A novel mobile device user interface with integrated social networking services

Available online at www.sciencedirect.com

1071-5819/$ - sehttp://dx.doi.org/

nCorrespondinE-mail addre

yanqing.cui@no1Tel.: þ358 5

Please cite thisComputer Stud

Int. J. Human-Computer Studies ] (]]]]) ]]]–]]]www.elsevier.com/locate/ijhcs

A novel mobile device user interface with integrated socialnetworking services

Yanqing Cuin, Mikko Honkala1

Nokia Research Center, Otaniementie 19, Espoo 02150, Finland

Received 21 May 2012; received in revised form 13 October 2012; accepted 1 March 2013

Abstract

Modern mobile devices support accessing Web-based social networking services from the user interface (UI) of Web browsers, applications,and mobile widgets. While effectively accessing these services, people may find it tedious to switch between multiple user interfaces in order tobe aware of the latest content. Aiming for an improved user experience, we experimented with integration of these services into mobile devices'main user interface. The integrated content is presented beyond application silos and automatically filtered to highlight the relevant elements.A mobile system called LinkedUI was developed and deployed in one lab test and one field study. Three findings emerge from these studies.Firstly, it is feasible to construct an alternative device UI that supports integration of Web content across applications and services viahyperlinking. Time, publisher (e.g., contacts), content types, and geographical locations are key dimensions for association of content. Secondly,the alternative device UI enables better usability of accessing social networking services than accessing them from individual Web sites on mobiledevices. It helps people to be aware of the latest content during microbreaks. Thirdly, automatic filtering, on the basis of one user's data, is onepromising approach to identifying relevant content. Given filtered content, most people using the automatic filtering approved the functionalityand experienced a better sense of control that is arguably due to the reduced information volume.& 2013 Elsevier Ltd. All rights reserved.

Keywords: Mobile Web; Social networking services; Hypertext navigation; Automatic filtering; User experience

1. Introduction

People have begun forming habits of regularly checkingpushed live content on their mobile devices. When having aminute or two, they often glance at their devices to maintainawareness of what is going on in their social networks or inthe public world (Church and Oliver, 2011; Oulasvirta et al.,2012; Taylor et al., 2008). This common use case, however, isnot well supported by the current mobile user interfaces (UIs):mobile Web browsers, mobile applications and mobile widgets(Cui et al., 2011; Kaikkonen, 2009). These conventional UIsconfine the content to separate silos. Navigation among thesestructures is time-consuming and error-prone (Cui et al., 2011;Marsden and Jones, 2001; Robbins et al., 2008). The prolif-eration of mobile functions inevitably leads to a broad and

e front matter & 2013 Elsevier Ltd. All rights reserved.10.1016/j.ijhcs.2013.03.004

g author. Tel.: þ358 44 5213648; fax: þ358718036857.sses: [email protected],kia.com (Y. Cui), [email protected] (M. Honkala).0 5704102; fax: þ358 718036857.

article as: Cui, Y., Honkala, M. A novel mobile device user interfacies (2013), http://dx.doi.org/10.1016/j.ijhcs.2013.03.004

deep hierarchy, which makes it hard for people to locate targetfunctions from multiple sources (Marsden and Jones, 2001;Ziefle et al., 2007).With mobile devices, people may also find it difficult to

access personally relevant content amid a multiplicity of feeds.This user problem becomes apparent when a system combinescontent from multiple services – for example, in Unified Inbox(Sohn et al., 2010), Motorola' Motoblur (Bentley et al., 2010),Windows Phone's Hubs, and HTC's Friend Stream. Thesesystems started to emerge recently. Automatic filtering prior-itizes and highlights relevant content without much userinvolvement. It is potentially preferable over customizationwherein people manually hide or highlight content on the basisof their own criteria. It is not easy to create customizationoptions that suit all people; neither is it easy to engage peoplein customizing these tools (Paek et al., 2010).To address the aforementioned user problems, we experi-

mented with an alternative device UI that integrates Web-based social networking services with mobile devices.The research questions in this work are:How can social networking

e with integrated social networking services. International Journal of Human-

www.elsevier.com/locate/ijhcs

dx.doi.org/10.1016/j.ijhcs.2013.03.004

www.elsevier.com/locate/ijhcs

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

mailto:[email protected]



dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

Y. Cui, M. Honkala / Int. J. Human-Computer Studies ] (]]]]) ]]]–]]]2

feeds be integrated and automatically filtered for presentation onmobile devices? How do people perceive and use these mobilesystems with integrated Web services? We answer these questionsby building a novel mobile system: the Linked Internet UI Concept,or LinkedUI, and evaluating the system in user studies.

LinkedUI is an alternative device UI covering all functionsin a mobile device. The device UI incorporates two keyfunctionalities. Firstly, LinkedUI fetches and associates socialnetworking feeds, and it uses hypertext navigation to structurethe entire device UI. Social networking feeds are accessible inmany device views, including home and search. Secondly, toalleviate information overload, LinkedUI supports automaticfiltering. It learns a user's interests, predicts what content theuser might click, and filters content accordingly. From atechnical perspective, it is an on-device filtering solutionwherein only a single user's click data are used for analysis.It predicts user clicks instead of user ratings because userclicks are the most common user actions in real-world systems(Chen et al., 2011; Paek et al., 2010; Wang et al., 2010).

The novelty of this paper is twofold: (1) LinkedUI is notyet another mobile application but an alternative device UI.It effectively incorporates hypertext navigation and automaticfiltering for structuring and presenting all content and functionsin a mobile device. Explorations of alternative UIs areimportant for the evolution of mobile UIs. However, a limitednumber of the previous studies address this topic. The relatedstudies typically envisioned alternative device UIs but seldomfully implemented them due to the development effortsrequired (Björk et al., 2000; Marsden and Jones, 2001; Sohnet al., 2010). (2) We conduct systematic user evaluations withthe new device UI. For example, automatic filtering maycompromise a user's sense of control, comprehensibility andprivacy when applied to personal social networking feeds(Jameson, 2008; Ozenc and Farnham, 2011). These userproblems are important but are often overlooked in earlyevaluation studies (Van velsen et al., 2008).

The bulk of this paper addresses design details of LinkedUIsystems (Section 3), system implementation (Section 4), anduser evaluation results. The system was tested in one lab test(Section 5) and one field study (Section 6). The main purposeof these studies is to evaluate LinkedUI functionalities withreal usage data and user opinions.

2. Related work

LinkedUI experiments with an alternative device UI withintegrated social networking services. Its key elements includestructuring the mobile UI with hypertext navigation, fetchingand aggregating content from multiple services, and automa-tically filtering the integrated content. In this section, wereview the work in these key fields.

2.1. Alternative mobile device UIs based on hypertextnavigation

The conventional device UIs typically organize functionsusing hierarchies such as applications and Web sites. These

Please cite this article as: Cui, Y., Honkala, M. A novel mobile device user interfacComputer Studies (2013), http://dx.doi.org/10.1016/j.ijhcs.2013.03.004

UIs confine the functions into separate information “silos”.To complete a task, people must follow distinctive hierarchystructures of multiple services (Cui et al., 2011; Marsden andJones, 2001; Robbins et al., 2008). This process can be slowand error prone on mobile devices considering the limited input andoutput capabilities and dynamic mobile contexts (Church andOliver, 2011; Cui and Roto; 2008; Taylor et al., 2008).LinkedUI experiments with an alternative UI navigation in

place of hierarchy structures. It organizes Web content togetherwith other native content via hypertext. The relevant literatureis traced back to early hypertext systems. Hypertext is thestructure of using nodes and links as a medium of thinking andcommunication for users (Conklin, 1987). In a meta-analysisof earlier empirical studies, Chen and Rada concluded thathypertext users tend to be more effective than non-hypertextusers, particularly when the users do not aim at specific goals(Chen and Rada, 1996). This conclusion holds true for mobileusers (Ziefle et al., 2007). As one main limitation, hypertextnavigation was used in Web applications (Buchanan et al.,2001; Chan et al., 2002; Kaikkonen and Roto, 2003; Ziefleet al., 2007), but not in native applications on mobile devices.A menu hierarchy is, instead, used in the native UI. Thisdichotomy is confusing for mobile users (Freyne et al., 2010;Kiljander, 2004).In line with LinkedUI design principle, some earlier studies

have explored the feasibility of using hypertext to structure themobile device UI. Marsden and Jones envisioned organizingthe entire mobile UI (both WAP and local applicationfunctions) in hypertext style; but they did not fully implementthe new system (Marsden and Jones, 2001). PowerViewintroduced linkage for navigation of information on mobiledevices. It constructed links between data of different typesand used these to generate the presentation of all informationrelated to the current view. Unlike hyperlinks, the linkage wasnot used to provide navigational shortcuts but insteademployed to offer a context for the current view (Björket al., 2000). Diehl created an associative PDA to storepersonal information through a network of associations onubiquitous devices (Diehl, 2009). Diehl's work focused on theunderlying principles, and it did not address the UI presenta-tion of associated content. Falke continued to sketch howassociations can be created and used in UI presentation, usingnote-taking applications as an example (Falke, 2008). Thisconcept work, however, has not been pursued further in theliterature. Overall, these studies envisioned the seminal mobileUIs but seldom fully implemented these UIs due to thedevelopment efforts required.

2.2. Service aggregation on mobile devices

Previous studies experimented with services aggregation,ranging from early studies about augmented device views tothe recent studies about social networking aggregators. Theseservice aggregation studies typically took place as workaroundin conventional hierarchy based UIs. Although not aiming forgenuine solutions to replace hierarchy-based UIs, these studies


dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

Y. Cui, M. Honkala / Int. J. Human-Computer Studies ] (]]]]) ]]]–]]] 3

explore some UI designs resembling some functionalities ofLinkedUI.

Some early systems experimented with augmenting mobileviews with presence information before the popularity of Webservices. For example, ContextContacts (Oulasvirta et al.,2007), Connecto (Barkhuus et al., 2008), Motion Presence(Bentley and Metcalf, 2007), and Friendlee (Ankolekar et al.,2009) integrate social information into mobile phonebooks.The use case is to help users plan communication, coordinatein-person get-togethers, express themselves, or just keepambient awareness of each other. Similarly, EmotiPix(Cowan et al., 2010) and mGuide (Milic-Frayling et al.,2007) experimented with augmenting home and map viewswith user generated content, mainly images. The related userstudies show that these designs help users handle multiplestreams of content without distracting them from their work-flow. These earlier systems inspire our work, although theydeal with different kind of content.

Recent studies explore mobile versions of social networkaggregators that “combine popular social media feeds inseparate tabs or in one feed and allow posting status updatesto multiple sites” (Jacovi et al., 2011). For example, UniversalInbox combines e-mail messages, text messages, RSS feeds,and Twitter and Facebook updates and allows people tocreate “Lenses” to control the collections of all items (Sohnet al., 2010). Commercial systems, such as Microsoft'sWindows Phone, Motorola's Motoblur (Bentley et al., 2010),Vodafone 360, and HTC Friend Stream, aggregate contentfrom social networking services and use the aggregated feedsto augment phonebook and home views. These systems startedto emerge after we published LinkedUI in 2009. Theyresemble LinkedUI in the logic of integrating Web servicesinto mobile devices. Their appearance indicates the validity ofthis research direction.

2.3. Automatic filtering for social networking feeds

LinkedUI aggregates social networking feeds to mobiledevices. This may accumulate a large amount of content, andmake it necessary to automatically filter the integrated content.

The most common content on mobile devices is socialnetworking feeds, e-mail messages, contacts, SMS content,and phone calls. These kinds of content are not publiclyaccessible, which dictates we should rely on content-basedfiltering rather than collaborative filtering (Adomavicius andTuzhilin, 2005). In content-based filtering, a model is builtthrough examination of the content that interests a user. Then,similar new content is highlighted (Cai et al., 2012).The process does not require access to data from other usersand can be executed on the user's private device. There islimited number of related studies in the literature. Wang et al.(2010) conducted a study in similar setup with computer-users.Their study reveals the importance of “actor”, “actor type,”“activity type,” and “application”. Paek et al. (2010) conducteda study with a hybrid filtering system. Their result supports theimportance of textual features.


Mobile users need to click often to view content due tolimited display size. Such implicit user actions are morecommon on mobile devices than on computers with largedisplays. We use these implicit user actions as prediction targetin this study. This distinguishes our work from earlier studiesthat typically predict explicit user ratings (Paek et al., 2010;Wang et al., 2010). The latter studies have limited applicabilityin that a user rating function is not available in most systems.

2.4. A summary of related work

In summary, we review early studies about alternativemobile device UIs, and two separate research tracks of closerelevance to LinkedUI functionalities: (1) aggregation andhyperlinking of content from other sources with mobiledevices and (2) automatic filtering for social networking feedson PCs. These research tracks were not yet synergized tofacilitate using Web-based social networking services onmobile devices. In this paper, we considered all these potentialfunctions, and redesigned an alternative device UI. We payspecial attention to user experience issues. Automatic integra-tion and filtering of social networking feeds could threatensome key user experience elements, such as sense of control,unobtrusiveness, comprehensibility, privacy, and breadth ofexperience (Jameson, 2008; Ozenc and Farnham, 2011).

3. Concept design

LinkedUI is not standalone mobile application, but a holisticdevice UI governing all functions on a device. Its mainpurpose is to facilitate the use of multiple services on mobiledevices, in particular, social networking services supportingstructured Web content.

3.1. Integration of social networking services with analternative device UI

LinkedUI fetches content from multiple Web services andinterlinks it with native mobile functions. Users can see theintegrated feeds in all logical device views instead of throughmultiple Web sites or applications. Fig. 1 depicts a typicalnavigation sequence and illustrates how a user can navigatebetween services via hyperlinks, back button, and history.LinkedUI combines content from multiple services and pre-sents it in aggregated time (shown in pane b), publisher (pane c),and location views (pane e). From any content item, the user canjump to other content generated by the same publisher (panes b–d)or in the nearby locations (panes d–e). Pane f shows the entirenavigation sequence in the history view.The LinkedUI home view and the LinkedUI activity stream

view emphasize the logic of presenting content in time order,as depicted in Fig. 1's pane a and pane b. LinkedUI constantlychecks for updates of integrated services and notifies users ofthe new content. It allows a user to keep up to date on recentcontent without visiting each service separately. Furtherfiltering of content is made possible for advanced users tofollow via individual services in pane b.


dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

View

This task inthe history

view

Maps: San-F

Image: Union

yanqing cui

All updates

Home

Sequence of viewsClicked touch area

Fig. 1. LinkedUI key views and hypertext navigation among these views. (a) Home view, (b) activity stream view, (c) contact view, (d) content item view, (e) mapview and (f) history view


The LinkedUI contact view is an activity stream published by anindividual contact, as shown in pane c. This view is accessiblefrom any content related to a contact – e.g., a message or a Flickrcomment. To enable this view, LinkedUI support a semi-automaticprocess of associating a given person's various online identities andpieces of contact information: The system matches identitiesautomatically, using unique identifiers such as e-mail address orphone number. It also provides suggestions to assist a user inmanually linking remaining disjoined identities.

The LinkedUI search provides a full-text search function forall cached content of the user and that shared by his or hercontacts. This search can be initiated from any device view.The search is filtered for items related to content types of thecurrent view; however, it is always possible to deselect thefilter and thus search all content instead.

3.2. Automatic filtering for integrated social networking feeds

In its automatic filtering, LinkedUI first predicts the relevance ofnew content items on the basis of earlier user actions and metadataof older items. From the inferred relevance, it highlights somecontent with high inferred relevance as the main user interface andfilters out the rest.


Automatic filtering is a global functionality availablein many LinkedUI views. As shown in Fig. 2(b), when landingin the activity stream view, users by default see the Suggestedtab, which displays content with high derived relevance.The filtered-out content is visible in the All tab only afterfurther user clicks. In a similar fashion, the home view showsthe top three items in the filtered-in content set; the search vieworders the search results according to the predicted relevance.We configured two versions of LinkedUI for the user studies.

One version came with an automatic filtering functionality asillustrated in Fig. 2(b). The other version did not have thisfunctionaltiy; instead, it showed content in its full timeline uponlanding in a view. For example, in the activity stream view, userssaw the All tab by default, as presented in Fig. 2(a). In the homeview, users saw the three latest items from the All tab in thenotification area. These two versions of LinkedUI are identicalapart from the automatic filtering functionality.

4. Implementation

We have developed and deployed functional prototypes forthis research. It supports online and offline operations and isable to synchronize the user's own content and content shared


dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

High-level architecture of the LinkedUI functional prototype

SQL DBSQLite

Linked RecordClasses, Relations, Queries (OR Mapping and

Persistence)

LinkedUI Ontology

Service Adapter

Service-specific Ontology

ViewsObject UI

SystemViews

search, home, contacts, …

Object UI

Supporting Libraries(UI Toolkit, packages, etc.)

Synchronization +Ontology Adaptation

Link

deU

IJa

vaS

crip

tR

untim

eC

/C++ Browser Engine (JS, HTML, CSS)

QTWebKit

Flickr Ovi TwitterInternet Services

RSS, JS APIs, REST, WS, JSON, ...over HTTP

Extends

…

ActionsSystemActions

JS S

QL

API

Fig. 3. High-level architecture of the LinkedUI functional prototype.

Fig. 2. Automatic filtering functionality in the activity stream view: a) with the filtering functionality off and (b) with the filtering functionality on.


by his or her contacts with services. It also synchronizesmetadata describing the content, such as tags, comments, andgeographical data. Our current prototype supports Facebook,Twitter, Flickr, Gmail, and other services, for demonstrationand user evaluation purposes; and it can be extended to covernew views, content types, and services.

4.1. The LinkedUI platform

The LinkedUI platform implemented for this study isdepicted in Fig. 3. The system runs completely on the user'sclient device. The UI and logic layers (middle of Fig. 3) wereimplemented with Web technologies (HTML5, CSS, andJavaScript). The main benefit of using CSS and other Webtechnologies in the UI layer is to increase the abstraction levelin comparison to native programming, thus improving agilityfor UI changes between user tests and aiding in cross-platformdevelopment. The runtime part is written in Cþþ and C (at thebottom in Fig. 3) and is based on Open Source componentsWebKit and SQLite.


The service adapters (at the upper right in Fig. 3) connect tothe Internet services via HTTP and synchronize service data bymeans of public service APIs. The main benefit of thisapproach is the stability of the APIs and resulting adaptersas compared to, e.g., screen scraping. For handling andaggregation of somewhat heterogeneous data from differentservices, a common data repository and a common ontologyhad to be built. In LinkedUI, the data repository, calledLinkedRecord, is a metadata persistence layer. The LinkedUIontology provides the system-wide ontology describing theconcepts that the system supports and understands. As a result,all data in the system are stored in the typed metadata graphprovided by the LinkedRecord system. LinkedRecord wasimplemented as an object-relational mapping and uses a SQLdatabase back end.The object UI allows pluggable UI implementations for

different object classes. System views provide the main views,such as Search and Home. Service adapters are used tointegrate third-party Internet services into the system. Theysynchronize and adapt the data between the LinkedUI ontologyand service-specific data representation. They can also extend


dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004


the system with service-specific ontology extensions, views,object UIs, and actions. This way, the resulting UIs providevarying levels of service-specific content and functions ratherthan only the lowest common denominator of all the services.

The runtime consists of an open-source Web engine,WebKit's Qt port, which was optimized to run well on aNokia N900 smartphone running the Linux-based operatingsystem Maemo 5.

4.2. Automatic filtering

On-device content-based filtering methods were used in thisstudy for the automatic filtering functionality. As depicted inFig. 4, a content-based filtering system takes as input old andnew content from the integrated feeds and the click history ofthe single user. It builds a model of the user's behavior, basedon these datasets. Building the model and using it for filteringis done completely on the user's personal device. As output,the system provides filtered feeds for that user. Inside thefiltering system are the predictor, which predicts user clicks,and the postprocessor, which applies filtering and sorting.

The predictor functions in two, separate stages: it firstlylearns a model from items in the user's history and the user'sclicks on these, and then it predicts click probabilities fornew items.

The postprocessor filters items on the basis of the predictedclick probabilities for the new items. For this study, a staticfiltering threshold ratio F was used for all users. Finally, thepostprocessor orders the items for presentation.

We have explored two alternative types of predictors, apersonalized PageRank predictor and a Bayesian predictor.We also tested a system that combines them into an ensemblepredictor.

Postprocessor

Predict Click Probabilities

Current Items

Old Items + Clicks

Learn Model

Rank

Predictor

Filter

Order byTime

FilteredFeed

Fig. 4. Automatic filtering system.

Table 1The nodes and edges used to create the graph for the PageRank predictor.

Node/edge List of type names in our system

Node Publisher (e.g., User or Contact), Item (e.g., FlickrImage or TwitterUpEdge IdentityUnification (Person-OnlineIdentity), Service (Service-OnlineIde

Commenting (Comment-Item).


In this study, we evaluated the offline accuracy of the threepredictors and further exposed the personalized PageRankpredictor to some users for user acceptance evaluation of thefunctionality.

4.2.1. Personalized PageRank predictorsFor PageRank, the Web content from the integrated feeds

was represented as a graph wherein publisher, services, andcontent items are modeled as nodes and semantic connectionsbetween these are modeled as edges. Examples of connectionsare “the author of a message” and “online identity of a person.”The edges and nodes that were used for creation of the graphare covered in Table 1. The user is considered as an agenttraversing that graph, and the click probability is computed bymeans of the PageRank method.Pure PageRank can be used to compute the “relevance” of

nodes on the basis of their connections within the graph.In addition, we introduce a personalization vector derived fromthe user's click history (Page et al., 1998) that adapts therelevance determination to each user. Since all weights arepropagated in the graph, the weights accumulated for publish-ers are further propagated to new content.

4.2.2. Bayesian predictorsWith a Bayesian predictor, the basic idea is to use a typical

machine-learning setup and transform each content item into afeature vector to be used in training and prediction. The modelis trained via a binary class label clicked (0 or 1) along with thefeature vector. At prediction time, the click probability of anitem is the probability that the item's class label clicked¼1.We implemented a set of features that were intuitively good

for prediction. The set of features used is given in Table 2.Media properties such as the publisher of a post and the type ofthe item were computed, as was a context parameter (time of day).This formulation of the problem allows use of any supervisedclassifier that outputs classification probabilities for the class labelvalues. We concentrate on Naïve Bayes in Section 6.2 aboutprediction accuracy evaluation. Even though it is a simple model, itis optimal under the feature independence assumption and has beenshown to work well even when this assumption does not hold(Zhang, 2004).

4.2.3. Ensemble predictorsWe experimented with a few ways of combining the

PageRank and Bayesian predictors into ensemble predictors.When validating these ensemble predictors, we found that thebest results were obtained with the PageRank value used as anumeric feature for the Bayesian predictor in addition to all the

date), Comment, Service, OnlineIdentity.ntity), Authorship (OnlineIdentity-Item and OnlineIdentity-Comment),


dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

Table 2The features used for the Bayesian predictor.

Feature Explanation of the feature

Click Binary indicator denoting whether the user clicked or did not click the item (prediction class label).Comm Comment history. The numerical rate of the user’s comments that target earlier posts by the author of this item. This history captures also comments

made before joining the study, to improve cold start.Publisher The categorical (non-numeric) identifier of the source who published this post.Type The categorical type of the post (Twitter status update, Facebook comment, Flickr image, etc.).Hour The hour of the day (1, 2, 3, ..., 24) as a categorical feature.

Table 3Usability metric differences between LinkedUI and the benchmark.

Test Tasks Benchmark LinkedUI

T1. No. contacts checked in 3 minutes 4.18 5.27T2. Task completion time (in seconds) 106 37T3. Task completion time (in seconds) 125 67


other features as listed in Table 1. A rank-averaging methodcame quite close in its results. We expect that these predictorscould be successfully combined in a number of ways.

5. A lab test of hypertext navigation

To test the usability of LinkedUI, particularly the hypertextnavigation, we conducted a lab test and compared it with thebenchmark of services' use through a mobile Web browser.Automatic filtering was not implemented in the system tested.The study was conducted in January 2010. It was first reportedin a conference paper (Cui et al., 2010).

Twelve Finnish users (8 male, 4 female) participated in thestudy (names coded as U1–U12). All were 21–34 years old(M¼27.42, SD¼3.78). Each participant subscribed to 42–205contacts in total (M¼98.75, SD¼48.60). Each participant wasrewarded with a 20- or 30-euro gift card.

Three tasks were employed, all using users' own content.The first test task (T1) was to browse freely what was going onin the relevant services within three minutes. The second testtask (T2) was to follow one of their contacts in multipleservices and attempt to make a call. The third test task (T3)was to locate a previously seen Flickr photo and a Twitter post.

Each participant performed all tasks both in LinkedUI and withthe benchmark. Six users started on the benchmark device, theother six with the LinkedUI one. For each device, they completedthe test tasks and rated their subjective impressions on a seven-point scale. In the end, the users stated which device they preferred,and they explained their decisions.

Fig. 5. Average subjective ratings for LinkedUI and the benchmark in the lab test.The result is on a seven-point Likert scale. 1 represents “strongly disagree” and 7stands for “strongly agree.” Note: The values for Q7 are different from an earlypaper reporting this study (Cui et al., 2010). The early publication contains errors inthat wrong divisors were used to divide the grand totals.

5.1. Usability metrics

The key usability metrics are summarized in Table 3. Whenfreely browsing services in T1, the participants were able toaccess significantly more contacts in LinkedUI than with thebenchmark, according to the analysis of vocal protocols(X2(10)¼2.63, p¼0.01). When probed about their opinions,nine participants commented that the activity stream view putthe content from all services in one place, which helped themto scan all updates quickly.

When trying to locate contacts or updates in T2 and T3, allparticipants succeeded in both tasks in LinkedUI. One participantfailed in T2, and one user failed in T3 with the benchmark. It wasfaster to track contacts or content in LinkedUI than with thebenchmark: (T2: t (10)¼4.92, p¼0.00; T3: t (10)¼3.60,p¼0.00). In the user interviews, the performance advantage of

Please cite this article as: Cui, Y., Honkala, M. A novel mobile device user interface with integrated social networking services. International Journal of Human-Computer Studies (2013), http://dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

Fig. 6. Setup for the four-week field study.


LinkedUI was largely attributed to its contact-centric design andsearch functionality.

5.2. Subjective opinions

Describing device preference, eight out of 12 users clearlypreferred LinkedUI and praised it in providing overviews oflive events from multiple services. U6 commented, “I reallyenjoy the fact that everything I can have a look at with, like,one application, instead of having zillions of Web pages.”Three users commented that LinkedUI and the benchmarkcomplement each other. U10 said that LinkedUI “is best whenI am away from my home, where I only want an overview ofmy friend things. But the browser is the device for home whereI want a full experience, since everything is possible there.”One user felt overwhelmed by the “ambitious” changes inLinkedUI and preferred the benchmark. She liked the contentfrom each service to be separate, in order to make it feel“organized.” She had tried a service aggregator on the Webbefore the study but was disappointed by its design.

The users rated a list of statements about their experience ona seven-point scale where 7 represents “strongly agree,” 4means “neither disagree nor agree,” and 1 stands for “stronglydisagree.” Fig. 5 presents their average ratings. On average, theusers preferred LinkedUI over the benchmark “to track anyindividual contact across services,” (5.67 vs. 3.50; t (11)¼3.34, p¼0.00), and “to combine content from multipleservices,” (5.58 vs. 4.00; t (11)¼1.89, p¼0.04). They thoughtthat they would “know their contacts better if keeping usingthe system”, significantly higher than the benchmark figure(5.09 vs. 3.73; t (11)¼3.32, p¼0.00). As U10 commented, “Itis easy for you to know a person. Click the person, and youwill see everything. You have friends in different services, soyou don’t have to log in to all services anymore”.

6. A field study of automatic filtering

To evaluate accuracy and user acceptance of automatic on-device filtering, we ran a field trial involving two groups ofusers. A control group only used LinkedUI without filtering; atest group used LinkedUI with filtering. We used log data fromthe control group to test accuracy of predictors, and usedopinion data from both groups to reveal user acceptance of thefiltering functionality. This field study was conducted betweenAugust 2010 and March 2011. Some of its results was earlierpublished in conference papers (Cui and Honkala, 2011;Honkala and Cui, 2012).

Fourty users (26 male, 14 female) completed the study, 20 usersin the control group (code name as A1–A20), and 20 users in thetest group, (code name as B1–B20). They were 17–53 years old(M¼34.98, SD¼7.79), all living in southern Finland. All theusers were actively using social networking services, in terms ofcontacts per person (control group: median¼285, M¼377,SD¼325; test group: median¼279, M¼298, SD¼92) andupdates per user per day (control group: median¼276,M¼304, SD¼234; test group: median¼194, M¼218,SD¼95). Each user received a 50-euro gift card as reward.


All users were given a Nokia N900 device with LinkedUIpreinstalled for the duration of the four-week study period.Fig. 6 depicts the overall procedure of this study. Both groupsused LinkedUI without filtering in the four-day training period.The filtering system required some training before it func-tioned well. From the fifth day, the control group continuedwith the same unfiltered system while the test group receivedthe filtering functionality. To keep all the users blind to thefiltering functionality, both user groups were primed to expectsome changes and were supplied with a few UI changes (titlechange in the home view) at the “switch” point.In this study, we logged the metadata of the content and the

click history of each user after user consent. The metadata loggedinclude publishing contacts, source services, content types (e.g.,status update, comment, or photo), comments, and time of arrivalin LinkedUI. The usage data include whether the user clicked anitem and which views the users visited. All the log data wereanonymous and abstracted. Twenty-six users were covered in thelog data analysis: 13 from the control group and 13 from the testgroup. These users clicked one or more items each day for at least14 days. The other 14 users were excluded from logging analysisbecause they reported that they only used the tested device assecondary phone due to the slowness of LinkedUI prototype. Weuse 5 min between neighboring accessed views to break usagesessions. In all, there were 5609 sessions.We gathered the user acceptance data from an online ques-

tionnaire and user interviews in the end of the study.All users were covered in the user opinion analysis. In all, 37participants completed the online questionnaire in time, and all 40users took part in the face-to-face interviews. The interview noteswere processed in an affinity diagram (Beyer and Holtzblatt, 1998).In the next few sections, we report main results of this field trial.

We firstly explore how people used and perceived integratedpresentation of social networking feeds. After that, we continue toreport the results about accuracy and user acceptance of automaticfiltering. User quotations are used to support the research results.A1–A20 are code names for the users in the control group.B1–B20 are code names for the users in the test group.

6.1. General results related to LinkedUI systems

An analysis of the filed study results reveals some usagepatterns and user experience characteristics of using social


dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004


networking feeds in LinkedUI. These results aim to substanti-ate the findings about the effects of this device UI as revealedin the lab test.

Fig. 7. Three main predictors and the baseline predictors with individualfeatures. The Bayesian predictor and the PageRank predictors have similarprediction accuracy. Both are better than the baseline predictors that consideronly the person, comm, type, or hour feature.

6.1.1. Frequent check-up sessions and awareness experienceIn the log analysis, we found that the users checked

LinkedUI at short intervals. The median of all breaks was37.63 min. As one example, B14 explained the use case toplace the LinkedUI next to his computer display and glanced atthe screen frequently. In the user interviews, nearly all usersfelt that they used the integrated services more often onLinkedUI than before. As quoted below, a user did not checksome of the integrated services regularly on mobile devicesbefore this study.

A13:“LinkedUI gets things faster, when I unlock the device,they are all there. I end up using these services more often. I donot check these services this often from other devices. Here ithappens all at once, which is nice.”

Awareness was repeatedly highlighted in the user inter-views. The users commented that “it feels like the world isunder my fingertips,” that “it supports ‘always-on’ feeling,”and that “it keeps me up-to-date.” One user commented, “I feelI am better connected to my friends as I am better aware whatthey are doing.” The constant flow of new content did not leadto anxiety. None of the users complained about this potentialproblem. To the contrary, A19 even reported notification ofnew content facilitated his autonomy as quoted below.

B19: “It is kind of “stress relieving.” I do not have to go tosome Web site to pull the data all the time. I can just checkthem out here and then leave it. A good number counter alsohelps the stress relieving. You will see that you have that muchtoday, and you have already read them. So there is no reason toworry.”

Nine users spontaneously reported that they came acrossinteresting content when handling phone calls. LinkedUIshows the latest content published by the caller on callhandling dialog. All these users were positive about thisfunctionality. For example, one user referred to his wife's postto impress her, as quoted below. Another user was happy toread a half-year-old item from her dad. She would not be ableto bump into as old content by other means.

A1: “One thing I really like, I thought that is delightfulfeature. You see the latest updates of a person when he iscalling. I really like it, especially when my wife is calling.She would ask me, did you like my status update onFacebook? I could say, no, but I read it.”

Most users reported that they noticed time-critical contentthey would have not had noticed otherwise. For example, postsabout an incoming vacation, being in a bad mood or under theweather, open invitations and offerings. Four users proceededto offline interaction after reading the content. For example,one user went to drink with a friend after reading the friend'sopen invitation such as “9:00 pm, Bar Amsterdam, Join.”Another user got two concert tickets from one friend who gavethem away on Facebook.


6.1.2. Brief sessions and challenges of locating relevantcontentBrevity was a key characteristic of mobile checkup sessions.

The median of all sessions was 1.08 min in duration. In atypical session, the users only managed to glance at orcarefully read one or two items. Among 5609 sessions, theusers did not click any content in 2970 sessions (53%), clickedone item in 1077 sessions (19%), clicked 2–5 items in 1108sessions (20%), and clicked 5þ items in 454 sessions (8%).In all, the users only clicked 4.9% of the incoming content(8556 out of 176208). In LinkedUI, users need to click an itemto read its full text and its complete conversation history.Before that, they can only see the first words and relatedcontacts.Micro breaks – the brief moments between planned activ-

ities, such as time waiting for a bus or for a friend to arrive –

are one common mobile context (Böhmer et al., 2011; Cui andRoto, 2008). With a short period of time to spare, users neededa quick access to the most relevant content. Three usersborrowed the “newspaper” metaphor to explain their readinghabits. B7 compared the automatic filtering to a newspaper'sfront page to explain why he liked the functionality.B7: “The ‘suggested’ was like the front page of a news-

paper. It contained the stuff that it thinks interests me. Prettyoften, it was right. I went there to check that stuff first. I alsowent to the ‘All’ tab for other information, especially when Iam on the go and do not have other devices.”

6.2. Accuracy of the automatic filtering

In this section, we explore accuracy of the automaticfiltering functionality based on data from the control group.The data was split into two evenly sized partitions (test andtraining sets) for each user. To measure the accuracy, eachpredictor was trained with data from the training set, and testedusing data from the testing set. We used a well-knownbalanced-accuracy measure, the phi correlation coefficient(Yule, 1912) between the observed clicks and the predictions,as the evaluation metric. The phi values were averaged over allstudied users' individual phi values by means of the FisherZ-value back-transform. Fisher's r-to-Z transformation was


dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004


used for statistical tests. For this paper, we used phi instead ofa precision/recall metric because of the low and varying clickrates, which resulted in two unbalanced prediction classes(click, no_click). On average, the users in this field study onlyclicked 4.9% of the content items.

We selected 0.24 as the filtering threshold for this study.This F value yielded the best phi value for almost allindividual users. In another word, it reflects a good tradeoffof precision and recall accuracy. This filtering threshold wasalso in line with an earlier study (Paek et al., 2010).

6.2.1. Predictor accuracyThe offline accuracy of our predictors is shown in Fig. 7,

which presents averaged phi correlations between the predictedand the observed clicks for three main predictors: PageRank,Bayesian, and Ensemble. In this study, the phi value rangedfrom 0 (a random prediction) to 0.490 (a perfect predictor,correctly detecting all clicked items). The maximum phi valueis below 1.0 because our targeted designs show top-24% items,about five times the percentage of user clicks.

As shown in Fig. 7, the Bayesian predictor and thePageRank predictor are similar in prediction accuracy. t(1,12)¼0.46, p¼0.65. The Ensemble of these two predictorsproduces the best result. Ensemble vs. Bayesian: t(1,12)¼0.1.54, p¼0.15; Ensemble vs. PageRank: t(1,12)¼3.21,p¼0.01. This indicates that these two predictors may capturedifferent aspects of user behavior.

6.2.2. Significance of person, content type, hour, and recencyTo analyze the significance of the individual features, we

created a number of baseline Bayesian predictors, each havingjust one feature and the prediction class label. Fig. 7 describesthe performance of each of the three predictors compared,along with these baselines (refer to Table 2. for the definitionof these features). These classifier-based variations commu-nicate the contribution of each classifier, and serve as fourdifferent benchmarks to evaluate the accuracy of our predic-tors. For example, a system that considers only the publisherhighlights the content from contacts clicked earlier and

Fig. 8. Three main predictors with the recency feature and the recency baseline.


de-emphasizes content from others. Overall, the main predic-tors – PageRank, Bayesian, and Ensemble – are better than thebaseline predictors that consider only the person, comm, type,or hour feature. The differences are significant or marginallysignificant except the comparison between the PageRankpredictor and the publisher baseline. publisher vs. PageRank:t(1,12) = 0.77, p = 0.45; publisher vs. Bayesian: t(1,12) =2.07, p = 0.06; publisher vs. Ensemble: t(1,12) = 2.12,p = 0.06.Recency, or how old an item was, has ambiguous meaning in

this study setup. LinkedUI shows the recent content at the top.When a user clicked an item, we cannot determine what led to theclicking: the item was new to the user, or the item was simply atthe top of the list. An earlier study shows that users tend to clickwhatever items on the top (Hjelmeroos et al., 1999). In Fig. 8, wereport the accuracy result of the main predictors with the additionof this ambiguous recency feature and the baseline Bayesianpredictor with just the recency feature. Overall, the recencyfeature had strong prediction power, and that it significantlystrengthens the accuracy of the Bayesian and PageRank pre-dicators. Bayesian vs. bayesianþr: t(1,12)¼4.86, p¼0.00;PageRank vs. pagerankþr: t(1,12)¼3.13, p¼0.01. Futurestudies should continue to disambiguate this recency feature.

Fig. 9. Average subjective ratings for the control and test groups in the fieldstudy. The result is on a seven-point Likert scale. 1 represents “stronglydisagree” and 7 stands for “strongly agree.”


dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004


6.3. User acceptance of automatic filtering

The users in the control and test group filled out aquestionnaire with a seven-point Likert scale. Here, 1 repre-sented “strongly disagree” and 7 stood for “strongly agree.”Fig. 9 presents the means of user ratings of the key statementsin the post-test questionnaire.

Overall, the filtering did not greatly change the overall userexperience of LinkedUI. (S1: 4.94 vs. 4.58), t(1,35)¼0.65,p¼0.526. But it did make users feel less stressed by the largeamount of content in LinkedUI, (S3: 5.67 vs. 4.58), t(1,35)¼1.77, p¼0.09. None of experience dimensions in S4–S7emerged as a user problem (Jameson, 2008). To the contrary,the users mostly reported that filtering improved their experi-ence dimensions, especially in their sense of control (S5: 4.11vs. 2.84); t(1,35)¼2.52, p¼0.02. The user interviews indicatethat this was associated with his or her ability to manage thecontent volume. In the control group, most users complainedabout being overwhelmed by the news feeds. The users in thetest group had fewer similar complaints.

Twenty users in the test group were using LinkedUI with theautomatic filtering. Sixteen users were positive about thefeature: Twelve users agreed with showing the filtered feedsby default; four argued for showing them only when solicited.The other four users were negative about the filtering andsuggested removing it.

Sixteen users were positive about automatic filtering. Theywere not worried about missing content on mobile devices.Most of these users realized that LinkedUI considered theirclicks in filtering content. These users preferred this over thealternative of purely relying on publishing and commentinghistories. They speculated that the latter would reveal popularbut irrelevant content items, and that the relevant content itemsmight not surface because they typically did trigger lengthyconversations.

Four users were negative about filtering feeds. When askedthe reasons, these users referred to a general tendency towards“smart” systems: they simply did not like a system to providehelp proactively, as quoted below. None of these users citedaccuracy issues as the main reason for rejection. They seemedto reject the functionaltiy a priori and did not give it a chance.Their strategy to reduce information volume was to customizethe services' Web sites – for example, hiding or removingcontacts – so that they could handle everything in theseservices.

B11: “I do not use the ‘suggested’ tab since I do not knowwhat is left out. I am the kind of person who wants to seeeverything. I do not throw a newspaper away before glancingat the titles.”

7. Discussion

In this paper, we experimented with a mobile device UI withintegrated Web-based social networking services. A lab testpointed to usability advantages of hypertext navigation overuse of services in a mobile Web browser. A field trial validated


the technical feasibility and the user value of the automaticfiltering functionality to emphasize relevant elements.

7.1. Mobile device UIs with integrated Web services: statusand future

LinkedUI is an alternative device UI solution. It effectivelyincorporated hypertext navigation and automatic filtering tosupport integration of Web-based services with mobiledevices. This contributed to the literature about holistic deviceUI evolutions. A device UI covers all the interactive aspects ofan operating system and determines the broad interaction style.The related studies were often conducted in industry settingsbut seldom documented in scientific literature. Previousscientific studies typically envisioned alternative mobile UIsbut seldom fully implemented them (Björk et al., 2000;Marsden and Jones, 2001; Sohn et al., 2010).After publicizing the LinkedUI concept in 2009, we witnessed

some of its functionalities started to emerge in commercial systems(Bentley et al., 2010; Sohn et al., 2010). For example, Microsoft'sWindows Phone attempts to integrate online content and present itin unified contact, photo, and timeline views. There are, however,notable differences between these commercial systems and Linke-dUI. These systems were typically aimed as a workaround in theconventional hierarchy-based UIs, but did not provide a genuinesolution of in-depth integration of multiple SNSs to devices asLinkedUI did. Neither did these systems support automatic filteringfunctionality that we explored in this study, thus they did not helpusers handle the information overload problem.Given the privacy policies of social networking services, the

automatic filtering in this paper used an on-device content-based approach. To move forward, we recommend usingclassification-based methods such as the Bayesian predictors,or to further improve them by means of an ensemble with thePageRank predictors. Future work should consider includingother implicit behavior metrics, such as display duration of anitem, in the predictors. A previous study showed a positivecorrelation between user attention as measured by eye gazeand user memory/interest and also between user attention andre-tweeting behaviors (Counts and Fisher, 2011). Beyond thescope of social networking services, future studies should alsoconsider the popular collaborative filtering for publicly acces-sible Web and social content.The following issues deserve attention in future design and

research. (1) Construct links for content from differencesources. We envision future systems as, ideally, maximizingautomation and encouraging people in linking of data – forexample, automatically linking a person’s identities wherepossible and aiding users in manual linking of other identities.The linked information in ontology could be kept on onlineservers for a better cross-device experience, which rewardspeople for creating links. (2) Accommodate new services andnew functions. Future systems should support regular manualreconfiguration to keep the system up to date. They maybenefit from third-party development support, which requiressophisticated security mechanisms. (3) Support both universalsearching and “associative” browsing use cases. Universal


dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004


searching is to support goal-oriented tasks. Associative brows-ing is to support the other non-goal-oriented use cases (Lindleyet al., 2012). For example, people explore information thatinterests them, whether on a person, a location, or an event.For these use cases, most of the content should be fetched “onthe fly,” especially if the content comes from a practicallyunlimited source, such as a Web search engine.

This paper addresses user experience issues with mobile useof social networking services. The solutions we proposed mayhave impact on business models of these services.For example, most social networking service providers relyon advertising as their main source of revenue. They may notendorse the integration designs unless aligned with theirbusiness models. An integration system is typically requiredto obtain permission from service providers. In our study, wereceived approval from each social networking service beforewe started the study, and highlighted the brand identity for theservices being considered in LinkedUI. A feasible businessmodel in the future is likely to require some kinds of revenuesharing between mobile manufactures, application developers,and social networking service providers.

7.2. Mobile social networking experience

In the study, we found that people attend to LinkedUI infrequent short bursts of activity. They glance at the deviceviews to be aware of their subscribed content. This usagepattern is in agreement with an early mobile application study(Böhmer et al., 2011) and shows more frequent, briefersessions than in use on stationary computers (Benevenutoet al., 2009). These use sessions contribute to mobile check-uphabits, or “automated behaviors where the device is quicklyopened to check the standby screen or information content in aspecific application” (Oulasvirta et al., 2012). As did an earlystudy (Oulasvirta et al., 2012), we found that people overallwere not annoyed by this frequent checking activity. In thestudy, all users agreed that LinkedUI should constantly fetchthe subscribed content. Only two users asked for the option oftemporarily stopping the constant information flow for the sakeof battery life. When using unfiltered LinkedUI, some usersgave low ratings to their sense of control. These low ratingswere attributed to lack of support in accessing the relevantelements from a large volume of content, not to the fact thatLinkedUI fetched social networking feeds in real time per se.

Integration of social networking services with mobiledevices helped people to gain a strong sense of connectedness(Rettie, 2003), or a feeling of being “in touch with” otherpeople by observing their published content. This result is inline with user studies with earlier systems (Bentley andMetcalf, 2007; Cowan et al., 2010; Oulasvirta et al., 2007).The system occasionally triggers online and offline socialinteractions, but overall the frequency of these interactions islow. This may be due to the established user practices onFacebook, the main tested service in our study. Bumgamerconclude that “the primary way in which Facebook contributesto socializing isn't by offering a medium through which peoplecan meet and communicate with others. Instead, it's by acting


as a virtual watering hole that dispenses information aboutpeers” (Bumgarner, 2007).To reduce information volume, we introduced on-device

automatic filtering of the integrated social networking feeds,which was positively received by most users in the field trial.Given the small slices of time and information needs in mobilecontexts, people expect to see “just enough” from their mobiledevices instead of expecting a system to present all thefunctionality of a full rich client. This highlights the “satisfi-cing” principles in design of the mobile UI. The term, coinedas “satisfice” by Herbert Simon (Simon, 1956), refers to adecision-making strategy of settling with a “good enough”solution rather than searching for the best solution to aproblem. The majority of users approved the fully automaticfiltering system, and experienced an improved sense of control.Future studies need to explore whether automatic filteringperforms better than do other designs, such as user customiza-tion or random filtering, to reduce information load.

7.3. Methodological notes

The studies we performed involved users who actively usesocial networking services. People who are less active in theseservices may not be as keen on constant access. The users inthis study were mainly employed adults. A study with youngeruser groups such as teenagers and students may reveal moreevent coordination cases (Barkhuus and Tashiro, 2010; Cuiand Wang, 2012). The field study lasted for four weeks, whichshould remove novelty or learning effects. However, a longerduration could still be needed to explore issues associated withvery long term usage. People may add new subscription contacts,or start to ignore some existing contacts. They may also changetheir personal interest over time. Even better, the system should bedeployed in the wild for test. People can naturally use it in theirlife.Design details of LinkedUI may bias some findings. For

example, it employed user clicks as the indication of userinterest. This logic is not always valid. Some items were soshort that a user could see the entire post without clicking.It emphasized reverse chronological order in sorting ofcontent. This made it difficult to quantify the recency featurein user click predictions. A new item might be clicked simplybecause it is at the top of the list.Our studies focused mainly on Facebook and Twitter

micro-blogging feeds. These feeds can be naturally pre-sented in time- and contact-related views. Future studiesshould cover other types of services. For example, in sociallocation-based services, map views may be the naturalpresentation. In a news site, audio and/or video services,or e-commerce, other content indices, such as news topics,may emerge as the key structure for organization of content.In addition, LinkedUI was designed for personal devices,and was not for devices shared by multiple anonymoususers, i.e., a group of people using the devices withoutsigning in with their identities.


dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004


8. Conclusion

LinkedUI introduces an alternative device user interface.It aggregates content from several Web services and commu-nication channels into a single consistent UI based onhypertext navigation. This exploration contributed to thelimited literature on alternative device UI solution. The relatedstudies were often done in the industry settings but seldomdocumented in scientific literature. The scientific publications,on the other hand, typically envisioned but seldom implemen-ted alternative mobile UIs (Björk et al., 2000; Marsden andJones, 2001; Sohn et al., 2010).

We focused on user experience characteristics when evalu-ating LinkedUI. The new mobile UI supports better usabilitythan accessing Web sites from a mobile Web browser. It helpspeople be aware of their social networks with minimal attention. Indisagreement with previous publications (Jameson, 2008; Ozencand Farnham, 2011), we learned that automatic filtering enhanceduser's sense of control—probably due to the reduced informationflow—when comparing the people who used the functionality andthe people who did not. This indicates information load influenceuser's sense of control. Given small slices of time and instant needsin mobile contexts, people expect to see “just enough” instead ofall functionalities of a service from their mobile devices.

We analyzed the user activities to evaluate LinkedUI. Ouranalysis supports the feasibility to predict what content a user willclick by observing the user's behavior and the metadata of thecontent. In terms of accuracy metrics, these predictors usingmachine-learning algorithms perform better than do simple pre-dictors considering publishers, comment history, or content types.

Mobile and ubiquitous devices are changing the landscape ofsocial networking experiences. These devices may benefit from analternative device UI supporting in-depth service integration.The designs we explored in this paper may also be valid for othertype of Web services supporting structured data, for example, newsservices, video services, and e-commerce sites. The key design aimis to ensure that people can concentrate on the content interestingfor them, rather than spending time exhausting diverse userinterfaces, being Web sites or mobile applications.

Acknowledgments

The authors wish to thank other team members for theircontribution to LinkedUI: Kari Pihkala and Kimmo Kinnunenimplemented parts of the systems. Guido Grassel supervisedthis project. Mika Rautava, Olli Immonen, Virpi Roto, ElinaOllila, David Arter, Shumeng Ye, Shruti Ramiah, Zac Fitz-Walter, Namita Savla, Melanie Wendland, and Katja Tallbergcontributed to concept work. We express thanks to allparticipants and to John Markow for recruitment aids.

References

Adomavicius, G., Tuzhilin, A., 2005. Toward the next generation ofrecommender systems: a survey of the state-of-the-are and possibleextensions. IEEE Transactions on Knowledge and Data Engineering 17 (6),734–749.


Ankolekar, A., Szabo, G., Luon, Y., Huberman, B. A., Wilkinson, D., Wu, F.,2009. Friendlee: a mobile application for your social life. In: MobileHCI'09,pp. 1–4.

Barkhuus, L., Brown, B., Bell, M., Sherwood, S., Hall, M., Chalmers, M.,2008. Friendship, awareness and repartee: sharing location on the go. In:CHI'08, pp. 497–506.

Barkhuus, L., Tashiro, J., 2010. Student socialization in the age of Facebook.In: CHI'10, pp. 133–142.

Benevenuto, F., Rodrigues, T., Cha, M., Almeida, V., 2009. Characterizinguser behavior in online social networks. In: IMC'09, pp. 49–62.

Bentley, F.R., Kames, J., Ahmed, R., Zivin, R.S., Schwendimann, L., 2010.Contacts 3.0: bring together research and design teams to reinvent thephonebook. In: CHI EA'10, pp. 4677–4690.

Bentley, F. R, Metcalf, C.J., 2007. Sharing motion information with closefamily and friends. In: CHI'07, pp. 1361–1370.

Beyer, H., Holtzblatt, K., 1998. Contextual Design: Defining Customer-Centered Systems. Morgan Kaufmann, San Francisco.

Björk, S., Redström, J., Ljungstrand, P., Holmquist, L. E., 2000. POWER-VIEW: Using information links and information views to navigate andvisualize information on small displays. In: HUC'00, pp. 46–62.

Buchanan, G., Farrant, S., Jones, M., Thimbleby, H., Marsden, G., Pazzani,M., 2001. Improving mobile internet usability. WWW'01, pp. 673–680.

Bumgarner, B., 2007. You have been poked: Exploring the uses andgratifications of Facebook among emerging adults. First Monday 12, 11.

Böhmer, M., Hecht, B., Schöning, J., Krüger, A., Bauer, G., 2011. Fallingasleep with Angry Birds, Facebook and Kindle: a large scale study onmobile application usage. In: MobileHCI'11, pp. 47–56.

Cai, X., Bain, M., Krzywicki, A., Wobcke, W., Kim, Y., Compton, P.,Mahidadia, A., 2012. Collaborative filtering for people to people recom-mendation in social networks. In: AI'10, pp. 476–485.

Chan, S., Fang, X., Brzezinski, J., Zhou, Y., Xu, S., 2002. Usability for mobilecommerce across multiple form factors. Journal of Electronic CommerceResearch 3 (3), 187–199.

Chen, J., Nairn, R., Chi, E., 2011. Speak little and well: recommendingconversations in online social streams. In: CHI'11, pp. 217–226.

Chen, C., Rada, R., 1996. Interacting with hypertext: a meta-analysis ofexperimental studies. Human-Computer Interaction 11 (2), 125–156.

Church, K., Oliver, N., 2011. Understanding mobile Web and mobile searchuse in today's dynamic mobile landscape. In: MobileHCI'11, pp. 67–76.

Conklin, J., 1987. Hypertext: An introduction and survey. Computer 20 (9),17–41.

Counts, S., Fisher, K., 2011. Taking it all in? Visual attention in microblogconsumption. In: ICWSM'11.

Cowan, L., Griswold, W. G., Barkhuus, L., Hollan, J. D., 2010. Engaging theperiphery for visual communication on mobile phones. In: HICSS'10.

Cui, Y., Honkala, M., 2011. The consumption of integrated social networkingservices on mobile devices. In: MUM'11, pp. 53–62.

Cui, Y., Honkala, M., Pihkala, K., Kinnunen, K., Grassel, G., 2010. LinkedInternet UI: A mobile user interface optimized for social networking. In:MobileHCI'10, pp. 45–54.

Cui, Y., Oulasvirta, A., Ma, L., 2011. Event perception in mobile interaction:Toward better navigation history design on mobile devices. InternationalJournal of Human-Computer Interaction 27, 413–435.

Cui, Y., Roto, V., 2008. How people use the Web on mobile devices. In:WWW 2008, pp. 905–914.

Cui, Y., Wang, L., 2012. Motivations for accessing social networking serviceson mobile devices. In: AVI'12, pp. 636–639.

Diehl, J., 2009. Associative personal information management. In: CHI EA'09,pp. 3101–3104.

Falke, E., 2008. The associative PDA 2.0. In: CHI EA'08, pp. 3807–3812.Freyne, J., Berkovsky, S., Daly, E. M., Geyer, W., 2010. Social networking

feeds: recommending items of interest. In: RecSys'10, pp. 277–280.Hjelmeroos, H., Ketola, P., Räihä, K.J., 1999. Coping with consistency under

multiple design constraints: The case of the Nokia 9000 WWW browser.In: MobileHCI'99, pp. 51–56.

Honkala., M., Cui, Y., 2012. Automatic on-device filtering of social network-ing feeds. In: NordiCHI '12, pp. 721–730.


dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004


Jacovi, M., Guy, I., Ronen, I., Perer, A., Uziel, E., Maslenko, M., 2011. Digitaltraces of interest: deriving interest relationships from social mediainteractions. In: ECSCW'11, pp. 21–40.

Jameson, A., 2008. Adaptive interfaces and agents. In: Jacko, JA (Ed.),Human Computer Interaction Handbook: Fundamentals, Evolving Tech-nologies and Emerging Applications 2nd ed. CRC Press, pp. 433–458.

Kaikkonen, A., 2009. Mobile Internet- Past, present, and future. InternationalJournal of Mobile Human Computer Interaction 1 (3), 29–45.

Kaikkonen, A., Roto, V., 2003. Navigating in a mobile XHTML application.In: CHI'03, pp. 329–336.

Kiljander, H., 2004. Evolution and Usability of Mobile Phone InteractionStyles. Dissertation, Helsinki University of Technology.

Lindley, S.E., Meek, S., Sellen, A., Harper, R., 2012. “It's simply integral towhat I do”: enquiries into how the Web is weaved into everyday life. In:WWW'12, pp. 1067–1076.

Marsden, G., Jones, M., 2001. Ubiquitous computing and cellular handsets –are menus the best way forward? In: SAICSIT'01, pp. 111–119.

Milic-Frayling, N., Hicks, M., Jones, R., Costello, J., 2007. On the design andevaluation of Web augmented mobile applications. In: MobileHCI'07,pp. 226–233.

Oulasvirta, A., Petit, R., Raento, M., Tiitta, S., 2007. Interpreting and acting onmobile awareness cues. Human-Computer Interaction 22 (1), 97–135.

Oulasvirta, A., Rattenbury, T., Ma, L., Raita, E., 2012. Habits makesmartphone use more pervasive. Personal and Ubiquitous Computing 16(1), 105–114.

Ozenc, F.K., Farnham, S.D., 2011. Life “modes” in social media. In: CHI'11,pp. 561–570.

Paek, T., Gamon, M., Counts, S., Chickering, D.M., Dhesi, A., 2010.Predicting the importance of newsfeed posts and social network friends.In: AAAI'10.


Page, L., Brin, S., Motwani, R., Winograd, T., 1998. The PageRank citationranking: Bringing order to the Web. Stanford University Database Group.⟨http://ilpubs.stanford.edu:8090/422/⟩ (accessed 26.04.12.).

Rettie, R., 2003. Connectedness, awareness and social presence. In: Proceed-ings of the 6th International Presence Workshop, Aalborg.

Robbins, D.C., Lee, B., Fernandez, R., 2008. TapGlance: Designing a unifiedsmartphone interface. In: DIS'08, pp. 386–394.

Simon, H.A., 1956. Rational choice and the structure of the environment.Psychology Review 63 (2), 129–138.

Sohn, T., Seltur, V., Mori, K., Kaye, J., Horri, H., Battestini, A., Ballagas, R.,Paretti, C., Spasojevic, M., 2010. Addressing mobile information overloadin the universal inbox through lenses. In: MobileHCI'10, pp. 361–364.

Taylor, C.A., Anicello, O., Somohano, S., Samuels, N., Whitaker, L., Ramey,J.A., 2008. A framework for understanding mobile internet motivations andbehaviors. In: CHI EA'08, pp. 2679–2684.

Van velsen, L., Van der geest, T., Klaassen, R., Steehouder, M., 2008. User-centered evaluation of adaptive and adaptable systems: a literature review.Knowledge Engineering Review 23 (3), 261–281.

Wang, Y., Zhang, J., Vassileva, J., 2010. Personalized recommendation ofintegrated social data across social networking sites. In: SAS-WEB'10,pp. 19–30.

Yule, G.U., 1912. On the methods of measuring the association between twoattributes. Journal of the Royal Statistical Society Series A: The Statistician75, 579–652.

Zhang, H., 2004. The optimality of naive Bayes. In: FLAIRS'04, pp. 562–567.Ziefle, M., Schroeder, U., Strenk, J., Michel, T., 2007. How younger and older

adults master the usage of hyperlinks in small screen devices.In: CHI'07, pp. 307–316.


)http://ilpubs.stanford.edu:8090/422/*

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

dx.doi.org/10.1016/j.ijhcs.2013.03.004

Documents

A novel mobile device user interface with integrated social networking services