Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Look up keyword
Like this
9Activity
0 of .
Results for:
No results containing your search query
P. 1
"To Switch or Not To Switch: Understanding Social Influence in Recommender Systems" - HP Labs research

"To Switch or Not To Switch: Understanding Social Influence in Recommender Systems" - HP Labs research

Ratings: (0)|Views: 6,775 |Likes:
Published by Hewlett-Packard
read more at http://h30507.www3.hp.com/t5/Data-Central/Changing-minds-through-social-media-HP-study-shows-it-happens/ba-p/98721
read more at http://h30507.www3.hp.com/t5/Data-Central/Changing-minds-through-social-media-HP-study-shows-it-happens/ba-p/98721

More info:

Published by: Hewlett-Packard on Sep 15, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

11/06/2012

pdf

text

original

 
 
To Switch or Not To Switch: Understanding SocialInfluence in Recommender Systems
ABSTRACT
We designed and ran an experiment to test how often
 people’s choices are reversed by others’ recommendations
when facing different levels of confirmation and conformitypressures. In our experiment participants were first asked toprovide their preferences between pairs of items. They werethen asked to make second choices about the same pairs
with knowledge of others’ preferences. Our results showthat others people’s opinions significantly sway people’s
own choices. The influence is stronger when people arerequired to make their second decision sometime later(22.4%) than immediately (14.1%). Moreover, people aremost likely to reverse their choices when facing a moderatenumber of opposing opinions. Finally, the time peoplespend making the first decision significantly predictswhether they will reverse their decisions later on, whiledemographics such as age and gender do not. These resultshave implications for consumer behavior research as well asonline marketing strategies.
 
Author Keywords
Social influence, Recommendation systems
ACM Classification Keywords
H.5.3 [Information Interfaces and Presentation]: Group andOrganization Interfaces
 – 
Collaborative computing, Web-based interaction; K.4.4 [Computers and Society]:Electronic Commerce
 – 
Distributed commercialtransactions.
General Terms
Experimentation.
INTRODUCTION
Picture yourself shopping online. You already have an ideaabout what product you are looking for. After navigatingthrough the website you find that particular item, as well asseveral similar items, and other people
s opinions andpreferences about them provided by the recommendationsystem. Will other people
preferences reverse your own?Notice that in this scenario there are two contradictingpsychological processes at play. On one hand, whenlearning of 
other’s
opinions people tend to select thoseaspects that confirm their own existing ones. A largeliterature suggests that once one has a position on an issue,one
s primary purpose becomes defending or justifying thatposition[19]. From this point of view, if others
 recommendations contradict their own opinion, people willnot take this information into account and stick to their ownchoices. On the other hand, social influence and conformitytheory[8]suggest that even when not directly, personally,or publicly chosen as the target of others
disapproval,individuals may still choose to conform to others andreverse their own opinion in order to restore their sense of belonging and self-esteem.To investigate whether online recommendations can swaypeoples
own opinions, we designed an online experimentto test how often people
’s choices
are reversed by others
 preferences when facing different levels of confirmationand conformity pressures. We used Rankr[18]as the studyplatform, which provides a lightweight and efficient way tocrowdsource the relative ranking of ideas, photos, orpriorities through a series of pairwise comparisons. In ourexperiment participants were first asked to provide theirpreferences between pairs of items. Then they were askedto make second choices about the same pairs with theknowledge of others
preferences. To measure the pressureto confirm people
s own opinions, we manipulated the timebefore the participants were asked to make their seconddecisions. And in order to determine the effects of socialpressure, we manipulated the number of opposing opinionsthat the participants saw when making the second decision.Finally, we tested whether other factors (i.e. age, genderand decision time) affect the tendency to revert.Our results show that others
 people’s
opinions significantlysway people
s own choices. The influence is stronger whenpeople are required to make their second decision later(22.4%) rather than immediately (14.1%) after their firstdecision. Furthermore, people are most likely to reversetheir choices when facing a moderate number of opposingopinions. Last but not least, the time people spend makingthe first decision significantly predicts whether they willreverse their decisions later on, while demographics such asage and gender do not.
Haiyi Zhu*, Bernardo A. Huberman, Yarun Luon
Social Computing Group, HP LabsPalo Alto, California, CAhaiyiz@cs.cmu.edu, {bernardo.huberman, yarun.luon}@hp.com
* Haiyi Zhu is also a PhD student in Human Computer InteractionInstitute, Carnegie Mellon University, Pittsburgh, PAThis is unpublished work. It will be submitted to
CHI 2012,
May 5-10,2012, Austin, TX, USA.If you need permission to make digital or hard copies of all or part of thiswork for personal or classroom use, please contact the authors.
 
 
RELATED WORKConfirming Existing Opinions
Confirmation of existing opinions is a long-recognizedphenomenon[19]. As Francis Bacon stated severalcenturies ago[2]:
“The human understanding when it has once
adopted an opinion (either as being received opinion or as being agreeable to itself) draws allthings else to support and agree with it. Althoughthere be a greater number and weight of instancesto be found on the other side, yet these it either neglects and despises, or else by some distinction
 sets aside and rejects”
 
This phenomenon can be explained by Festinger
sdissonance theory: as soon as individuals adopt a position,they favor consistent over inconsistent information in orderto avoid dissonance[11].A great deal of empirical studies supports this idea (see[19] for a review). Many of these studies use a task invented byWason[26], in which people are asked to find the rule thatwas used to generate specified triplets of numbers. Theexperimenter presents a triplet, and the participanthypothesizes the rule that produced it. The participants thentest the hypothesis by suggesting additional triplets andbeing told whether it is consistent with the rule to bediscovered. Results show that people typically testhypothesized rules by producing only triplets that areconsistent with them, indicating hypothesis-determinedinformation seeking and interpretation. Confirmation of existing opinions also contributes to the phenomenon of belief persistence. Ross and his colleagues showed thatonce a belief or opinion has been formed, it can be veryresistant to change, even after learning that the data onwhich the beliefs were originally based were fictitious[21].
Social conformity
In contrast to confirmation theories, social influenceexperiments have shown that often people change their ownopinion to match others
responses. The most famousexperiment is Asch
s[1]line-judgment conformityexperiments. In the series of studies, participants wereasked to choose which of a set of three disparate linesmatched a standard, either alone or after 1 to 16confederates had first given a unanimous incorrect answer.Meta-analysis showed that on average 25% of theparticipants conformed to the incorrect consensus[4].Moreover, the conformity rate increases with the number of unanimous majority. According to Latané, the relationshipbetween conformity and group size follows a negativeaccelerating power function[15]. More recently, Cosleyand his colleagues[10]conducted a field experiment on amovie rating site. They found that by showing manipulatedpredictions, users tended to rate movies toward the shownprediction. Researchers have also found that socialconformity leads to multiple macro-level phenomenons,such as group consensus[1], inequality and unpredictabilityin markets[22], unpredicted diffusion of soft technologies[3]and undermined group wisdom[17]. There are informational and normative motivationsunderlying social conformity, the former based on thedesire to form an accurate interpretation of reality andbehave correctly, and the latter based on the goal of obtaining social approval from others[8]. However, the twoare interrelated and often difficult to disentangletheoretically as well as empirically. Additionally, bothgoals act in service of a third underlying motive to maintainone
s positive self-concept[8].Both self-confirmation and social conformity are extensiveand strong and they appear in many guises. In what followswe consider both processes in order to understand users
 reaction to online recommender systems.
Online recommender systems
Online recommender systems supplement recommendationsprovided by peers such as friends and coworkers, expertssuch as movie critics, and industrial media such as
Consumer Reports
by combining personalized
recommendations sensitive to people’s interests andindependently reporting other peoples’ opinions and
reviews. One popular example of a successful onlinerecommender system is the Amazon product recommendersystem[16].
Users
reaction to recommender system
In computer science and HCI, most research inrecommendation systems has focused on creating good andeffective algorithms (e.g.[5]). There are fewerinvestigations of the basic psychological processesunderlying the interaction of users with recommendations;and none of them addresses both self-confirmation andsocial conformity. As we mentioned, Cosley and hiscolleagues[10]studied conformity in movie rating sites and
showed that people’s ra
ting are significantly influenced by
other users’ ratings.
But they did not consider the effects of self-confirmation or the effects of different levels of socialstrength. Schwind et al studied how to overcome users
 confirmation bias by providing preference-inconsistentrecommendations[24]. However, they representedrecommendations as search results rather thanrecommendations from humans, and thus did notinvestigate the effects of social conformity. Furthermore,their task was more related to logical inference rather thanpurchase decision making.In the area of marketing and customer research, studiesabout the influence of recommendations are typicallysubsumed under personal influence and word-of-mouthresearch[23]. Past research has shown that word-of-mouthplays an important role in consumer buying decisions, andthe use of internet brought new threats and opportunities formarketing[23,13,25]. There were several studiesspecifically investigating the social conformity in product
 
 evaluations[7,9,20]. Although they found substantial
effects of others’ evaluations on people’s own judgment
s,the effects were not always significantly stronger when
others’ opinio
ns were more uniform
1
. In Burnkrant andCousineau
s[7]and Cohen and Golden
s[9]experiments,subjects were exposed to evaluations of coffee with highuniformity or low uniformity. Both results showed thatparticipants did not exhibit significantly increasedadherence to others
evaluation in the high uniformitycondition (although in Burnkrant
s experiments, theparticipants significantly recognized the difference betweenhigh and low uniformity). On the other hand, in Pincus andWaters
s experiments (college students rated the quality of one paper plate while exposed to simulated qualityevaluations of other raters), it was found that conformityeffects are stronger when the evaluations were moreuniform[20].In summary, while previous research showed
that others’
opinions
can influence people’s own decision
s, none of thatresearch addresses both the self-confirmation and socialconformity mechanism that underlie choice among severalrecommendations. Additionally, although previousresearchers concluded that people are more likely to beinfluenced when the social conformity pressures arestronger (i.e. more people uniformly oppose them), theempirical results were mixed. By contrast our experimentsaddress how often people reverse their own opinions whenconfronted with others people preferences, especially whenfacing different levels of confirmation and conformitypressures.
EXPERIMENTAL DESIGN
We conducted a series of online between-subjectsexperiments. All participants were asked to go to thewebsite of Rankr
2
 [18]to make a series of pairwisecomparisons with or without knowledge of others
 people’s
 preferences (Figure 1)
3
. We wanted to determine whether
 people reverse their choices by seeing others’
preferences.
1
Unlike other conformity experiments such as line- judgment where the strength of social conformity ismanipulated by increasing the number of 
unanimous
 majority, experiments about social influence in productevaluation[7, 9, 20]usually manipulate the strength of  social influence by changing the degree of uniformity of opinions. As discussed in[9], since it is seldom that novariation exists in the advice or opinions in reality, the latter
method is more likely to stimulate participants’ real
reactions. We also use the latter method in our experimentby manipulating the ratio of opposing opinions versussupporting opinions.
2
3
The pictures were collected from Google Images.
Conditions
The experimental design was 2x3x4 measuring (babypictures and loveseat pictures) versus (strong confirmation,weak confirmation and control) versus (ratio of opposingopinions versus supporting opinions: 2:1, 5:1, 10:1 and20:1).
Participants were recruited from Amazon’s
Mechanical Turk and were randomly assigned into one of six conditions (baby
 – 
strong, baby-weak, baby
 – 
control,loveseat-strong, loveseat-weak and loveseat-control) andmade four choices with different levels of conformitypressure.In the baby condition, people were asked to comparetwenty-three or twenty-four pairs of baby pictures byanswering the question
which baby looks cuter on a babyproduct label
. In the loveseat condition, the question was
your close friend wants your opinion on a loveseat fortheir living room, which one do you suggest
; and peoplealso needed to make twenty-three or twenty-four choices.In the strong confirmation condition, people first comparedtwo pictures on their own and they were then immediatelyasked to make another choice with available informationabout others
preferences. When the memories were fresh,
 
a. Comparing
baby pictures, not showing others’ preferences
 
b. Comparing loveseats, showing
others’ preferences
 Figure 1. Example pairwise comparisons in Rankr

Activity (9)

You've already reviewed this. Edit your review.
1 hundred reads
1 thousand reads
Lukas Fischer liked this
Arun Rangarajulu liked this

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->