An In-Situ Study of Mobile App& Mobile Search Interactions

Juan Pablo CarrascalPompeu Fabra University

jp.carrascal@upf.edu

Karen ChurchYahoo Labs

kchurch@yahoo-inc.com

ABSTRACTWhen trying to satisfy an information need, smartphone usersfrequently transition from mobile search engines to mobileapps and vice versa. However, little is known about the na-ture of these transitions nor how mobile search and mobileapps interact. We report on a 2-week, mixed-method studyinvolving 18 Android users, where we collected real-worldmobile search and mobile app usage data alongside subjectiveinsights on why certain interactions between apps and mobilesearch occur. Our results show that when people engage withmobile search they tend to interact with more mobile appsand for longer durations. We found that certain categories ofapps are used more intensely alongside mobile search. Fur-thermore we found differences in app usage before and af-ter mobile search and show how mobile app interactions canboth prompt mobile search and enable users to take action.We conclude with a discussion on what these patterns meanfor mobile search and how we might design mobile searchexperiences that take these app interactions into account.

Author KeywordsMobile Search; Mobile Apps; User Study

General TermsHuman Factors

ACM Classification KeywordsH.5.2 Information interfaces and presentation: Miscellaneous

INTRODUCTIONPeople adopt a range of ingenious methods when trying to sat-isfy their daily information needs but the Internet has becomethe dominant means of finding information for smartphoneusers in particular [6]. A recent comScore report1 shows thatmobile devices accounted for 55% of Internet usage in theUnited States in January 2014. Native mobile apps made upthe majority of that traffic (85%), while the remaining 15%came from mobile browsers. With a wealth of informationseeking apps at their fingertips, mobile users can now issue1Mobile apps overtake PC Internet usage in U.S., See: cnn-mon.ie/1cfHe0Z

Permission to make digital or hard copies of all or part of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full cita-tion on the first page. Copyrights for components of this work owned by others thanACM must be honored. Abstracting with credit is permitted. To copy otherwise, or re-publish, to post on servers or to redistribute to lists, requires prior specific permissionand/or a fee. Request permissions from Permissions@acm.org.CHI 2015, April 18 - 23 2015, Seoul, Republic of Korea Copyright is held bythe owner/author(s). Publication rights licensed to ACM. ACM 978-1-4503-3145-6/15/04$15.00 http://dx.doi.org/10.1145/2702123.2702486

queries directly from within these apps. For example, Yelpcan be used to search for local businesses, Google Maps canbe used to find directions, while IMDB can be used to look formovie reviews. Thus the boundaries between mobile searchengines and mobile apps have become blurred.

Smartphone users sometimes switch between mobile searchand other mobile apps to satisfy their information needs.Brown et al. [4] describes one such switching interactionin which mobile search and mobile maps are used inter-changeably — with search being used to find informationabout certain places and maps being used to get directionsto those places. It’s likely that such switching interactionsoccur across a diverse range of everyday apps, i.e. beyondjust maps applications. One of the goals of this work is tounderstand such switching behaviors.

Past research has also shown that mobile search can be trig-gered by our conversations and our surroundings [1, 7, 9, 24].Furthermore, the results of mobile search can feed into nextactions and plans [4]. Thus it’s likely that some of these mo-bile search triggers and actions may relate to or be supportedby interactions with mobile apps. For example, a text mes-sage from a friend may prompt a search or after searching fora local restaurant, a person may call to make a dinner reser-vation. The goal of this paper is to explore these interactionsand to investigate if and how mobile search and app usagerelates to mobile search triggers and actions.

While patterns of mobile app usage and mobile search havebeen studied separately [3, 18, 11, 2, 8, 13, 14, 26, 27, 22],the relationships between them have not been examined in-depth. A detailed study of in-situ app and search engineuse will provide insights into how we can support more en-riching cross app and search engine experiences, and assistmobile users in completing their daily information seekingtasks. To shed light on these interactions we conducted a 2-week mixed-method study involving 18 Android users. Wecombined interviews, a daily online diary and real-world mo-bile usage data to gather rich insights into mobile informationseeking behaviors. Over the 2-week period we collected ap-proximately 54,000 mobile app launches, over 840 mobilesearch sessions and over 500 diary entries. Our results high-light that when users engage in mobile search activity, theyinteract more with other mobile apps and for longer dura-tions. Furthermore certain categories of mobile apps are moreprevalent in sessions involving search, compared to sessionswith no search activity. Our core contributions are:

• We define and introduce cross app and search interactionsas an important and challenging research opportunity in thefield of mobile HCI and mobile information retrieval.

• We analyze and characterize app and search interactionsfrom a unique perspective, combining real-world in-situsearch and app usage with qualitative insights.

• We discuss what our findings mean for future mobile infor-mation seeking services.

RELATED WORKSignificant effort has been placed on understanding people’sdaily information needs and how those needs are addressed[10, 6]. Studies focused on mobile information needs high-light that 72% of mobile needs are influenced by explicit con-textual factors including activity, location, time and conver-sation [21]. The act of fulfilling information needs is also af-fected by contexts [5], with location being particularly influ-ential [15]. Furthermore, the internet is a prominent means ofaddressing mobile information needs [21] in particular amongexperienced mobile Internet users [12].

As mobile device use has become more prominent, there hasbeen a growing interest in understanding how and why peopleaccess and consume online content via their mobile phones.Some of this research focuses on behaviors of early-stagesmartphone devices, e.g. [16, 9], while other work attempts toshed light on modern device use [4]. Qualitative studies havehighlighted that the majority of mobile Web access occursin stationary, familiar environments like home and work [20,7]. The main motivations for general mobile Web usage areawareness and staying up-to-date [23], while mobile searchmotivations are more inclined to relate to curiosity [7]. Mo-bile search has been shown to be a social act, often sparkedby conversations [9] and conducted in the presence of oth-ers [7]. Similar social influences have been found for localmobile search [1, 24]. While these qualitative studies haverevealed interesting aspects of mobile search behavior, theirreliance on interviews, diaries and survey data mean they lackdetails on actual search interactions.

Other research focused on analyzing actual Web usage pat-terns of real mobile phone users using datasets from commer-cial search engines like Google as well as operator-specificsearch services [2, 8, 13, 14, 26, 27, 22]. Overall these stud-ies highlight that mobile queries tend to be shorter than theirdesktop counterparts, users tend to submit fewer queries persession compared to desktop users and that adult related con-tent is very popular. Recent studies show signs of evolvingpatterns, highlighting that the increased popularity of modernsmartphones is having an impact on search behavior. For ex-ample, mobile queries are becoming less homogenous, mo-bile query length is steadily increasing and interest in localcontent is rising. These studies have helped shape our un-derstanding about how mobile users seek online information,however their focus on logs-based analysis means they lackinsights about why certain behaviors have emerged.

As the adoption of smartphones has increased dramatically inthe last few years, the number and diversity of native mobileapplications — so called apps — have also risen. This hasgiven rise to a number of studies aimed at understanding howpeople engage with native mobile apps. Bohmer et al. [3]studied app use among over 4,100 Android phone users and

highlighted the prevalence of communications related appsthroughout the day, with certain categories of apps being usedmore often at certain times, e.g. Clock in the morning. Tossellet al. [25] looked at Web use on smartphones and found thatnative internet applications like Mail, Facebook, Maps andWeather are visited twice as much as browsers. These studiesprovide us with a concrete understanding on the types of mo-bile app interactions among modern smartphone users, how-ever they suffer from a lack of understanding on the specificsof what’s being done within the applications themselves.

There are of course a few exceptions. For example, McGregoret al. [18] use a video-data collection method to record 100days of iPhone use and highlight 3 distinct usage patterns:micro-breaks, digital knitting and reading. Lee et al. [17]explored if and how app usage relates to smartphone overuseand addiction in college studies. And most recently Ferreriaet al. [11] looked at app micro-usage and found that 40% ofall app launches last less than 15 seconds and such short in-teraction happen mostly when a person is at home and alone.

Work by Brown et al. [4] is particularly relevant. Using inno-vative video data collection of everyday smartphone use, theyfocus on what prompts the use of particular applications atspecific times or in specific situations, focusing on web searchand maps use. They highlight ‘occasioned search’, that issearch triggered by the environment or local events. Our goalis to expand upon these insights to explore the unique inter-action between mobile search and general mobile app usage,specifically how search is influenced by app usage and howapp interactions influence mobile search actions. Brown etal. also highlight the use of the mobile Web and mobile mapsinterchangeably in information searches. In this paper we ex-amine such interactions but explore switching between searchand a range of mobile apps that smartphones users engagewith daily, i.e. beyond just maps applications.

In summary, past work has explored mobile search and mo-bile app usage primarily in isolation. Thus questions abouthow mobile search and app usage is interlinked remainlargely unanswered. In this paper we present an in-depth in-vestigation on the relationship between mobile apps and mo-bile search, exploring the triggers, actions and app interac-tions in and around mobile search. Furthermore, we studythis behavior from a unique perspective, combining both real-world in-situ search and app usage with qualitative insights.

STUDY METHODWe conducted a 2-week in-situ field study in June and July2014 where we collected both objective and subjective data.Objective data was collected via two apps installed on par-ticipant’s smartphones, which tracked their actual mobile appand mobile search usage. Subjective data was gathered at thestart and end of the study in initial and final in-person inter-views as well as daily via an online diary where participantsreviewed and described specifics of their mobile searches.

ParticipantsWe recruited 18 participants (10 male; 8 female) from 8 dif-ferent cities in the Greater San Francisco Bay Area using a

professional recruiting agency. Participants ranged in age be-tween 18-48, with an mean age of 32.7 (sd: 9.8) and had adiverse set of occupations including students, administrativeassistants, social workers, managers, chemists, homemakersand construction workers. Their education levels ranged fromhigh school to college degree. All participants were activeusers of mobile search engines like Google and Yahoo. Allof them owned Android smartphones as their main commu-nication device, the majority of which were Samsung Galaxyphones (N=14). Participants were compensated for the time.

ProcedureThe study was conducted between June and July 2014 andwas comprised of three stages:

1. Initial interview and app installationThe initial interview was semi-structured and lasted from 30minutes to 1 hour, depending on the scope and diversity ofparticipant’s mobile search and app usage. The interviewscovered (a) their daily mobile device habits; (b) their generalmobile search engine use; (c) concrete examples of their mostrecent mobile search, focusing on aspects such as what trig-gered the search in the first place and any actions that tookplace as a result of the search; and (d) apps they use forsearching and situations in which they choose an app overa search engine for satisfying their needs. To conclude the in-terviews, we installed two custom apps on their smartphones:

AppLogger: a logging app that ran as an android service inthe background of the participants’ phone and kept track ofall device usage events including app launches, turning thescreen on and off, unlocking the phone, and accessing thehome screen. Each logged usage event includes an app orevent name, a timestamp, and the duration of the event.

MSearch: a mobile search app which embedded the searchfunctionality of a well known commercial search engine.This app collected time stamped search data including searchqueries and search interactions (i.e. clicks). Participantsmoved the app icon to their homescreen during the initial in-terview to ensure the search app was easily accessible.

2. Two-week mobile search and online diary studyParticipants were asked to use the MSearch app for all theirmobile search needs over the 2-week study period and to re-view their mobile searches in a Web-based online diary tool.All queries and clicks generated through our search app weresent to a server for processing. Queries and clicks weregrouped into sessions using a session delta of 5 minutes andthese search sessions formed the basis of the participants’ di-ary. The online diary was designed to capture the motivations,triggers and actions surrounding their mobile searches. Tominimize the time burden on participants, the diary presentedparticipants with a maximum of 3 of their mobile search ses-sions per day, which were selected at random. Below is a listof the 9 questions we asked for each of the 3 daily search ses-sions. In parenthesis we indicate if a question was multiplechoice (closed), or freeform text (open):

1. What information were you looking for? (open)2. What were you doing at the time of the search? (open)

3. Where were you at the time and did your location influencethe search? If so, how? (closed/open)

4. Who were you with at the time and did the people aroundyou influence the search? If so, how? (closed/open)

5. Did you share the information you found with other peo-ple? If so, how? (closed/open)

6. How important was it to find the information you searchedfor? (closed, 5-point Likert scale)

7. How urgent was it to find the information you searched for?(closed, 5-point Likert scale)

8. Could you find the information? If not, what alternativeapproaches did you try to find the information? (closed)

9. What did you do with the information you found during thesearch? Did you take any additional actions? (open)

Note that participants were encouraged to access and fill theironline diary each evening, a task which took approx. 5-10mins to complete. Participants were also sent a daily emailfor the duration of the study to remind them about the study.

3. Final interview and app removalAt the end of the 2-week study, participants attended a finalin-person interview. Prior to the interview, we reviewed theirmobile search and mobile app log data as well as their diaryentries to extract specific usage behaviors we wanted to fol-low up on. We probed participants about two or three of theirreported mobile searches and asked them more details regard-ing the triggers, actions and associated app interactions in andaround those searches. To conclude, both the AppLogger andMSearch apps were uninstalled from the participants phone.All in-person interviews were audio recorded and transcribed.

RESULTSQuantitative data from both mobile search and mobile appusage was analyzed per-session and in aggregate to under-stand the types and nature of mobile search interactions; thenature and duration of mobile app use; and the interaction be-tween mobile app and mobile search. Qualitative data fromthe transcribed interviews and the diary responses were an-alyzed using grounded theory-based affinity analysis, an ap-proach commonly used to organize and group large quantitiesof subjective data into a logical set of themes or categories.We extracted over 3,000 individual quotes from the qualita-tive data, which made up individual data items in our analy-sis. These data items were iteratively reviewed and groupedby the authors to find repeating themes across participants.

Basic Descriptive StatisticsThe results reported here are based on the mobile app andsearch logs gathered between 23rd June and 13th July 2014.Although the mobile search part of the study lasted 14 days,the AppLogger ran for an average of 16 days (sd : 2.57) dueto scheduling final participant interviews. Before exploringthe interactions between mobile apps and mobile search, wemuch first analyze app usage and mobile search separately.

Mobile App UsageOver 189K mobile phone usage events were logged over theentire study duration with an average of 10,515.3 events per

Table 1. App usage summary showing number of unique apps, app launches and average usage time of apps grouped by app categoryCategory # Launches % Launches # Unique Apps Avg Dur(secs) App ExamplesSocial networking 9714 17.98 17 107.6 Facebook, Twitter, Instagram, OkCupidBrowser & Search 7727 14.30 10 112.9 Chrome, Google Search, FirefoxSMS/Texting 5964 11.04 2 63.6 Build in MessagingPhone & Audio 5456 10.10 10 25.8 Dialer, Google Voice, Skype, VoxerEmail 3928 7.27 4 54.0 Gmail, Yahoo mail, SolMailGames 2780 5.15 73 264.1 Slotmania, CandyCrush, WordsContact 2574 4.76 4 53.0 Contacts, Mr. NumberMusic & Audio 2468 4.57 16 74.0 Pandora, iHeartRadio, SoundCloudPhotography 2172 4.02 19 36.4 Shutterfly, Gallery, Camera, PicCollageSystem & Settings 1904 3.52 12 29.5 Software updates, Android SettingsTools & Utilities 1777 3.29 45 30.8 Calculator, Clean Master, Media StorageInstant Messaging 1831 3.39 10 79.6 Snapchat, ChatOn, Kik, TrillianMedia & Video 1338 2.48 10 275.8 YouTbe, Netflix, BitTorrentProductivity 1169 2.16 24 61.4 Calendar, Quickoffice, LastPass, GnotesTravel & Local 1035 1.92 13 111.6 Maps, Muni Alerts, YelpEntertainment 370 0.68 19 239.7 Meme Generator, Series Guide, IMDBUnknown 346 0.64 23 142.5 Not applicableShopping & Retail 254 0.47 12 138.3 eBay, Groupon, Macy’s, LivingSocialFinance 252 0.47 15 58.1 Wells Fargo, Chase, PayPal, MintHealth & Fitness 228 0.42 10 156.7 S Health, Fitit, Push Ups, My Diet CoachEducation 181 0.34 1 22.2 SMCCLifestyle 169 0.31 12 65.8 dscout, 7-Eleven, AIDS WalkWeather 146 0.27 5 19.2 The Weather Channel, Weather WidgetNews & Magazines 124 0.23 7 73.0 Flipboard, yahoo, GlamlifeBusiness 71 0.13 7 112.5 VPN Client, CamCard, Job SearchPersonalization 28 0.05 7 129.9 Zedge, Live wallpaper, travel wallpaperSports 9 0.02 1 47.8 theSocreBooks & Reference 7 0.01 6 49.4 Dictionary, Wikipedia, Audible

participant (Mdn : 11165.5, sd : 6124.7). Of these, 54,022corresponded to app launches, with an average of 3,001.2 applaunches per participant (Mdn : 2867.5, sd : 2003.9). The re-maining events were device events such as turning the displayon/off, unlocking the phone and accessing the homescreen. Atotal of 394 unique mobile apps were launched across our 18participants during the study, with an average of 53.6 uniquemobile apps launched per participant (Mdn : 52, sd : 20.6).

While prior work shows that users spend an average of 1 hourusing mobile apps [3, 19], our participants spent significantlymore time. The average time our participants spent interact-ing with mobile apps is > 4.5 hours per day (Mdn : 248.2mins, sd : 173.3 mins). In contrast, Table 1 highlights thatindividual mobile app usage is often short, lasting an averageof just 90 seconds (Mdn : 17, sd : 241.5), slightly longerthan Bohmer et al.’s 71 second average. Overall we foundthat 73.5% of app uses were one minute or less in durationand approx. 40% of uses are 15 seconds or less which is inline with findings by Ferreria et al. [11].

To get a better sense of the types of apps used, we manu-ally classified the 394 unique apps into corresponding GooglePlay categories. We did this mapping by searching the Playstore based on app name. Similar to the Bohmer et al. studywe made some minor changes to the Google Play categoriza-tion. Specifically, we opted for one high-level games categoryinstead of multiple micro-categories of Games (e.g. arcade,brain & puzzle, etc.). We opted to have a separate Browsers &Search category and a separate Systems & Settings categoryfor handling the default Android settings apps. Finally weopted to break-out communications-related categories suchas Contacts, SMS/Texting and Instant Messaging. Table 1highlights that the top 3 categories are Social networking,

Browsers & Search Engines and SMS, while almost 55% ofapp usage in our study relates to Communications.

Table 2. Distribution of search topics across all 882 unique queriesTopic # Queries % QueriesEntertainment 144 16.3Trivia 144 16.3Local 133 15.1Shopping & Coupons 106 12.0Travel & Commuting 84 9.5Technology 49 5.6Health & Fitness 45 5.1General Information 42 4.8Cooking, recipes & ingredients 28 3.2Sport 24 2.7Auto 24 2.7Misc 17 1.9Search & Navigational 12 1.4Stocks & Finance 12 1.4News & Weather 10 1.1Employment 6 0.7Education 1 0.1

Mobile Search UsageParticipants issued 882 unique queries through the MSearchapp resulting in 2794 webpage visits. These webpage visitsinclude both click-throughs and follow-on links. This corre-sponds to an average of 50.1 unique mobile search queries(Mdn : 47.5, sd : 23.9) and 158 webpage visits (Mdn : 112,sd : 122.4) per participant over the 2 week study period.

A mobile search session is a sequence of queries and searchinteractions (i.e. clicking on the next page of results or on anindividual search result) issued by a single user within a smalltime period. Using a standard mobile search session delim-iter of 5 minutes [13], we identified a total of 843 search ses-sions. We found an average of 1.6 unique queries per session

(Mdn : 1, sd : 1.2). Approximately 78% of sessions resultedin at least one click-through (647 sessions) while the averagenumber of webpage visits (click-throughs and follow-ons) persession is 4.9 (Mdn : 3, sd : 5.7).

Compared to previous studies of mobile search patterns weobserve that our participants issued longer search queries.The average number of words per query is 3.55 (Mdn : 3,sd : 2.1), compared to 3.05 reported in a 2013 Bing study[22]. Finally to get a better sense of the types of queriesour participants issued, we manually classified all 882 uniquesearch queries into a set of search categories. Cohen’s kappawas used to measure intercoder reliability. After two itera-tions of independently coding 200 unique search queries anddiscussing any conflicts, a Cohen’s Kappa value of 0.74 wasreached. The remaining dataset was then divided, and eachauthor independently coded his or her part of the dataset. Theresulting classification is shown in Table 2. We found thatour participants predominately issued mobile search queriesrelated to Entertainment, Trivia, Local, Shopping & Couponsas well as Travel & Commuting.

The Online DiaryParticipants contributed a total of 535 diary entries describ-ing details of specific mobile searches. This equates to 29.7entries per user over the course of the study. Accordingto their answers, most of the searches were conducted athome (70.1%), followed by work/school (7.7%) or commut-ing (7.7%). Other locations mentioned included bars, cafesor restaurants, someone else’s home, the gym, shopping andrunning errands. In 26% of the cases, participants stated thattheir location at the time influenced their search.

In contrast to past research on mobile search, we foundthat participants reported that they were mostly alone whensearching (51.2%). The rest of the time they were with theirpartner or spouse (22.4%), relatives (12.9%), friends (7.3%),work colleagues or school mates (3.9%) or others (2.2%). In15.5% of the time, the person accompanying the participanthad an influence on searching.

Most of the time the information sought was found (80.6%).In 17.4% of cases the participants considered the informa-tion to be partially found. When participants couldn’t findthe information, they either did nothing (29.4%) or reliedon alternative methods including: asking somebody (23.5%),using other search engines (23.5%), reading the informa-tion somewhere—in a map, etc.—(2.9%), or something else(20.6%). In 17.4% of cases, the information found wasshared with somebody else. In the following section we beingour investigation of cross app and mobile search interactions.

Cross App and Mobile SearchTo explore the relationship between mobile search and mo-bile app usage we must first define cross app and search in-teractions. Figure 1 presents the sequence of actual mobileapp and search interactions for a 1 hour period (12-1pm) ona single day for a real participant in our study. It shows that auser’s day is typically comprised of a number of “Sessions”.Similar to Bohmer’s definition of an app chain [3], we definea session as a sequence of interactions that occur without the

device being in standby mode, i.e. the display switching off,for longer than 30 seconds. In other words, if the phonesdisplay is off for more than 30 seconds, all following app in-teractions belong to the next session. While if the phonesdisplay is off for 30 seconds or less, the following app inter-actions belong to the preceding session.

Some of these sessions involve the user turning their dis-play on without actually launching any apps. For exampleto check the time or to check for any missed notifications,both of which would be visible on their lock screen and/orhomescreen of the phone. In other sessions, the user launchesand interacts with specific mobile apps. Sometimes this maybe a single app launch, while in other cases this can involveopening a chain of apps. The diagram highlights that this par-ticular user interacts with her device regularly throughout thehour but more intensely in the first 30 minutes. She primarilyuses SMS but also plays a number of games (e.g. Pet Res-cue Saga), browses the Internet and interacts with Facebook.Thus we define sessions in which at least one mobile app waslaunched as “App Sessions”.

Aside from mobile app interactions, the figure also high-lights that the participant engaged with mobile search on twooccasions. First submitting the query ‘advance screeningssan francisco’ and visiting a website advancescreenings.com.Minutes later she submits the query ‘rotten tomatoes’ and vis-its another website. These queries (Q) and search clicks (C)are identified in bold in the figure. Thus some “App Sessions”involve the user engaging with mobile search, while other appsessions do not. We define these as AppSessionssearch andAppSessionsnonsearch respectively.

Note that this particular user interacts with SMS before, dur-ing and soon after her mobile searches. This implies that heroverarching task — deciding on a movie — involved crossapp and search interaction. Note that final interviews with theparticipant confirmed she went to a movie as a result of thissearch. It appears that the participant switched between mo-bile search and other mobile apps. Likewise given most ofthis switching occurs with SMS, it appears as though she wascommunicating with someone at the time. Thus it’s likely thatthe conversations within these interactions with SMS mayhave prompted the mobile search and in turn influenced theresulting action (e.g. making plans to go to the movies). Thegoal of this work is to understand more about the nuancesof these cross app and search interactions. In the followingsection we dive into our analysis.

Are Search and Non-Search Sessions Different?To investigate the mutual influence and interaction betweenmobile search and mobile app usage, we explore differencesbetween AppSessionssearch and AppSessionsnonsearch interms of app launches, unique apps used and session dura-tion. To do this we extracted a list of all App Sessions in ourdataset, that is sessions that include at least one launch of atleast one mobile app. We excluded the first app session ofevery participant from all subsequent analyses since it corre-sponds to the installation of the MSearch app which was doneduring the initial interview. We found a total of 12, 307 appsessions in our dataset. Our MSearch app was used in 913

Figure 1. Participant timeline for a 1 hour period of a single day showing sequences of mobile device interactions in the form of mobile app launches,home screen interactions and mobile searches.

Figure 2. (A) Number of app launches in AppSessionssearch vs. AppSessionsnonsearch; (B) Number of unique apps used in AppSessionssearchvs. AppSessionsnonsearch; (C) Session duration of AppSessionssearch vs. AppSessionsnonsearch

sessions, and in remaining 11, 394 it was not. We thereforesplit the dataset in two: 913 AppSessionssearch and 11,394AppSessionsnonsearch for comparison purposes.

Figure 2(A) shows the distribution of AppSessionssearch(red) and AppSessionsnonsearch (blue) grouped by the to-tal number of app launches. The figure suggests that appsessions involving mobile search include a higher numberof app usages compared to non-search sessions. We foundthat AppSessionsnonsearch include an average of 3.9 applaunches (Mdn : 2, sd : 6.9), while AppSessionssearch in-volve an average of 10.8 app launches (Mdn : 5, sd : 25.7).A non-paired Welch’s t-test confirmed that there is a signif-icant difference between the groups (t = 8.14, p < 0.01,Cohen’s d = 0.72). Note that Cohen’s d here measures themagnitude of the difference. Thus when mobile users engagewith search, the intensity of their app interactions (i.e. applaucnhes) increases. Figure 2(B) shows a similar pattern forthe number of unique apps used in search vs. non-search appsessions. That is when users engage in AppSessionssearch,they interact with a higher number of unique mobile apps.We found that AppSessionsnonsearch include an averageof 2.2 distinct mobile apps (Mdn : 1, sd : 2.2), whileAppSessionssearch involve using an average of 4.6 uniqueapps (Mdn : 3, sd : 3.9). This implies there is greater di-versity in their behaviors during search sessions, that is whenusers engage with search, they interact with a higher numberof different apps compared to when users are not engaged in

search. We tested this and again found this difference to besignificant (t = 18.68, p < 0.01, Cohen’s d = 1.18).

Finally we explored temporal differences in the form ofoverall session duration. Figure 2(C) highlights thatapp sessions involving search typically last longer thanapp sessions with no search. The average duration ofAppSessionssearch is 1377 seconds (Mdn : 372, sd :4940.9) compared to 348.2 seconds (Mdn : 52, sd : 1275.6)for AppSessionsnonsearch. This difference was also foundto be significant (t = 6.28, p < 0.01, Cohen’s d = 0.57)between the AppSessionssearch or AppSessionsnonsearchdatasets. These results indicate that when users engage withsearch, they engage for longer periods than when they arenot engaged with search. All in all these results indicatethat AppSessionssearch tend to be more app-intensive thanAppSessionsnonsearch on a number of levels. To investigatewhy these differences occur, we turn our analysis to app inter-actions within these sessions to see if certain apps or certaincategories of apps tend to be used more intensively in andaround mobile search activity.

How are Search and Non-Search Sessions Different?In this section we investigate differences in usage of dif-ferent categories of apps between AppSessionssearch andAppSessionsnonsearch. We opted to make these compar-isons at a category level as opposed to single app level fortwo reasons: (1) because of individual preferences, differ-ent participants used different apps from the same category

Table 3. T-test (non-paired) results for comparing launch frequencyand usage time of app categories between search and non-search ses-sions. These categories displayed significantly higher launch frequencyand significantly longer usage time in search sessions compared to non-search sessions.

Launch frequency Usage durationCategory t∗ Cohen’s d t∗ Cohen’s dBrowsers 15.10 1.29 23.19 1.60Email 3.80 0.16 8.80 0.57Entertainment 2.62 0.23 4.12 0.35Finance 1.98 0.10 2.02 0.10Games 5.15 0.42 6.06 0.44Media & Video 1.98 0.15 3.02 0.21Photography 2.22 0.10 2.84 0.14Shopping & Retail 2.60 0.20 3.90 0.23SMS/Texting 3.07 0.23 4.14 0.24Social networking 3.81 0.29 3.87 0.26Tools & Utilities 2.80 0.30 4.94 0.33Unknown 2.91 0.16 2.93 0.21Weather 2.87 0.13 3.39 0.17∗All coefficients are significant at p < .05.

(e.g. Gmail and the Android email client, the native Internetbrowser and Google Chrome); and (2) we wanted to limit theinfluence of apps that are only used by single participants.

To perform the comparisons, we extracted the frequency ofapp launches and the usage time of every category of appon a per-session basis. Then we compared the frequencyand duration of categories of apps in AppSessionssearchand AppSessionsnonsearch. Comparisons were assessedwith a non-paired Welch’s t-test. Table 3 highlights thatwe found a significantly higher launch frequency and signif-icantly longer usage time in AppSessionssearch comparedto AppSessionsnonsearch, for certain app categories. Cate-gories including Browsers, Email, SMS, Social Networking,Shopping & Retail and Entertainment were all used more in-tensively when people engaged with mobile search, both interms of app launches and duration of app usage. This impliesthat when users are in information seeking mode, they engagemore heavily with certain types of apps. To investigate fur-ther, we turn our analysis to the triggers of mobile search tosee if certain app interactions prompt users to search.

Triggers & App InteractionsThrough qualitative analysis of diary and interview data, weidentified 6 key triggers of mobile search. While understand-ing the prompts of mobile search is not new, our goal here isto shed light on the relationship between triggers and app in-teractions. The 6 triggers we identified were either external orinternal. External triggers are sensory stimuli, that is thingsyou can see, touch or hear. In contrast internal triggers areconnected with our thoughts, emotions, body, or pre-existinghabits. External triggers included (1) media, e.g. watching tv,listening to music, reading a newspaper; (2) conversations,face to face as well as as conversations across messaging,email and social networking app; (3) Tangible external trig-gers, that is noticing something material in their physical sur-roundings; (4) Activities & Events; (5) Physiological signalslike hungry and stress and finally (6) State of Mind, e.g. ran-dom thoughts.

While some of the so-called triggers are unlikely relatedto app usage, Media and Conversations are the exception.Given prevalence of entertainment and media related apps aswell as messaging apps, it’s likely that some of these app in-teractions prompted mobile searches. P11 for example dis-cussed this situation: “I was listening to the music in thereand I remember that I think that particular person was sup-posed to be on this podcast”; he then did a search for thepodcast in question. He was listening to music on his phoneat the time. Similarly P15 described a situation in which shewas texting a friend about flights which prompted a search,“Our conversation about her wanting to fly. She had lookedup some prices and I said ’that seems really high!’. I was like’let me check’”.

To understand more about how app interactions might triggermobile searches we extracted the first apps launched withina search session (excluding our search app) and found thatFacebook (8.3%), VEVO (4.8%), Calendar (3.2%), Messag-ing (3.1%) and Gmail (3%) were the top apps launched firstwithin search sessions. We decided to take a manual look atthe data to see if we could make any connections between thefirst app launched in a session and the query (mobile search)immediately succeeding that app interaction. P5, for exampleuses SeriesGuide which helps people manage (re)watchingtheir favorite TV shows. Looking at the data we found that P5issued a query ‘Friday night dinner tv wiki’ immediately afteran interaction with SeriesGuide. The topic of P5’s query afterinteracting with SeriesGuide, an entertainment app, was alsoentertainment related. P1 issued two media & music relatedqueries ‘video2mp3’ and ‘worldstar hiphop’ immediately af-ter launching his music player. P18 for example, issued thequery ‘closer Hollywood sign’ while vacationing in Los An-geles immediately after interacting with Google Maps. Thatis she issued a query with local intent (i.e. tied to a physicallocation) after interacting with a location-based maps appli-cation. P8 interacted with Yelp and immediately afterwardssearched for ‘ca dmv wait time’. Again this highlights a con-nection between an app that enables people to search for localbusinesses and services and a query for information tied to alocal business. Finally P4 queried for ‘bcbg generation jellythong sandals’ a type of footwear after using the Macy’s de-partment store app. Again her interaction with a shoppingapp appears to have sparked a search for a specific product.

While these interactions are not generalizable across all ourusers, nor are they indicative of the interactions of all smart-phone users, they do provide anecdotal evidence of real con-nections between apps and mobile search and help shape ourunderstanding of specific app-to-search use cases. Overallthese insights demonstrate that interactions with apps can infact prompt mobile searches.

Apps Before & After SearchNext we dig deeper into mobile search sessions and exploredifferences in app usage before and after mobile search. Wefind 8391 app launches within AppSessionssearch, 2715(32%) happen before the first launch of our app within ses-sions, and 4746 (57%) occur after the first launch of ourapp within sessions. The remaining 11% relate to launches

of the MSearch app. Additionally, we found that out ofthe 913 AppSessionssearch, 338 (36.3%) sessions beganwith a launch of the MSearch app. This indicates that forAppSessionssearch, there is frequently more app activity af-ter mobile search than before.

To explore differences in pre and post search interactions, weextracted the frequency of app launches per app category be-fore the first appearance of the MSearch app and after the lastappearance of the MSearch app, on a per-session basis. Thisallowed us to obtain additional insights about how certain cat-egories of apps might be related to mobile search triggers andactions. We assessed this by a paired t-test comparing thelaunch frequency of every app category before and after thesearch activity. We found significant differences for some cat-egories, in particular after mobile search activity (Table 4).These differences indicate that within search sessions, cer-tain app categories tend to be launched more frequently af-ter mobile search has taken place. Overall we found the useof browsers is more frequent after mobile search. Likewisecommunications apps like Email, SMS and Phone & Audioare used more often after mobile search than before.

Table 4. T-test (paired) results from comparing launch frequency ofapp categories before and after the search activity. These categories dis-played significantly more launches after searching.

Category t∗ Cohen’s dBrowsers 6.63 0.31Email 4.38 0.20Games 2.74 0.13Phone & Audio Communication 2.35 0.11Photography 3.65 0.17SMS/Texting 2.60 0.11Social networking 2.06 0.10System & Settings 2.25 0.11Tools & Utilities 3.57 0.17Unknown 2.49 0.11∗All coefficients are significant at p < .05.

We were interested in exploring if these post-search app in-teractions relate in anyway to mobile search actions. Basedagain on data from interviews and diary responses we iden-tified 9 key mobile search actions among our participants,namely: (1) consuming content, i.e. watching, reading,listening; (2) sharing information; (3) keeping informationwhich involved saving it digitally or taking down notes aswell as making mental notes; (4) buying goods/services;(5) booking something; (6) planning something; (7) visitingsomewhere; (8) contacting a person, business or place; (9)making/doing, e.g. cooking a recipe. Again drawing fromthe qualitative data, we find that some of these actions indeedmap directly to app usage. For example:

• P4 “Downloaded new music to my phone before proceed-ing to the gym”

• P2 “Called the store and confirmed they have what I’mlooking for and will stop by after work Tuesday to buy...”

• P1 “Posted the picture of a Mohawk warrior on Face-book.”

• P6 “Sent information back and forth to sister by Voxer.”

We find that sharing actions in particular were done directlyfrom the mobile device. This involved emailing links to web-sites, sharing details in threads of conversations in messagingapps in particular, as well as taking and sending screenshotsof interesting content. This taking and sending of screen-shots likely relates to the prevalence of Photography apps af-ter search compared to before search. Note that this categoryof app includes the phones in-built photo gallery app which isthe primary source for sharing screenshots on smartphones.

Overall our results highlight differences in app interactionsbefore and after search, and once again anecdotal evidencesuggests that app interactions are used to support certain mo-bile search actions. In the following section we turn our at-tention to switching between search and other apps.

Switching between Apps in SearchWe found that 337 out of the 913 total AppSessionssearchinvolved 2 or more launches of the MSearch app. Thismay be down to participants using mobile search twice withina single session or it may be because participants switchedto/from other mobile apps in between their mobile searches.The later appears more likely. To investigate we extract thetop 10 apps (in terms of app launches), within these 337 ses-sions. Table 5 shows that most of these top apps relate to com-munications, primarily email, texting and social networking.

Table 5. Top 10 apps (in terms of launch frequency) in sessions thatinvolve 2 or more launches of the MSearch app

No App Freq Perc1 Msearch 1001 18.02 Facebook 572 10.33 Internet 306 5.54 Email 266 4.85 Messaging 196 3.56 GO SMS Pro 174 3.17 Chrome 174 3.18 Messages 165 3.09 Gmail 164 3.010 System 136 2.4

In the interviews, participants shed light on their sometimesvery complex switching between apps and services to eitheraddress their information needs or to share their findings. P8for example is an avid coupon user and actively uses appslike Groupon and LivingSocial to find discounts and deals.He explained an interesting example of how he uses both mo-bile search and other apps when looking for concert tickets:“I will actually go to Ticketmaster, see the price and thenopen up another window or another Groupon like that andthen if I get one [groupon code], I’ll copy it. Get out of thereand go back to Ticketmaster to where it usually has a littleicon where it says you can enter a coupon code, so paste it.”.At times when he cannot find tickets, he will also visit siteslike craigslist. This highlights the disjoint and fragmented se-quences of interactions involved. Groupon gives him discountcodes, Ticket Master enables the purchase, while search en-gines allow him to find upcoming concerts in the first place.

P6 described similar sequences of interactions this time usingVoxer with her sister “She had me sign up for Voxer so thatshe can communicate with me, which is actually kind of coolbecause while I’m searching for stuff, we can talk back and

forth or we could text or we can send pictures between us”.In an attempt to find information about a local beach she ex-plained how her searching the Web and communicating withVoxer were interlinked, “I will give her the information andshe’ll look at it, and then she’ll text me something back andthen we’ll both look at it, so we were just going back and forthwith ideas”.

To shed light on reasons for such switching behaviors, in theinterviews we asked participants if and why they use andswitch to other apps like Yelp, Maps and IMDB, etc. forsearching. It appears that search engines are used for broadinformation needs, while apps are used for more specificquestions. One participant likened the interaction to start-ing with search and drilling down with apps. Search engineswere also seen as offering more options, enabling participantsto cast a wide net. This was particularly useful for shoppingrelated needs where options and prices are important, e.g. P4“I would look at the search first because then it gives me anidea because sometimes there are some things that talk aboutprices, like what range and where you would find that pricefrom what app or what location”. Heimonen [12] found asimilar trend comparing search to known websites.

DISCUSSIONIn this paper we have taken a first in-depth look at the com-plex cross app and search interactions surrounding mobile in-formation seeking. Our results indicate that users who en-gage in search activity, i.e. search sessions, interact morewith other mobile apps and for longer durations within thosesessions. We found significant differences in the categoriesof apps used within search sessions compared to non searchsessions. Specifically, browsers, email, SMS, social network-ing, shopping & retail and entertainment related apps are usedmore intensively when people engage with mobile search.

Using both qualitative insights and quantitative analysis wehighlight that some app interactions lead to searches, whileother app interactions are used by participants to take actionafter a search. Anecdotal evidence highlights that categoriesof app usage and search topics are linked, e.g. an interac-tion with Yelp leads to additional local search queries issuedvia a search engine. Our results also show that there is moreapp activity after mobile search than before, suggesting thatparticipants tend to start their app sessions with the intentionof searching. In line with related work [4], this implies thatinteractions with the material world tend to create more infor-mation needs and information seeking behaviors than virtualinteractions.

Our insights into the categories of apps used before and aftermobile search, as well as complex switching between appsand search, point to an overarching theme of task completion.In our earlier examples, we find that for one participant thetask was going to a concert with his girlfriend. For anotherparticipant it was going to a beach with her family. What wascommon across a number of these tasks was that our partici-pants used multiple information sources and often shared theinformation found with others to make joint decisions on dayto day things. This involves using a range of apps both forthe finding and sharing phases of these tasks. This highlights

not only that tasks are more important than individual mobileapp usage, but that tasks are often collaborative or social innature. Future mobile search experiences should take suchsocial interactions into account.

While beyond the scope of this research, it would be interest-ing to gather similar data for a longer time period and withmore users to determine if such patterns could be mined toprovide predictive mobile search capabilities. For example,perhaps the probability of issuing certain queries is higher af-ter using certain types of apps. And likewise the probablyof using certain apps is higher after engaging with differenttypes of mobile search interactions. If such probabilities canbe detected, future mobile search could pre-empt these behav-iors and offer more proactive search experiences. Essentiallysupporting users in task continuation and task completion.

While many of these insights would not have been possiblewithout a detailed analysis of both the quantitative and qual-itative data from our study, there are of course some limi-tations to our approach. We built a custom mobile searchapp for the purpose of the study, thus we were only aware ofsearches conducted within our MSearch app. This meanswe may be missing searches conducted in other search en-gines and other apps. Likewise the fact that participants be-haviors were tracked may have resulted in some deviationfrom the participants’ normal behavior. That said our anal-ysis does shed significant light on the nuances of search andapp interactions. Albeit challenging, we would encourage fu-ture work to extend upon this study and perhaps incorporatesearches across apps.

It’s also important to highlight that we have explored mobilesearch and mobile app interactions within sessions, using a30 second display-off window as a delimiter. Based on inter-views with our participants we found that people also conductsearches that span multiple sessions to address a given need,and these sessions can span differing hours, days, weeks, evenmonths. Particularly when searches were related to biggerthings like buying a car or planning a vacation. We wouldencourage future research to consider search and app interac-tions across multiple sessions and multiple timeframes to seeif other interesting patterns emerge.

Finally we should mention our participants who all live inthe Greater San Francisco Bay area. While we made everyeffort to recruit a diverse sample of participants, we are awarethat mobile search and app usage patterns may differ in otherparts of the world. Thus we would encourage future workin this space in other cities and other countries around theworld. Similarly our sample size was small — 18 Androidusers — thus we cannot claim that our pool of participantsis large enough to be representative of the entire populationof Android users. However we are confident that they werediverse enough to provide us with rich insights and to offerconcrete hypotheses that could be used in future larger-scalestudies.

CONCLUSIONSIn this paper we build upon and extend past research on mo-bile search and mobile app usage, focusing on the complex

interactions and relationships between native apps and mo-bile search engines. By taking a step back from search andexploring the interactions in and around search, we have ex-plored mobile information seeking in a new light. We pro-vide a detailed understanding of cross app and search useand discuss what these findings mean for mobile informationseeking and for improving future search experiences. As thelandscape of smartphones continues to evolve and the linesbetween apps and search engines continue to blur, we wouldargue that more studies of this nature will be needed.

REFERENCES1. Amin, A., Townsend, S., Ossenbruggen, J., and

Hardman, L. Fancy a drink in canary wharf?: A userstudy on location-based mobile search. In Proceedingsof INTERACT ’09:, Springer-Verlag (2009), 736–749.

2. Baeza-Yates, R., and Velasco, J. A study of mobilesearch queries in japan. In Query Log Analysis: Socialand Technological Challenges, WWW Workshop (2007).

3. Bohmer, M., Hecht, B., Schoning, J., Kruger, A., andBauer, G. Falling asleep with angry birds, facebook andkindle: a large scale study on mobile application usage.In Proceedings of Mobile HCI’11, ACM (2011), 47–56.

4. Brown, B., McGregor, M., and Laurier, E. iphone invivo: Video analysis of mobile device use. InProceedings of CHI’13, ACM (2013), 1031–1040.

5. Chua, A. Y. K., Balkunje, R. S., and Goh, D. H.-L.Fulfilling mobile information needs: a study on the useof mobile phones. In Proceedings of ICUIMC ’11, ACM(2011), 92:1–92:7.

6. Church, K., Cherubini, M., and Oliver, N. A large-scalestudy of daily information needs captured in situ. InTOCHI 21, 2 (2014), 10.

7. Church, K., and Oliver, N. Understanding mobile weband mobile search use in today’s dynamic mobilelandscape. In Proceedings of MobileHCI’11, ACM(2011), 67–76.

8. Church, K., Smyth, B., Cotter, P., and Bradley, K.Mobile information access: A study of emerging searchbehavior on the mobile internet. In TWEB 1, 1 (2007), 4.

9. Cui, Y., and Roto, V. How people use the web on mobiledevices. In Proceeding of WWW ’08, ACM (2008),905–914.

10. Dearman, D., Kellar, M., and Truong, K. N. Anexamination of daily information needs and sharingopportunities. In Proceedings of CSCW’08, ACM(2008), 679–688.

11. Ferreira, D., Goncalves, J., Kostakos, V., Barkhuus, L.,and Dey, A. K. Contextual experience sampling ofmobile application micro-usage. In Proceedings ofMobileHCI’14, ACM (2014).

12. Heimonen, T. Information needs and practices of activemobile internet users. In Proceedings of Mobility ’09,ACM (2009), 1–8.

13. Kamvar, M., and Baluja, S. A large scale study ofwireless search behavior: Google mobile search. InProceedings of CHI’06, ACM (2006), 701–709.

14. Kamvar, M., Kellar, M., Patel, R., and Xu, Y. Computersand iphones and mobile phones, oh my!: a logs-basedcomparison of search users on different devices. InProceedings of WWW’09, ACM (2009), 801–810.

15. Komaki, D., Hara, T., and Nishio, S. How does mobilecontext affect people’s web search behavior?: A diarystudy of mobile information needs and search behaviors.In Proceedings of AINA, IEEE (2012), 245–252.

16. Lee, I., Kim, J., and Kim, J. Use contexts for the mobileinternet: A longitudinal study monitoring actual use ofmobile internet services. International Journal ofHuman-Computer Interaction 18, 3 (2005), 269–292.

17. Lee, U., Lee, J., Ko, M., Lee, C., Kim, Y., Yang, S.,Yatani, K., Gweon, G., Chung, K.-M., and Song, J.Hooked on smartphones: an exploratory study onsmartphone overuse among college students. InProceedings of CHI’14, ACM (2014), 2327–2336.

18. McGregor, M., Brown, B., and McMillan, D. 100 daysof iphone use: Mobile recording in the wild. InProceedings of CHI ’14 EA, ACM (2014), 2335–2340.

19. Nielsen. An era of growth: The cross-platform report.http://bit.ly/1pcoQFU, March 2014.

20. Nylander, S., Lundquist, T., and Brannstrom, A. Athome and with computer access: why and where peopleuse cell phones to access the internet. In Proceedings ofCHI’09, ACM (2009), 1639–1642.

21. Sohn, T., Li, K. A., Griswold, W. G., and Hollan, J. D. Adiary study of mobile information needs. In Proceedingsof CHI’08, ACM (2008), 433–442.

22. Song, Y., Ma, H., Wang, H., and Wang, K. Exploringand exploiting user search behavior on mobile and tabletdevices to improve search relevance. In Proceedings ofWWW’13, ACM (2013), 1201–1212.

23. Taylor, C. A., Anicello, O., Somohano, S., Samuels, N.,Whitaker, L., and Ramey, J. A. A framework forunderstanding mobile internet motivations andbehaviors. In Proceedings of CHI’08 EA, ACM (2008),2679–2684.

24. Teevan, J., Karlson, A., Amini, S., Brush, A. B., andKrumm, J. Understanding the importance of location,time, and people in mobile local search behavior. InProceedings of MobileHCI’11 (2011).

25. Tossell, C., Kortum, P., Rahmati, A., Shepard, C., andZhong, L. Characterizing web use on smartphones. InProceedings of CHI ’12, ACM (2012), 2769–2778.

26. Vojnovic, M. On mobile user behaviour patterns. InInternational Zurich Seminar on Communications, IEEECommunications Society (2008).

27. Yi, J., Maghoul, F., and Pedersen, J. Deciphering mobilesearch patterns: a study of yahoo! mobile search queries.In Proceedings of WWW’08, ACM (2008), 257–266.