444 Accidental Influentials JOURNAL OF CON FIGURE 2 of these assumptions is demonstrabl clearly correct either—the empirical ev SCHEMATIC OF NETWORK MODEL OF INFLUENCE inconclusive. Thus we will also presen variations of the basic model that relax and the randomness assumptions. Another advantage of formally deﬁ work, even with such a simple mod deﬁne more precisely what we mea Previous empirical work has address should be considered inﬂuential, b mains elusive (Weimann 1991). Clas of Coleman et al. (1957) and Merton individuals who directly inﬂuence m of their peers should be considered cent market research studies have co ber may be as high as 14 (Burson-M studies, by contrast, deﬁne inﬂuent terms: Keller and Berry (2003), fo ﬂuentials as scoring in the top 10% ership test, while Coulter et al. (2002 treat the top 32% as inﬂuentials. Here we follow the latter approac ential as an individual in the top q% tribution p (n). From a theoretical pers value of q that we specify is necessa we have already argued that dichoto tween opinion leaders and followers a1 inﬂuence can only ﬂow from opinion leaders to fol- derived nor empirically supported. Olowers, in ﬁgure 2, it can ﬂow in either direction. Second, ever, is not to defend any particular din ﬁgure 2 inﬂuence can propagate for many steps, [D. Watts,but to examine the claim that inﬂuen P. Dodds JSTOR 2007]whereas in ﬁgure 1 it can propagate only two. We note, reasonable, self-consistent manner—however, that, in both cases, ﬁgure 2 is consistent with of diffusion processes. From this persavailable empirical evidence—arguably more so than ﬁg- deﬁnition has the advantage (over dure 1. Numerous studies, including that of Katz and La- absolute numbers) that it can be app
Dataset Users technology-savvy, design-conscious Pictures technology, lifestyle, music, design and fashion9,316 users uploaded 6,395 pictures and submitted 13,893 votes.
Identification Trends A simple burst detection method Spotters/Makers Spotter Score: how many, early, popular of the trends Maker Score: how often Typical UsersAll active users (>=2 votes/uploads) who is not spotter or a maker. 140 Makers; 671 Spotters; 1,705 Typical Users
Characterizations Features Activity Content Network Geographical Hypotheses [Kolmogorov-Smirnov tests] Spotters/Makers vs. Typical Users Spotters vs. Makers
ResultsSpotters/Makers vs. Typical users More active, more popular Spotters vs. Makers More votes, less uploads,wider spectrum of interests
Prediction Features ｛ Activity Content Trend SpottersEvery User Social Network Trend Makers User Space Geography User Space
Predictors Follower Geo Span Upload Diversity Daily Uploads Vote Diversity Daily Votes Wandering #Followers #Followees Life Time Age Life Time 0.21 Activity Daily Uploads 0.02 -0.12 Daily Votes 0.05 -0.09 0.47 ⇤ Upload Diversity 0.02 0.09 0.40 ⇤ 0.08 Content Vote Diversity 0.04 0.08 0.22 0.08 0.42 ⇤ Wandering 0.004 0.13 0.16 0.11 0.06 0.05Geographical Follower Geo Span 0.05 0.12 0.16 0.10 0.12 0.11 0.23 #Followers 0.03 0.23 0.37 ⇤ 0.14 0.22 0.16 0.44 0.16 #Followees 0.05 0.17 0.52 ⇤ 0.31 ⇤ 0.29 ⇤ 0.22 0.56 ⇤ 0.21 0.64 ⇤ Network Network Clustering 0.03 0.13 0.22 0.04 0.24 0.23 -0.001 0.27 ⇤ 0.08 0.22 Spotter Score 0.07 0.18 0.03 0.01 0.05 0.10 0.04 0.07 0.13 0.11 0.15 Maker Score 0.07 0.10 0.06 0.01 0.07 0.06 0.02 0.12 0.12 0.09 0.10 Table 5. Pearson Correlation coefﬁcients between each pair of predictors. Coefﬁcients greater than ±0.25 with statistical signiﬁcant level < 0.05 are marked with a ⇤. Practical Implications CONCLUSION The ability of identifying trend spotters and trend makers has A community is an emergent system. It forms from the ac- implications in designing recommender systems, marketing tions of its members who are reacting to each other’s behav- campaigns, new products, privacy tools, and user interfaces. ior. Here we have studied a speciﬁc community of individuals who are passionate about sharing pictures of items (mainly Recommender Systems. Every user has different interests fashion and design items) using a mobile phone application. and tastes and, as such, might well beneﬁt from personalized This community has a speciﬁc culture in which a set of habits, suggestions of content. These suggestions are automatically attitudes and beliefs guide how its members behave. In it, we produced by so-called “recommender systems”. Typically,
y trend spot- Age 2e-04 0.001 preliminary Life Time 0.006 * 0.001 *ers opens up Successful Spotters/Makers Daily Votes (Daily Uploads) 0.007 * 0.16 *ferences be- Vote Diversity (Upload Diversity) 0.38 * 0.14 * Wandering -6e-15 -7e-15 #Followers 2e-05 0.009 * Network Clustering 0.08 0.28 * no previous spotters andl hypotheses (b) Linear Regressioner that trend Features log(Score) tend to vote Spotters Makerscompared to Age 0.36 * 0.01 (H3.1), vote Life Time 0.19 * 0.0001 Daily Votes (Daily Uploads) 0.16 - -1.03 * ote more di- Vote Diversity (Upload Diversity) 7.28 * - -1.09 * ). After run- Wandering -2.1e-13 -1.4e-15nd that trend #Followers -0.06 0.01 * ers who, by Network Clustering 2.75 - -0.64 *oth H3.1 and R2 0.15 0.65 ote, we ﬁnd Adjusted R2 0.14 0.64oad and vote s vote items kers act in a Table 3. Coefﬁcients of the linear regression. A correlation coefﬁcientd in [20, 18] within 2 standard errors is considered statistically signiﬁcant. We high- light and mark them with *.ality content. items in the le they vote of followees, daily uploads, daily votes, and content diver- de spectrum
Summary Successful Spotters Early adopters who vote items from various categories. Successful MakersUsers who upload items belonging to speciﬁc categories, tend to be followed by users from different social clusters.
Conclusions Who Create Trends?Regular individual with speciﬁc interests connected with early adopters with diverse interests.