submitted (See Figure 5). Clicking on the query opens a
more detailed information page (See Figure 6). The de-
tailed page shows relevant query details but also displays
a Google map of where the user was at the time the query
was executed. It also shows a list of any answers/comments
submitted for the query and a form for entering new an-
swers/comments. In this way Facebook users can see what
queries their friends have executed on the go, where and
when their friends executed these queries and any answers
provided to these queries. To envisage how the SocialSearch- Figure 6. Facebook application showing the answer details page.
Browser would work, imagine the following scenarios:
in this location. Amy doesn’t see any queries on the map re-
lated to tapas so she decides to submit her own query. Thus,
Amy enters the query “good tapas” via the SocialSearch-
Browser interface. Amy is presented with a localized list of
Web search results for her query. At the same time a notifi-
cation is sent to Amy’s facebook friends indicating that Amy
is in Barcelona and that she’d like some help with a query. A
few minutes later Amy is alerted that one of her friends has
submitted an answer to her query. Amy returns to the map,
clicks on her query and is shown the answer(s) submitted by
her friend(s). Perfect, now Amy knows exactly where to go
for great tapas!
David is in the middle of Dublin city center, sipping on a
coffee and is thinking about what to do this weekend with
friends. He takes out his iPhone, opens the browser and
Figure 4. The new SocialSearchBrowser interfaces which allows mobile connects to SocialSearchBrowser. David is presented with a
users to filter the set of queries to display queries entered by friends. map centered at his current physical location. David is able
to see straight away that other users have entered queries
Amy is wandering around Plaza de Catalunya in Barcelona like “coffee to go” and “salsa classes” in this location. David
as part of her weekend away in Spain. She wants to know decides he wants to explore what else other people in this lo-
where she can find a nice restaurant that serves tapas but she cation have been interested in. He moves the temporal slider
wants to avoid touristy places. Amy takes out her iPhone, towards the earlier marker and the map is updated with lots
opens the browser and connects to SocialSearchBrowser. Amy of different queries entered in this location. David see’s lots
is presented with a map centered at her current location. The of queries related to comedy events. David then uses the so-
map displays other queries and user interactions that have cial filter to show only queries submitted by his friends and
taken place in her current location. Amy is able to get an he notices that his friend Tony was looking for tickets to see
idea of the types of needs that arose from other mobile users a comedy show last week. David decides to call Tony to see
if he’d like to go to a comedy show this weekend.
Ideally, when queries are submitted via SocialSearchBrowser,
a user’s friends will be online and will be able to offer help
immediately. This scenario could also be extended to allow
anyone to answer queries, but in this case, answers generated
by close friends of the user would be rated higher. Other so-
cial factors could also be explored. For example, in the cur-
rent prototype we include a social slider for filtering queries
so that only queries generated by friends are displayed. We
could also investigate filtering friend locations, i.e the set of
Figure 5. Facebook application showing the initial query list page. locations where your friends executed queries, etc.
8
Thus, SocialSearchBrowser provides an alternative means 2. D. Arter, G. Buchanan, R. Harper, and M. Jones.
of mobile search and information discovery, taking into ac- Incidental information and mobile search. In
count key mobile contexts such as location and time, while Proceedings of MobileHCI ’07, pages 413–420. ACM,
exploiting the social context of search. Users are encour- 2007.
aged to discover new, interesting content and perhaps new,
interesting places. The new prototype utilizes a users so- 3. S. Bao, G. Xue, X. Wu, Y. Yu, B. Fei, and Z. Su.
cial network to improve the information access experience, Optimizing web search using social annotations. In
allowing friends to provide helpful information through real- Proceedings of WWW ’07, pages 501–510. ACM, 2007.
time query answering. Furthermore the application enables 4. K. Church and B. Smyth. Who, what, where & when: a
a new form of social discovery by allowing friends to share new approach to mobile search. In Proceedings of IUI
queries and online interactions while mobile. ’08, pages 309–312. ACM, 2008.
CONCLUSIONS 5. K. Church and B. Smyth. Understanding the intent
Mobile information access is challenging, particularly from behind mobile information needs. In Proceedings of
a search perspective. In this paper we described two new IUI ’09, 2009. To Appear.
interfaces for improving mobile search and discovery. The 6. K. Church, B. Smyth, K. Bradley, and P. Cotter. A large
first interface, SearchBrowser, presents users with histori- scale study of european mobile search behaviour. In
cal query, comment and result-selection data on a rich map- Proceedings of MobileHCI ’08, pages 13–22. ACM,
based. The application takes important mobile contexts into 2008.
account such as location and time. Results of an initial user
trial were positive and demonstrated that the current inter- 7. J. Freyne, R. Farzan, P. Brusilovsky, B. Smyth, and
face design is easy to use, easy to learn and aesthetically M. Coyle. Collecting community wisdom: integrating
pleasing to end-users. Based on user feedback from this social search & social navigation. In Proceedings of
evaluation and results of a recent diary study of mobile infor- IUI ’07, pages 52–61. ACM, 2007.
mation needs, we developed an extended proof-of-concept
prototype, that explores the social context of mobile search. 8. J. Golbeck and M. M. Wasser. Socialbrowsing:
SocialSearchBrowser is an innovative interface that incorpo- integrating social networks and web browsing. In CHI
rates mobile contexts with social networking capabilities to ’07: extended abstracts on Human factors in
improve the search and information discovery experience of computing systems, pages 2381–2386. ACM, 2007.
mobile subscribers. SocialSearchBrowser allows friends to 9. M. A. Hearst. Tilebars: visualization of term
provide help to mobile users in the form of answers, com- distribution information in full text information access.
ments, suggestions, links and tags, through a Facebook ap- In Proceedings of CHI ’95, pages 59–66, 1995.
plication. Furthermore, the interface incorporates a social
filter which enables mobile users to filter the set of queries 10. M. Jones, G. Buchanan, R. Harper, and P.-L. Xech.
displayed to show only friend queries, helping to visualise Questions not answers: a novel mobile search
friend queries and interactions. technique. In Proceedings of CHI ’07, pages 155–158.
ACM, 2007.
We are currently investigating a number of different areas
relating to the prototype. Firstly, we are in the process of 11. A. K. Karlson, G. G. Robertson, D. C. Robbins, M. P.
implementing a fully functional working prototype which Czerwinski, and G. R. Smith. Fathumb: a facet-based
we plan to test and evaluate with real users in a live mobile interface for mobile search. In Proceedings of CHI ’06,
field study. We have also identified a number of interest- pages 711–720. ACM, 2006.
ing future directions that explore the social context of search 12. A. Mislove, K. P. Gummadi, and P. Druschel.
in more detail. We’d like to consider other social filtering Exploiting social networks for internet search. In
approaches. For example, we could show popular friend HotNets ’06: Proceedings of the 5th Workshop on Hot
queries and allow users to visualize the most common lo- Topics in Networks, 2006.
cations in which friends have had previous mobile informa-
tion needs. We could also exploit social networks to provide 13. S. Plant. On the mobile: the effects of mobile
personalized query and location recommendations. For ex- telephones on social and individual life. See
ample, a mobile user might be interested to learn about a http://www.motorola.com/mot/doc/0/234 MotDoc.pdf,
new street in their city where a number of their friends have 2001.
submitted previous mobile queries. We’re also interested
14. R. W. White, B. Kules, S. Drucker, and M. Schraefel.
in the automatic identification of ‘close friends’ vs ‘not so
Supporting exploratory search: Special issue.
close friends’, based on facebook activity, presence in mo-
Communications of the ACM, 49(4), 2006.
bile phone contacts and other informative resources.
15. K.-P. Yee, K. Swearingen, K. Li, and M. Hearst.
REFERENCES Faceted metadata for image search and browsing. In
1. O. Alonso, R. Baeza-Yates, and M. Gertz. Exploratory Proceedings of CHI ’03, pages 401–408. ACM, 2003.
search using timelines. In ACM SIGCHI 2007
Workshop on Exploratory Search and HCI, 2007.
9
No. Question Type
1 Overall, I am satisfied with the search LK
browser interface
2 The interface of the search browser appli- LK
cation was pleasant
3 I liked using the interface of the search LK
browser application
No. Question Type 4 I noticed the queries on the map LK
1 I found other people’s queries useful LK 5 I liked the query icons LK
2 I found other people’s queries informative LK 6 I found the query icons intuitive LK
3 I found other people’s queries intriguing LK 7 I noticed the time slider LK
4 I found other people’s queries distracting LK 8 I found the time slider intuitive LK
5 I interacted with other people’s queries LK 9 I liked the time slider LK
6 I found that other people’s queries helped LK 10 I noticed the query similarity slider LK
me form my own queries 11 I found the query similarity slider intu- LK
7 Many of the queries displayed helped me to LK itive
understand the sort of information that was 12 I liked the query similarity slider LK
relevant to the location being browsed 13 I was able to easily read information on LK
8 The ability to browse other people’s queries LK the interface
is an interesting way to discover new infor- 14 It was easy to interact with the interface LK
mation. 15 The organization of information on the LK
9 The queries helped me to learn about other LK map was clear
people in the area, their needs and prefer- 16 I was able to easily explore the various LK
ences map features
10 The ability to view other people’s past LK 17 The interface was intuitive LK
result-selections is useful
11 I interacted with other people’s past result- LK Table 6. List of interface questions presented to end-users. (LK = 7-
selections point likert scale).
12 The result-selection feature provided me LK
with additional information about the query
13 The result-selection feature helped me find LK No. Question Type
answers to the queries 1 Overall, I am satisfied with the search LK
14 I found the comments feature useful LK browser application
15 The comments associated with a query LK 2 It was simple to use the application LK
helped me learn more about the query 3 I felt comfortable using the application LK
16 The comments associated with a query LK 4 It was easy to learn to use the application LK
helped me find answers to the query 5 It was easy to find the information I LK
17 I viewed other people’s comments LK needed
18 I added comments to other people’s queries LK 6 The application had all the functions and LK
19 I added comments to my own queries LK capabilities I expect it to have
20 I found the time slider useful LK 7 Performing tasks is straightforward LK
21 I interacted with the time slider LK 8 I could imagine using this type of appli- LK
22 I liked being able to filter the queries dis- LK cation when out and about.
played on the map based on time 9 Leaving cost aside, would you use the Y/N
23 I found the preference slider useful LK search browser application if the service
24 I interacted with the preference slider LK was easily/readily available?
25 I liked being able to filter the queries dis- LK 10 What circumstances do you think you F
played on the map based on query similar- might find the search browser application
ity useful?
11 What did you like about the search F
Table 5. List of features questions presented to end-users. (LK = 7- browser application?
point likert scale). 12 What if anything did you find frustrating F
or unappealing about the search browser
application?
13 How could we make the search browser F
application more useful for you?
Table 7. List of general satisfaction questions presented to end-users.
(LK = 7-point likert scale, F = freeform, Y/N = Yes/No).
10