Source-linked AI summary
Analysis of Social Voting Patterns on Digg
Kristina Lerman, Aram Galstyan
TL;DR
Content producers need ways to promote work, and consumers need ways to find relevant material amid abundant user-generated content. This study examines social voting on Digg and finds that early vote locations indicate whether stories will become popular, with votes beyond the submitter’s neighborhood associated with greater popularity.
Problem
Producers and consumers face challenges promoting content and finding relevant material, while empirical evidence about information propagation on social networks has been sparse and conflicting.
Source: §1 · p. 1Method
The study empirically examines Digg voting and the spread of interest through users’ social networks, using data on stories, voters, and network links.
Source: §1 · p. 2, §3.1 · p. 2, §4 · p. 3Results
Takeaways & Limitations
The authors believe the findings may apply to other social media sites that use social networks to promote content.
Source: §6 · p. 6A more thorough analysis of how network structural properties affect voting dynamics is left for future work.
Source: §6 · p. 6
Abstract
from arXiv · showhide
The social Web is transforming the way information is created and distributed. Blog authoring tools enable users to publish content, while sites such as Digg and Del.icio.us are used to distribute content to a wider audience. With content fast becoming a commodity, interest in using social networks to promote and find content has grown, both on the side of content producers (viral marketing) and consumers (recommendation). Here we study the role of social networks in promoting content on Digg, a social news aggregator that allows users to submit links to and vote on news stories. Digg's goal is to feature the most interesting stories on its front page, and it aggregates opinions of its many users to identify them. Like other social networking sites, Digg allows users to designate other users as ``friends'' and see what stories they found interesting. We studied the spread of interest in news stories submitted to Digg in June 2006. Our results suggest that pattern of the spread of interest in a story on the network is indicative of how popular the story will become. Stories that spread mainly outside of the submitter's neighborhood go on to be very popular, while stories that spread mainly through submitter's social neighborhood prove not to be very popular. This effect is visible already in the early stages of voting, and one can make a prediction about the potential audience of a story simply by analyzing where the initial votes come from.
1. INTRODUCTION
Social networks could help content producers promote work and consumers find relevant material, but empirical evidence about information propagation was sparse and conflicting. This study examines Digg voting and finds that the locations of early votes indicate a story’s likely popularity.
1. INTRODUCTION
- Producers seek ways to distribute content, while consumers need ways to identify relevant material amid a vast stream of user-generated content.Prior empirical studies of information propagation produced conflicting results, including different findings about peer influence and product recommendations.Source: §1 · p. 1
- The study empirically examines how social networks affect voting on Digg, where users submit news links and vote on stories.Digg aggregates user opinions to identify stories for its front page, and its Friends Interface shows stories friends found interesting.Source: §1 · pp. 1–2
- Stories receiving early votes mainly outside the submitter’s social neighborhood are likely to become popular, while neighborhood-heavy voting predicts lower popularity.The pattern is visible early enough to predict a story’s potential audience from where initial votes come from.Source: §1 · p. 2
2. RELATED WORK
Prior Digg and social-media studies examine promotion, voting dynamics, and popularity from different perspectives. This paper focuses on social networks as a mechanism for the spread of interest in stories.
2. RELATED WORK
- Lerman [9] found that Digg users with larger social networks more often got stories promoted, even when those stories were not very interesting.The paper notes a potential “tyranny of the minority” from small groups of active, well-connected users, while arguing that networks can also help assess content quality.Source: §2 · p. 2
- Wu and Huberman [24] studied Digg stories after front-page promotion, finding that interest peaked at promotion and then decayed with an approximately one-day half-life.Their study examined aggregate story dynamics, whereas this paper proposes and studies social networks as a mechanism for spreading interest.Source: §2 · p. 2
- Crane and Sornette used YouTube video vote dynamics to identify high-quality videos, but examined aggregate statistics rather than microscopic spread.Their videos could be editor-selected or spontaneously popular.Source: §2 · p. 2
3. DIGG’S FUNCTIONALITY
Digg promotes stories using user votes and provides a friends interface for tracking activity. The study collects story-voting and network data to describe vote dynamics and the social structure of its sample.
3. DIGG’S FUNCTIONALITY
- Digg lists new submissions in a reverse-chronological upcoming queue and selects some for the front page using voting patterns, including vote counts and rates.The promotion algorithm’s details are secret and change over time; friendship links are asymmetric, and the Friends Interface summarizes friends’ recent activity.Source: §3 · p. 2
3.1 Digg dataset
- User activity was skewed: most users had one promoted story or voted on one story, while some submitted or voted on many.A few users voted on well over a hundred stories.Source: §3 · p. 2, §3.1 · p. 3
- The researchers scraped roughly 200 recently promoted Technology stories in June 2006, recording submitters and chronologically listed voters, then added final vote counts.The collected voting data covered over 16,600 distinct users.Source: §3.1 · pp. 2–3
- After front-page promotion, stories accumulate votes faster than in the upcoming queue, then slow and saturate at totals ranging from fewer than 500 to thousands.Figure 1 plots votes since submission; the rectangle marks the queue period and dashes mark the transition to the front page.Source: Figure 1 · p. 3, §3.1 · p. 3
- In the sample, about 20% of front-page stories received fewer than 500 votes and about 20% received more than 1,500.The distribution resembled a year-long Digg sample reported by Wu and Huberman [24].Source: §3.1 · p. 3
3.2 Social networks
- The researchers reconstructed incoming fan links for users in the voting dataset by combining a top-user network snapshot with later fan data and excluding fans who joined after June 2006.Top users tended to have more friends and fans than other users.Source: §3.2 · p. 3
4. INFORMATION SPREAD IN NETWORKS
Before front-page promotion, Digg stories can reach users through the upcoming queue and the Friends Interface. The study measures that network reach and the share of votes coming through voters’ fans.
4. INFORMATION SPREAD IN NETWORKS
- Before promotion, stories are visible through the upcoming queue and Friends Interface; the study examines how interest spreads through users’ networks.The authors note that the queue received more than 1,500 submissions daily, making it unmanageable for most users to browse.Source: §4 · p. 3
4.1 Information cascades
- After ten votes, almost half the stories were visible to at least 200 users through the Friends Interface; after 30 votes, all reached at least ten users.Figure 3(a) measures influence as the number of users able to see a story through the Friends Interface.Source: §4.1 · p. 3, Figure 3 · p. 4
- For 30% of stories, at least half of the first ten votes came from fans of previous voters; the in-network share grew as more votes arrived.After 20 votes, 28% had at least ten in-network votes; after 30 votes, 36% did.Source: §4.1 · p. 4
5. STORY INTERESTINGNESS
On Digg, early voting patterns distinguish stories that attract interest beyond the submitter’s network from those circulating mainly within it. The study tests whether those patterns can predict story interest, including for stories from top users.
5. STORY INTERESTINGNESS
- Digg generally predicts after about 40 votes, while top-user stories are harder to assess because dense friendship ties can increase their votes and promotion likelihood.Top users are more active and well connected, and are more likely to submit stories that receive votes from their friends.Source: §5 · p. 4
5.1 Social networks and interestingness
- More in-network votes among a story’s first 6–10 votes correspond to lower total-vote interestingness; interesting stories are defined as receiving at least 520 votes.The inverse relationship is visible early; the figure compares in-network votes among the first six, 10, and 20 votes with total votes.Source: §5.1 · p. 5, Figure 4 · p. 5
- The 520-vote threshold was raised from 500 to retain two top-user stories with 505 and 507 votes.Those cases were considered ambiguous, and each had five in-network votes.Source: §5.1 · p. 5
- A poorly connected user’s story with 185 votes still reached the front page after an early vote from kevinrose, Digg’s founder and most-followed user.The passage attributes the promotion to the extra visibility from that vote.Source: §5.1 · p. 5
- The findings suggest two spread mechanisms: broad-interest stories reach independent seed sites, while stories of interest to a narrow community remain within that community.The latter pattern is described as not being picked up by unconnected users.Source: §5.1 · p. 5
5.2 Predicting interestingness
- The C4.5/J48 tree uses v10, fans1, and a 520-vote label; 10-fold validation classified 174 of 207 stories correctly, and top-user testing retained 48 stories.Figure 5 presents the learned tree; the test set was filtered to top users with rank ≤100 and stories receiving at least 10 votes.Source: §5.2 · pp. 5–6, Figure 5 · p. 5
- P=0.57 for the algorithm versus P=0.36 for Digg on the 14 Digg-promoted stories, using the 520-vote interestingness threshold.The algorithm identified seven as interesting, four of which exceeded 520 votes; five of the 14 Digg-promoted stories exceeded that threshold.Source: §5.2 · p. 6
6. CONCLUSION
The study finds that social networks play a significant role in promoting Digg stories and that social voting patterns can predict how interesting a story will become. It identifies further analysis of network structure as future work.
6. CONCLUSION
- The Digg study finds that social networks play a significant role in promoting stories and that social voting patterns can predict story interest.The authors believe the conclusions may apply to other social media sites that use social networks to promote content.Source: §6 · p. 6
- Further work should analyze more thoroughly how network structural properties relate to voting dynamics.The conclusion notes that network structure can affect dynamical processes, including influence propagation.Source: §6 · p. 6