Edvertisements: Adding Microlearning to Social News Feeds and Websites
EEdvertisements: Adding Microlearning toSocial News Feeds and Websites
Geza Kovacs
Stanford UniversityStanford, [email protected]
Figure 1. Our extension can show interactive microlearning tasks (Ed-vertisements) in users’ Facebook news feeds.
ABSTRACT
Many long-term goals, such as learning a language, requirepeople to regularly practice every day to achieve mastery. Atthe same time, people regularly surf the web and read so-cial news feeds in their spare time. We have built a browserextension that teaches vocabulary to users in the context ofFacebook feeds and arbitrary websites, by showing users in-teractive quizzes they can answer without leaving the website.On Facebook, the quizzes show up as part of the news feed,while on other sites, the quizzes appear where advertisementsnormally would. In our user study, we examined the effective-ness of inserting microlearning tasks into social news feeds.We compared vocabulary learning rates when we inserted in-teractive quizzes into feeds, versus inserting links that leadthem to a website where they could do the quizzes. Our re-sults suggest that users engage with and learn from our em-bedded quizzes, and engagement increases when the quizzescan be done directly within their feeds.
Author Keywords microlearning; social feeds; facebook; advertisements;language learning
ACM Classification Keywords
H.5.2. User Interfaces: Graphical user interfaces (GUI)
Figure 2. Our extension can replace advertisements with interactive mi-crolearning tasks (Edvertisements) on arbitrary websites.
INTRODUCTION
Many people have long-term learning goals, such as wantingto learn a new language. However, they often fail to achievethese goals, citing the lack of time to study as a major rea-son [7, 13, 17]. Nevertheless, people do have spare time,as shown by their recreational web browsing and social net-work usage. American adults spend an average of 27 hoursper month surfing the web [16]. 71% of American adults useFacebook, and 63% of these visit Facebook daily [5]. 90% ofAmerican college students use Facebook, spending on aver-age of 30 minutes on Facebook each day [8].In this paper, we present
Edvertisements , which help userslearn during their spare time by showing them interactive mi-crolearning tasks as they surf the web and read their Facebookfeeds. We implemented a Chrome extension which shows Ed-vertisements in two ways: • On Facebook, the extension embeds Edvertisements di-rectly into the feed alongside regular posts (see Figure 1). • On other sites, the extension replaces web advertisementswith Edvertisements (see Figure 2). a r X i v : . [ c s . H C ] F e b ur research questions are: • Do users engage with and learn from Edvertisements thatwe insert into their Facebook feeds? • Do users engage more with Edvertisements if they cancomplete the microlearning tasks without leaving theirFacebook feeds (compared to external links)?In our user study, we examined engagement and vocabularyacquisition rates after embedding Edvertisements into users’Facebook feeds. We found that users interacted frequentlywith Edvertisements, improved their post-test results after aweek, and engaged more readily with Edvertisements whenthey did not have to leave their social news feeds.
RELATED WORKMicrolearning
Microlearning is the strategy of studying frequently in shortintervals throughout the day [11]. Several mobile applica-tions use microlearning to teach foreign language vocabularyand other materials [1, 7]; however, these require the user tointerrupt their routine to open a dedicated app for studying.Some systems try to solve this problem by embedding mi-crolearning into other contexts. There are games in whichusers complete learning tasks while playing [2], video playerswhich teach vocabulary while users watch foreign-languagevideos [15], screensavers that display facts while the screenis idle [12], and chat clients that show vocabulary while theuser is chatting [3].Compared to the learning contexts used by existing systems,we believe that web surfing and Facebook feeds are especiallypotent locations for embedding microlearning, because: • Unlike playing educational games or watching foreign-language videos, visiting Facebook is a daily habit for 45%of American adults with an internet connection. [5] • Web surfing and reading Facebook news feeds are recre-ational activities, so the embedded microlearning tasks willnot interrupt users’ work. • Users are already accustomed to a variety of rich contentappearing in their Facebook feeds, such as videos, games,recommendations, and advertisements, so they should notfind the added Edvertisements too distracting.
Using Social News Feeds to Trigger Desired Habits
Many apps post on users’ Facebook feeds to drive engage-ment. For example, Duolingo can share users’ study progress,and Strava can share users’ exercise history. These posts aimto get users’ friends to send them encouraging feedback, andto try the apps themselves. However, these posts are oftenviewed as bragging about trivial accomplishments, and re-ceive little attention [9].These posts are examples of triggers , which are calls to ac-tion designed to help users form habits [10]. Facebook appposts require the user to go to a different website to study, asFacebook’s API does not allow apps to post interactive con-tent. With Edvertisements, we lower the barrier to action byallowing the user to study without leaving the website.
Web Advertising and Ad-Blocking
Although advertisements are an important revenue source forwebsites, surveys indicate that 77% of users almost neverclick on ads, and 69% express interest in skipping or block-ing ads [19]. 16% of US web users use ad blockers, whichare browser extensions that prevent web ads from being dis-played. Ad blocker usage is growing – global ad blocking hasmore than tripled since 2013, and is posed to further grow asad blockers for mobile devices gain traction [18].In surveys, users of ad-blockers cite “distracting animationsand sounds”, and “offensive/inappropriate ad content” astheir top reasons for blocking ads [19]. Even users who do notinstall ad blockers tend to avoid looking at ads, a phenomenonknown as “banner blindness.” In fact, web surfers click onless than 0.5% of advertisements – a number which has beendeclining ever since banner ads were introduced in 1994 [4].Edvertisements repurpose this space for microlearning.
EDVERTISEMENTS SYSTEM
Our system is a Chrome extension that can show users mi-crolearning tasks – in our case, vocabulary quizzes – inusers’ Facebook feeds, or as they are browsing the web. Al-though we originally implemented the browser extension forChrome, we have also ported it to Firefox, and our techniquecan be implemented on any browser that supports extensions(Chrome, Firefox, Edge, Safari, etc). Our system features avariety of microlearning tasks for learning vocabulary in mul-tiple languages, but in this paper we will focus on learningJapanese vocabulary.
Inserting Edvertisements into Facebook Feeds
Our extension can insert Edvertisements into users’ Facebookfeeds as rectangular interactive quizzes mimicking the lookof a regular feed item, as shown in Figure 1. We chose toinsert 1 microlearning task for every 10 normal feed items,to mimic the approximate frequency we observed sponsoredcontent appearing in the feed.
Replacing Web Advertisements with Edvertisements
People spend considerable time on sites other than Facebook,so we also created a general mechanism for presenting mi-crolearning tasks as users browse the web. Our extension de-tects web advertisements on pages, and replaces them withmicrolearning tasks.We detect the presence of web advertisements using the sameapproach as ad blockers – by checking the URL the elementoriginates from, and comparing it against EasyList, a list ofknown URL patterns for advertisements maintained by Ad-block Plus. When we detect an element that is an advertise-ment, we replace it with an Edvertisement of the same size.Web advertisements follow standardized sizes, called the IAB(Interactive Advertising Bureau) Standard Ad Units. We havecreated microlearning tasks which fit 2 of the common sizes– 300x250 and 200x90 – corresponding to regular-sized andsmall ads. If a microlearning task in the appropriate size isnot available, we pick a smaller one and scale and stack it tofit the available space. For example we can fill a banner ad(728x90) with 3 small Edvertisements, as shown in Figure 2. igure 3. One type of quiz presents a noun in English (umbrella), andasks the user to select the correct translation into Japanese ( kasa ). Theuser has incorrectly selected fukuro , so the user is shown its meaning(bag), and tries again.Figure 4. Another type of quiz presents a noun in Japanese ( jikan ), andasks the user to select its meaning (time).
Quiz Types
One type of quiz presents a noun in English, and asks the userto select the corresponding Japanese word, as shown in Fig-ure 3. To ensure that users learn word associations in bothdirections, we also have a second type of quiz, which showsthe user a Japanese word and asks for the corresponding En-glish translation, as shown in Figure 4.We chose this multiple-choice quiz format because it tests theuser’s knowledge with a minimal amount of interaction – theuser simply clicks on a word to answer. Once the user an-swers a quiz correctly, a new quiz testing a different word isdisplayed. Thus, users can engage with an Edvertisement tocontinue study vocabulary for as long as they wish to.
Quiz Generation
We obtained words and translations from the Nouns sec-tion of Wiktionary’s 1000 Basic Japanese Words list.We excluded loanwords that users would easily recognize( pinku =pink), and words that become homographs when ro-manized ( hana =flower or nose). We focus on nouns, becausethey are the most common type of word [1].
Figure 5. The control condition in our user study inserted a link intousers’ Facebook feeds that led them to a site where they could do quizzes.
Spaced Repetition
Spaced repetition algorithms schedule items for review to en-sure long-term retention [14]. We modified the Memreflexalgorithm [6] to show the overdue word that appeared leastrecently in the feed (or to introduce a new word if there are nooverdue words), instead of always showing the most overdueword. This ensures that users will continue to see differentwords as they are scrolling through their feeds, even if theyare not always answering the in-feed quizzes.
USER STUDY
We conducted a preliminary user study to see how frequentlyusers would engage with Edvertisements, and compare the ef-fectiveness of embedded interactive quizzes that can be com-pleted without leaving the page, versus inserting static linksas is done by today’s advertisements and sponsored Facebookposts.In our study, we only inserted Edvertisements into Facebookfeeds, and did not manipulate advertisements, since manyusers were already using ad-blockers. Furthermore, manip-ulating the Facebook feed enabled us to better control the fre-quency and size of inserted quizzes, compared to repurposingexisting advertisement slots.
Participants
We recruited 14 users (6 female, 8 male; ages 18-29 withmedian age 20) who had not previously studied Japanese butwere interested in learning some basic vocabulary. They werevoluntary participants recruited from online forums and Face-book groups related to Japanese culture. All of our partici-pants self-reported that they were regular users of Facebook.
Materials
We selected 50 basic Japanese words from Wiktionary’s Ba-sic Japanese Words list as the study material. We presentedvocabulary words in romanized form instead of Japanesescript, since our users could not read Japanese script.
Conditions
Users were assigned to one of two conditions: • Users in the in-feed quiz condition had quizzes inserted di-rectly into their feeds, as shown in Figure 1. • Users in the link condition were shown links to a site wherethey could do the quizzes, as shown in Figure 5. igure 6. Vocabulary test scores for the in-feed quiz and link conditions,with standard error bars.Figure 7. Average number of events logged per user for the in-feed quizand link conditions.
Apart from the different items (quizzes/links) inserted intothe feed, the questions and quiz interfaces were identical inboth conditions. In both conditions, the items were insertedat a rate of 1 quiz/link per 10 feed items.
Procedure
The study was conducted entirely online. First, users took apre-test where they tried matching the 50 Japanese words totheir 50 English definitions. Then they installed our Chromeextension and used it to study the 50 words for a week. Aftera week, we asked users to take the post-test, which had thesame format as the pre-test.
RESULTSVocabulary Quiz Results
Figure 6 shows average vocabulary pre-test and post-testscores. On average, users in the in-feed quiz conditionlearned 13.2 new words, compared to 2.4 new words in thelink condition. However, this difference was not statisticallysignificant (t=1.38, p=0.20).
Engagement With Edvertisements
Figure 7 shows the number of times users practiced answer-ing quizzes. We also kept track of “study sessions”, which wedefined as the number of times the user clicked on the link tovisit the website (in the link condition), or first engaged withan Edvertisement (in the in-feed quiz condition).We found that there was high engagement with in-feed Ed-vertisements – on average, users answered 116 quizzes across21 study sessions and answered a question 4.4 days of the 5.7days they visited Facebook. Users in the in-feed quiz condi-tion answered significantly more quizzes than the link con-dition, engaged in more study sessions, and studied on moredays across the week.
Qualitative Feedback
Some users mentioned that they would prefer words to be ex-plicitly introduced first before they start appearing in quizzes.In addition, as we can see from the ratio of study sessions tofeed insertions, even in the in-feed quiz condition, users onlyinteract with 1/5 of quizzes they see. Hence, we need to en-sure that seeing items reinforces memory, even if users do notinteract with them. To address this issue, we later added newtypes of items to introduce new words and review old ones,such as the one shown in Figure 2.
CONCLUSION AND FUTURE WORK
Edvertisements are microlearning tasks that we show to usersas they are surfing the web. We have built a browser extensionthat can insert Edvertisements into Facebook feeds, or replaceweb advertisements with Edvertisements.In our user study, we inserted Edvertisements teaching vo-cabulary into users’ Facebook feeds. We found that usersengaged with and learned from Edvertisements, and that en-gagement rates were significantly higher when the quizzescould be done without leaving their Facebook feeds.Our current implementation can insert Edvertisements intoFacebook feeds, and can replaces web advertisements withthem. There are other online contexts where we might showusers Edvertisements – for example, between Youtube videos,chapters of an e-book, or in their email. Although we have fo-cused on microlearning, Edvertisements could also be used toremind people to do other small, beneficial tasks as they areidly surfing the web – for example, encouraging people to doa short exercise, or complete an item on their to-do list.Other directions for future work include making Edvertise-ments more personalized and contextually relevant, based onthe user’s browsing history and social networks. For example,if today is your Chinese friend’s birthday and your brows-ing habits indicate that you are learning Chinese, we mightshow an Edvertisement in your Facebook feed teaching youhow to wish him a happy birthday in Chinese. Or if you arereading an anti-vaccination webpage, an Edvertisement mightteach you scientific facts about vaccines, showing the namesof your friends who have also completed that Edvertisement.Or if many of your friends run, we might show you an Edver-tisement about the health benefits of running, showing yourfriends’ recent runs as part of the Edvertisement.Production of Edvertisements and monetization is anotherarea of future work. One potential approach is sponsored Ed-vertisements – for example, a local gym might sponsor anEdvertisement which teaches you a workout routine, hopingit will encourage you to visit their gym.People spend hours surfing the web, but many fail to investtime towards learning and other forms of self-improvement.By putting interactive microlearning tasks right in peoples’social feeds and websites, we aim to help users learn in theirspare time, one Edvertisement at a time.
EDVERTISEMENTS DEMO AND SOURCE CODE
Researchers interested in using or building on Edvertisementscan visit https://edvertisements.github.io/
EFERENCES
1. Jennifer S Beaudin, Stephen S Intille,Emmanuel Munguia Tapia, Randy Rockinson, andMargaret E Morris. 2007. Context-sensitivemicrolearning of foreign language vocabulary on amobile device. In
Ambient Intelligence . Springer, 55–72.2. Carrie J Cai. 2013. Adapting arcade games for learning.In
CHI’13 Extended Abstracts on Human Factors inComputing Systems . ACM, 2665–2670.3. Carrie J Cai, Philip J Guo, James R Glass, and Robert CMiller. 2015. Wait-Learning: Leveraging wait time forsecond language education. In
Proceedings of the 33rdAnnual ACM Conference on Human Factors inComputing Systems . ACM, 3701–3710.4. Chang-Hoan Cho and University of Texas at Austin)is an as. 2004. Why do people avoid advertising on theinternet?
Journal of advertising
33, 4 (2004), 89–97.5. Maeve Duggan and Aaron Smith. 2013. Social mediaupdate 2013.
Pew Internet and American Life Project (2013).6. Darren Edge, Stephen Fitchett, Michael Whitney, andJames Landay. 2012. MemReflex: adaptive flashcardsfor mobile microlearning. In
Proceedings of the 14thinternational conference on Human-computerinteraction with mobile devices and services . ACM,431–440.7. Darren Edge, Elly Searle, Kevin Chiu, Jing Zhao, andJames A Landay. 2011. MicroMandarin: mobilelanguage learning in context. In
Proceedings of theSIGCHI Conference on Human Factors in ComputingSystems . ACM, 3169–3178.8. Nicole B Ellison, Charles Steinfield, and Cliff Lampe.2007. The benefits of Facebook “friends:” Social capitaland college students’ use of online social network sites.
Journal of Computer-Mediated Communication
12, 4(2007), 1143–1168.9. Daniel A Epstein, Bradley H Jacobson, Elizabeth Bales,David W McDonald, and Sean A Munson. 2014. From“nobody cares” to “way to go!”: A Design Frameworkfor Social Sharing in Personal Informatics. (2014). 10. Brian J Fogg. 2002. Persuasive technology: usingcomputers to change what we think and do.
Ubiquity
Interactive ComputerAided Learning
Interactive ComputerAided Learning
Proceedings of the SIGCHI conference on Humanfactors in computing systems . ACM, 1097–1106.14. Jeffrey D Karpicke and Althea Bauernschmidt. 2011.Spaced retrieval: absolute spacing enhances learningregardless of relative spacing.
Journal of ExperimentalPsychology: Learning, Memory, and Cognition
37, 5(2011), 1250.15. Geza Kovacs and Robert C Miller. 2014. Smart subtitlesfor vocabulary learning. In
Proceedings of the 32ndannual ACM conference on Human factors in computingsystems . ACM, 853–862.16. Nielsen. 2014. The U.S. Digital Consumer Report.
TheNielsen Company (2014).17. Rebecca Oxford and Jill Shearin. 1994. Languagelearning motivation: Expanding the theoreticalframework.
The modern language journal
78, 1 (1994),12–28.18. Pagefair and Adobe. 2015. The cost of ad blocking.
PageFair and Adobe 2015 report (2015).19. Nevena Vratonjic, Mohammad Hossein Manshaei, JensGrossklags, and Jean-Pierre Hubaux. 2013. Ad-blockinggames: Monetizing online content under the threat of adavoidance. In