RETHINKING MOBILE SEARCH QUERIES USING CONTEXT
Noran Hassan, Sherif G. Aly, Ahmed Rafea and Ahmed Soliman
Department of Computer Science and Engineering, The American University in Cairo, Cairo, Egypt
Keywords: Mobile, Context Awareness, Searching, Ranking.
Abstract: In some search domains, user context is often related to user search intent or preference. Such context
however is rudimentarily used in search queries. Mobile devices, through their sensors and data content
however have an abundance of information that can indicate the user context. Such context information can
be used to influence, filter or re-rank search results to better match user needs. In this, paper we present
some of the previous work where user context was used to improve the mobile search experience, as well as
work that attempted to understand how user context is related to search intent. Our findings show that
previous work primarily focused user location, with great neglect to other types of context that may be of
great significance to search results. The work we present in this paper attempts to understand how a wide
range of types of context influence a particular search domain. The types of context we study include
location, time, day, weather and movement. We analyze how such context information can influence search
needs when searching for restaurants and movies. Our analysis is based on a survey that was taken by 179
respondents. We describe the survey, how it was authored and reviewed, and then analyze the results and
findings as deals with the most important contextual pieces of information that could be used to enhance the
mobile search experience.
1 INTRODUCTION
Search engines are expected to respond to query
requests with results that are characterized with high
precision and recall. However, many users continue
to be challenged with formulating proper search
query terms that match their true intent, often
impairing the ability of search engines to properly
return and rank appropriate results in harmony with
user intent. Furthermore, the large number of results
makes them practically impossible to be fully
browsed by any user.
The use of search engines however is no longer
limited to personal computers. Ubiquitous devices,
such as smart phones, are becoming more common
as a new channel for search. As presented in
(Church et al., 2009), 67% of people’s information
needs are delivered while they are mobile. Studies in
(Kamvar et al., 2009) showed that search patterns
initiated on phones vary significantly in query length
and topic diversity. Mobile phone searches tend to
have shorter queries that encompass a narrower
range of topics compared to other resourceful
devices. Moreover, limited screen space and
mobility makes it more difficult to browse through a
large number of results.
Location information has long been mistakenly
perceived as being the most important piece of
contextual information relevant to search queries, as
evident by most commercial search engines.
However, we hypothesize that further contextual
information obtained from the ambience of mobile
devices can be used to enhance search queries issues
from such devices, yet little is known in literature
about the type and usefulness of such information
for performing web searching under such conditions.
In this paper, we have a clear objective, which is
to understand the type of contextual pieces of
information that are most relevant in enhancing
search queries initiated from mobile devices. For
this purpose, we start off by presenting related work
that helps in understanding the relationship between
user context and mobile search intent. We then
describe a study that we conducted to identify and
weight the types of contextual information that
would be relevant in performing mobile
entertainment-related search queries, in which we
limit the entertainment domain to searching for
restaurants and movies.
237
Hassan N., G. Aly S., Rafea A. and Soliman A..
RETHINKING MOBILE SEARCH QUERIES USING CONTEXT.
DOI: 10.5220/0003804302370244
In Proceedings of the 2nd International Conference on Pervasive Embedded Computing and Communication Systems (PECCS-2012), pages 237-244
ISBN: 978-989-8565-00-6
Copyright
c
 2012 SCITEPRESS (Science and Technology Publications, Lda.)

2 RELATED WORK
In this section, we present two types of related work.
The first type is work which focuses on
understanding the relationship between user context
and search intent, especially in mobile
environments. The second type of work is that which
attempts to improve the mobile search experience by
using information about user context.
Many attempts have been made to define and
classify contextual information. More notably,
Ranganathan in (Ranaganathan et al., 2003) tried to
summarize and categorize the types of context to
include: Physical, environmental, informational,
personal, social, application, and system. Even
though categorization attempts are many, but to
mention a few examples in Ranaganathan’s
categories, physical context includes location and
time. Environmental context includes weather, light
and sound levels. Informational context includes
stock quotes, and sports scores. Personal context
includes health, mood, schedule, and activity. Social
context includes group activity, social relationships,
and collocation. Application context includes email
received, and websites visited. Finally, system
context includes network traffic, and the status of
printers.
2.1 Understanding Search Behavior
and Intent
To improve the mobile search experience, it is
important to understand mobile search patterns and
how they are different or similar to desktop search.
According to (Church et al., 2009), to understand
search behavior, the main two approaches have been
to understand what people search for, and why they
search for it. Analyzing search behavior and what
people search for involves the study of things like
the length of queries and their topics. Analyzing
search intent (why people are searching) can be
categorized to the following classifications (Church
et al., 2009): navigational (reach a site),
informational (learn about a topic or answer a
question), geographical (search for a location),
transactional (web-mediated activity e.g. games,
downloads), and personal information management
(find personal information).
Collecting data about mobile usage is a challenge
due to the difficulty of discreetly collecting data.
There have been several methods for collecting such
data such as interviews (Arter et al., 2007), log data
analysis, observation, diary study, or a combination
of two or more methods. Log data has the benefit of
providing realistic and large-scale data of usage.
However, in (Amin et al., 2009), it is argued that log
analysis is not sufficient as location-based needs are
not always explicitly expressed in the queries.
In (Church et al., 2009), results revealed that
67% of people’s information needs are while they
are mobile (i.e. away from home or work computer).
58% of these needs are informational, 31%
geographical and 11% is personal information
management. Another observation is that 75% of
geographical entries were generated while the user is
mobile. Most of the geographical information needs
were temporally dependent (i.e. only relevant at a
particular point in time), even though most queries
did not include explicit temporal cues. It was noted
that the user’s current activity has an important
factor that triggers the user to perform a search.
Regarding the topics searched, the most dominating
topics were 20% travel and commuting searches,
16% general information searches, 13% local
services searches, and another 13% were
entertainment searches.
In (Amin et al., 2009), a hybrid approach is used
to collect information on users’ search activities.
They data they collected included the search event
log (queries, clicks, etc.), location at the time of
search, a diary where participants logged more
details about their context at time of search such as
who they were with, their current activity, the
importance of this task, and the success of their task.
A post-study interview was conducted with the
participants where they clarified any unclear or
missing entries. The results showed that the main
domains of interest were stores (27%), foods and
drinks (24.5%), entertainment (14%), news (12%),
and transport (10%). Over 86% of tasks are goal-
oriented whether the goal is finding a specific piece
of information or to make a higher level decision.
Queries were analyzed with respect to the spatial,
temporal and social context. Regarding spatial
context, results indicated that most searches are
performed either at home, work, while commuting in
between, or at regularly visited places. With the
exception of weekends, participants followed regular
routes and visited regular places. Another discovery
was that the target places are more often related to
their regularly visited places rather than their current
location. The common places were: at
family/friends' home (6.5%), public places (8.5%),
at work (12%), on the move (20%), and at home
(53%). Temporally, the results showed that 66.1% of
searches were related to a spontaneous need (e.g.
need a number to make a call). 21.5% were less
urgent and related to something planned that day
PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems
238





(e.g. book a table in a restaurant). The remaining
12.4% were exploratory tasks and not urgent at all.
Socially, 76.1% percent of searches were performed
with the company of others. Tasks performed alone
were driven by necessity (e.g. weather inquiry,
directions), while tasks performed with people
around were mostly driven by conversations with
people. The reasons participants chose a place for
their desired local service were the availability of a
particular product or service (24%),
recommendation by others (16%), a decision made
with others, e.g. friends, family (13%), closeness to
current location (8%), whether it’s a favorite place to
them (8%).
2.2 Context-enhanced Mobile Search
In this section, we list related work that helps
enhance queries issued from mobile devices. Once
the relationship between context and a search query
is understood, the search experience can be
improved by concatenating context words to a
query. This disambiguates the query, refines the
search and thus returns fewer, more specific results.
Alternatively, context words can be used to filter
results and return the results most relevant to this
context. (Kraft et al., 2006) present 3 algorithms to
implement contextual search:
i. Query Rewriting: In this approach, context
words are simply appended to the original query.
This makes the search more specific and therefore
returns a smaller response set. This poses the danger
of low recall because if the search becomes too
specific, the engine may return too few results or
none at all.
ii. Rank Biasing: This approach also appends
context words to the original query but only as
optional ranking terms. Optional ranking terms may
also be given weights to capture the importance of
each. The advantage of this approach is that it
guarantees the same level of recall as the original
query. However, this requires a modified engine that
can accept such a complex weighted query.
iii. Iterative Filtering Meta-search: In this approach,
candidate context words are generated. Query
templates are also selected which are templates that
define the query and context combinations that will
be used. The template generates multiple queries,
each submitted to the search engine. The results
from the multiple queries are aggregated and re-
ranked.
There have been several attempts to use context in
desktop search to improve the user’s search
experience by trying to resolve any ambiguities via
analyzing the user’s context. The types of context
used in such applications include the previous
queries submitted by the user (Cao, 2009), the URLs
recently clicked by the user, the recently browsed
documents (Cao and Shen, 2009) (Rahurkar and
Cucerzan, 2008) and, the contents of the documents
on the desktop (Chirita et al., 2006), and the
activities the user is engaged in standard applications
(Leake and Scherle, 2001).
3 METHODOLOGY
After surveying many related work dealing with
contextual searching, and after making two major
conclusions: (1) That location is still perceived to be
the most important contextual information used in
searching and (2) That there is little understanding
about other types of context that may help mobile
contextual searching, we decided to conduct a study
that better understands the relevance of other types
of contextual information that may enhance the
results of mobile search queries.
In our methodology, we hypothesize that context,
beyond location,should be a reflection of user search
intent. We eventually designed and conducted a
survey on 179 respondents to understand mobile
search usage patterns and how the aforementioned
context influences their preferred search results.
Table 1 shows the type of context information we
are interested in and the sources that they can be
extracted from.
Our main hypothesis is that user context
especially physical, environmental and personal
context such as time of day, day of week, weather,
location, calendar events, and whether or not the
user is on-the-go does influence the user’s
preferences. We test this hypothesis with special
focus on the entertainment search domain. This main
hypothesis breaks down to the following specific
hypotheses:
 Time-of-day influences user preferences.
 Current location influences user preferences.
 Mobility influences user preferences.
 Weather influences user preferences.
 Day-of-week influences user preferences.The
type of event users are going out for influences their
preferences.
We built a survey with questions that aim to validate
these hypotheses. We concentrated on
restaurant/food search as the domain that the
questions tackle. In this section, we discuss the main
RETHINKING MOBILE SEARCH QUERIES USING CONTEXT
239

and minor goals of the survey and its structures.
Then we discuss how the survey was authored, then
reviewed by a focus group, and shared with the
respondents.
Table 1: Contextual information and sources of extraction.
Context Information How to extract?
Time of day Phone time settings
Day of week Date and time
Location GPS
Weather
Weather web service and
GPS
Movement Accelerometer
Type of events,
Meetings/deadlines
Calendar entries
3.1 Survey Goals
To figure out which types of context affect what
types of preferences when picking a restaurant to eat
at, we authored a survey to find out to what extent
respondents agree with our hypotheses. The main
goal of the survey is to find out how people’s
preferences regarding the restaurant/type of food
they want change as their context changes. A
secondary goal of the survey is to understand how
people use internet services from mobile phones.
The questions are tailored such that the answers
allow us to compare the importance of each type of
context and in what way it affects respondents’
choice of restaurant. Some questions were intended
to understand the respondent’s profile (e.g. age,
gender, profession). Another set of questions were
posed to understand how fluent the respondents are
with technology and how reliant they are on it. The
main bulk of the survey was meant to understand
how the user preferences are influenced by
contextual situations.
3.2 Survey Structure
The survey is composed of the following sections:
 Demographic and General Information: (e.g.
country, age, gender, profession).
 Technology Use: Question technology fluency.
 Search Use: Question whether and how search
engines are used to find information about
entertainment.
 Time: Question how the preferences are
influenced by the time of day.
 Location and Movement: Question how
preferences regarding location of restaurants
changes with mobility.
 Weather: Question how weather conditions
imply the attributes of the place they would like to
go to (e.g. outdoor vs. indoor seating)
 Calendar: Question how the type of
gathering/meeting influences the attributes of place
they would like to go to.
3.3 Survey Reviewing
The survey went through two cycles of review
before being published. In these review cycles, the
survey was shared with a focus group, a group of
people who were asked to both fill the survey and
provide comments and feedback about how they
found the questions. They were asked to identify any
flaws in the survey. We provided them with some
hints and guidelines while asking them not to limit
themselves to these guidelines. They were impelled
to point out ambiguous questions, redundant
questions, words that are hard to understand,
questions that seem to direct the user to give a
certain answer, or questions that seem invasive or
offensive. They were also encouraged to look at the
multiple choice answers for any missing possible
answers or overlapping answers. We also asked
them to point out if they found the survey to be too
long. The size of the focus group was five people in
each round.
The survey evolved as we made changes in
response to the focus group’s comments. Such
comments included pointing out difficult terms,
missing options in multiple-choice answers,
ambiguous words, and how they felt at certain points
(e.g. annoyed after a series of similar questions).
Some questions which were related and had the
same set of multiple choice answers were aggregated
in one tabular question. In some cases, answers with
numerical ranges, such as commute time that a
respondent finds reasonable, needed to be
aggregated into fewer bigger ranges to become more
meaningful.
3.4 Survey Sharing
To ease the distribution of the survey and the
consolidation of results, it was created as an online
survey. The tool used for this purpose was Survey
Gizmo. The snowballing approach was used to share
the survey. The survey was sent in mailing lists and
shared via Facebook. Friends with large networks
were messaged directly and asked to share the
survey with their friends.
PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems
240





4 FINDINGS
The survey was taken by 179 respondents. The
survey results and analysis is discussed in this
section. We discuss several aspects of the results; the
respondents’ demographics, their comfort with
technology and search, and how the context
influences their preferences about the type restaurant
or place that they would like to go to. The types of
context discussed are time of day, day of week,
weather, location and type of event.
It was also deemed important to be able to give a
value for the strength of the relationship between a
certain context and a user preference pair. This
allows us to filter out pairs with weak relationships
and rank them by significance. For this purpose we
calculate the confidence of the relationship between
each pair. The confidence of each relationship is a
value between 0 and 1 that is basically the
percentage of agreement. There are two question
types used in this case; 5-likert scale and checkbox-
based. How the confidence is calculated in each type
is explained below:
i. Five poInt Likert Scale Questions: Answer
options are: strongly agree (SA), agree(A), neutral
(N), disagree (D), strongly disagree(SD). The
confidence of the relationship presented in such
questions is calculated as:
(#SA + #A) / (# respondents) (1)
ii. Checkbox-based Questions: In these questions,
respondents check a checkbox if they agree that the
two items in question are correlated. The confidence
calculated as:
(# checks) / (# of respondents) (2)
4.1 Demographics
Around two-thirds of respondents are female.
Almost 90% of respondents are in the 18-35 age
group. There are no respondents in the 55 and above
age group and only couple in the under-18 age
group. Since the survey was spawned in Egypt, it is
not surprising that 82% of responses came from
Egypt. 11% of responses came from the United
States. Due to the nature of the network of survey
authors of there is a considerable percentage, 29%,
of respondents are from the
technology/programming domain. Other respondents
are students (13%) or came from research (11%),
construction (8%), education (12%), and other
domains.
In correlation with the respondents’ age groups,
99% of respondents have at least a bachelor’s
degree, with more than half of those with a post-
graduate degree too. Putting that in mind, we can
conclude that respondents whose profession is
student, are in fact mostly Master’s or PhD students.
4.2 Technology and Search Use
The vast majority of respondents have mobile
phones, are comfortable using the computer, and use
the internet on a daily basis. Regarding mobile
usage, we notice some important trends:
 73% of respondents use the internet from their
mobile phone at least a few times per month for one
purpose or another (figure 1a).
 23% use internet from the mobile phone all the
time, regardless of whether they have access to a PC
(Figure 1b). This confirms that internet usage from
mobile phones is becoming more common and more
of a main internet channel rather than just a backup
channel.
The top eight purposes mobile internet is used for
are: Email (88%), social networking (67%), instant
messaging and chatting (48%), maps (42%),
checking weather conditions (41%), seeking
information such as word definition, movie reviews
(37%), news and sport scores (36%), and searching
for local entertainment such as movie theatres and
restaurants (27%). 18% of respondents use mobile
internet to search for local services (e.g. pharmacy,
bookstore), 18% watch or download music and/or
videos, 10% download wallpapers and ringtones, 7%
play online games, and 5% shop online.

Figure 1: Mobile internet usage. a) Frequency of usage. b)
When do people mobile internet.
Only 28% of respondents have searched for
RETHINKING MOBILE SEARCH QUERIES USING CONTEXT
241

entertainment from their mobile phones, however,
83% have searched for entertainment from a
computer. With the growing prevalence of mobile
internet usage, we can forecast that searching for
entertainment from mobile phones will become
more common with time. Another interesting trend
is that 68% of outings are planned on the same day.
This supports the assumption that the timely context
is in fact relevant to the user’s searches when it
comes to searching for something like entertainment.
See Figure 2 to see how much time in advance
respondents have planned for outings before.
Figure 2: How much time in advance outings are planned.
4.3 Context Influence on Preferences
Focusing on restaurants, we posed questions that
aim to validate some hypotheses regarding how we
suspect that people’s choice of restaurants would
differ as their context changes. Such attributes
include the restaurant location, working hours,
menu, presence of shaded parking, view, WiFi
availability, ambience, indoor/outdoor seating, etc.
In this section, we discuss the context and restaurant
attribute pairs that gained more than 50% of
respondents’ agreement.
4.3.1 Time of Day
Table 2 shows the relationship between time-of-day
context and the restaurant attributes and the
confidence of each relationship.
Table 2: Time-of-day context, related restaurant attributes,
and the confidence of the relationship.
Context Restaurant attribute Confidence
Any time Open at current time 0.85
Time is before noon Breakfast menu 0.60
4.3.2 Day and Time of Week
75% of respondents agreed that their restaurant
choices differ between weekends and weekdays.
When it comes to restaurant location, on weekdays,
around 80% of respondents are willing to spend up
to 30 minutes maximum on the road to get to the
restaurant. On weekends, respondents are willing to
spend more time on the road, with around 80%
willing to spend up to one hour.
Table 3 shows the relationship between the
restaurant attribute and the day of week (weekend
vs. weekday) with distinction between mornings and
evenings, along with the confidence. It was also
concluded that:
 The majority of respondents want coffee and
light sandwiches on weekday mornings while
wanting a more sophisticated breakfast menu on
weekend mornings with a nice view to enjoy.
 On weekend evenings, respondents also want to
enjoy a nice view, have a decent meal, discover a
new place or cuisine, and have fun at a place that
offers fun activities such as pottery or karaoke.
 On weekday evenings, some people also want a
decent meal but would probably opt for fast food or
take-away food. This makes sense since people tend
to have less time for leisure on weekdays and opt for
the faster option.
Table 3: Day-and-time-of-week context, related restaurant
attributes, and the confidence of the relationship.
Day/time Restaurant Attribute Confidence
Weekday
evening
Take-away 0.65
Fast food 0.61
Dining (i.e. main meal) 0.57
Weekday
morning
Coffee 0.63
Light sandwiches e.g. cold cuts 0.54
Weekend
evening
A new cuisine/place to discover 0.81
Dining (i.e. main meal) 0.8
Nice view 0.66
Special activities e.g.
pottery/karaoke
0.59
Weekend
morning
Breakfast menu 0.7
Nice view 0.66
4.3.3 Weather
The attributes that respondents preferred in a
restaurant at changed depending on the weather
conditions (pleasant, cold, hot, raining, windy and
humid). If the weather is pleasant, 92% would rather
enjoy it and therefore prefer a place with an outdoor
seating area. In unpleasant weather conditions, the
preference is more towards indoor seating areas:
cold (66%), raining (64%), hot (56%), windy (55%)
and humid (51%). If the weather is hot, 53% want to
keep their cars cool in a shaded parking lot. If it’s
raining, 60% would want a nearby location. That’s
probably because people prefer to drive less in the
rain.
PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems
242





4.3.4 Location and Movement
Figure 3 demonstrates the percentage of respondents
who have indicated certain preferences regarding
restaurant location. Please note that respondents
were allowed to make more than one selection and
that is why the percentages do not add up to one
hundred in the indicated graph. Most notably,
whether stationary or on the move, and 41% would
rather have that location close to their homes.
4.3.5 Type of Event
The restaurant attributes that respondents
need/prefer are also related to the type of
gathering/event they are going out for. For example,
after work or school, most respondents want take-
away food. This is similar to weekday evenings. If
the meeting is for studying or working, respondents
need a quiet place that serves coffee and has WiFi. If
the event is a casual meeting or a special event e.g.
birthday, respondents are interested in dining, a nice
view to enjoy, and are willing to discover a new
place/cuisine. If it’s a special event, respondents
would like a place that has special activities such as
pottery or karaoke. See table 5 for details.
Table 4: Type of gathering/event, related restaurant
attributes and the confidence of the relationship.
Type of Event Restaurant Attribute Confidence
After school/work
Take-away (not necessarily fast
food)
0.59
Casual meeting
w/friends
A new cuisine/place to discover 0.62
Coffee 0.59
Dining (i.e. main meal) 0.54
Nice view 0.53
Breakfast menu 0.51
Special event e.g.
birthday
Nice view 0.62
Dining (i.e. main meal) 0.62
Special activities e.g.
pottery/karaoke
0.56
A new cuisine/place to discover 0.54
Studying/
working
Coffee 0.64
WIFI 0.64
Quiet environment 0.61
Figure 3: Preferred restaurant location. 
5 CONTEXT-SENSITIVE
SEARCH SYSTEM
We envision the use of our results in a context-
sensitive system. The system would be triggered if a
certain context situation is satisfied when the user
issues a query in a particular topic. A mapping of a
context situation to context-words would have to be
created as shown in Table 6 with confidence values
based on our results. The corresponding
context-
words would be looked up and used to refine the
search and/or filter/re-rank the results. The context-
word weights can be further fine-tuned based on the
user profile (e.g. age, gender, profession).
Ultimately, the confidence should evolve based on
the user’s previous searches and clicks and therefore
making the weights more personalized.
6 CONCLUSIONS
In this paper, we presented previous work in
understanding the relationship between context and
search needs/intents. We then presented our work in
developing and reviewing a survey intended to
understand how a wide range of context situations
influence user preferences in restaurant search. Our
results lay a foundation of understanding for
subsequently building a system of mobile search
queries that is sensitive to user context, and in line
with user intent. Future work includes developing a
system that builds on the conclusions of this analysis
to produce a context-sensitive mobile search system.
Further work would involve generalizing these
conclusions by analyzing how context influences
search in other domains.
Table 5: Sample context to context-word mappings.
Context situation
Context words (may be
expanded)
Confidence
Morning “Breakfast [menu]” 0.76
Pleasant weather
“outdoor seating [area]”,
“open-air”
0.92
Hot weather
“shaded parking”,
“underground parking”
0.53
Weekday evening
“Take-away”, “to go” 0.65
“Fast food” 0.61
Weekday morning “Coffee” 0.63
Weekend evening “Dining”, “main platters” 0.8
Studying/working “WIFI”, “internet” 0.64
RETHINKING MOBILE SEARCH QUERIES USING CONTEXT
243

ACKNOWLEDGEMENTS
This work was funded in part by a Google Research
Award. The authors would like to extend an
acknowledgement to Dr. Mohamed Elfeky from
Google Mountain View for his contributions.

REFERENCES
Amin, A., Townsend, S., Ossenbruggen, J., and Hardman,
L., (2009, August). Fancy a Drink in Canary Wharf?:
A User Study on Location-Based Mobile Search.
Proceedings of the 12th IFIP TC 13 International
Conference on Human-Computer Interaction: Part I.
Arter, D., Buchanan, G., Jones, M., Harper, R., (2007).
Incidental information and mobile search. Proceedings
of the 9th international conference on Human
computer interaction with mobile devices and services,
p.413-420.
Cao, H., Jiang, D., Pei, J., Chen, E., and Li, H., (2009).
Towards Context-Aware Search by Learning A Very
Large Variable Length Hidden Markov Model from
Search Logs. Proceedings of the 18th international
conference on World wide web, 191-200.
Cao, H., Hu, D., Shen, D., Jiang, D., Sun, J., Chen, E. and
Yang, Q., (2009). Context-Aware Query
Classification. Proceedings of the 32nd international
ACM SIGIR conference on Research and development
in information retrieval, 3-10.
Chirita, P., Firan, C., and Nejdl, W., (2006). Summarizing
Local Context to Personalize Global Web Search.
Proceedings of the 15th ACM international conference
on Information and knowledge management, 287–
296.
Church, K., and Smyth, B., (2009a). Understanding the
intent behind mobile information needs. Proceedings
of the 13th international conference on Intelligent user
interfaces.
Church, K., and Smyth, B., (2009b). Who, What, Where &
When: A New Approach to Mobile Search.
Proceedings of the 13th international conference on
Intelligent user interfaces, 309-312.
Kamvar, M., Kellar, M., Patel, R., and Xu, Y., (2009,
April). Computers and iphones and mobile phones, oh
my!: a logs-based comparison of search users on
different devices. Proceedings of the 18th
international conference on World wide web.
Kraft, R., Chang, C., Maghoul, F., and Kumar, R., (2006).
Searching with Context. Proceedings of the 15th
international conference on World Wide Web, 477 –
486.
Leake, D., and Scherle, R., (2001). Towards Context-
Based Search Engine Selection. Proceedings of the 6th
international conference on ntelligent user interfaces,
109 – 112.
Rahurkar, M., and Cucerzan, S., (2008). Using the Current
Browsing Context to Improve Search Relevance.
Proceeding of the 17th ACM conference on
Information and knowledge management, 1493-1494.
Ranaganathan, A., and Campbell, R., (2003). A
Middleware for Context-Aware Agents in Ubiquitous
Computing Environments. IFIP International
Federation for Information Processing, Lecture Notes
in Computer Science, 2672/2003, 998.
PECCS 2012 - International Conference on Pervasive and Embedded Computing and Communication Systems
244