Group formation in large social networks: membership, growth, and evolution

Lars BackstromDan HuttenlocherJon KleinbergXiangyang Lan

article2006KDD2,123 citations

Understanding how social groups form, expand, and evolve over time is critical for managing digital platforms, tracking organizational behavior, and forecasting the adoption of new products or ideas. Historically, research on social dynamics has lacked large-scale, time-resolved empirical data to measure these processes at individual and group levels. The article addresses this challenge by evaluating the specific structural network properties that govern community membership, overall group expansion, and the shifting topical focus between groups.

The investigation analyzes extensive real-world datasets across two domains: LiveJournal, an online social blogging platform encompassing millions of members with explicit user-created groups, and the DBLP computer science publication database, tracking hundreds of thousands of co-authors over decades with academic conferences acting as proxies for communities. Using statistical measurements, decision-tree modeling with ensemble averaging, and temporal burst analysis, the article identifies key predictors of individual adoption and long-term group trajectory.

The findings reveal several fundamental principles of social diffusion. First, an individual's probability of joining a group exhibits a diminishing returns relationship relative to the number of existing friends in that group; having a second connected friend significantly doubles the joining probability compared to having just one, but additional friends yield progressively smaller marginal increases rather than a traditional S-shaped adoption curve. Second, network topology among friends strongly influences joining decisions: an individual is significantly more likely to join a community if their friends within that group are also mutual friends with each other, demonstrating that localized trust and cohesion outweigh simple information diversity. Third, predicting whether a community will grow rapidly depends heavily on internal connectivity; notably, groups with an excessively high ratio of closed triangles to open triads expand significantly less rapidly, indicating that tight internal insularity or post-growth stagnation impedes external growth. Finally, temporal burst analysis shows that topical alignment between communities typically precedes the migration of individuals; shared interest in an emerging topic occurs roughly 50% more frequently than author migration bringing new topics into a community.

These results demonstrate that organizations and platform designers should look beyond raw counts of community members or member friends when forecasting group trajectories or promoting user adoption. Structural configurations—such as the mutual connectivity among an individual's contacts and the overall internal density of groups—provide far more predictive power for engagement and growth. These insights also challenge standard theoretical diffusion models by showing that emerging topics drive user migration rather than migrating users seeding new topics.

For platform operators and community managers, strategies to boost membership should focus on engaging interconnected clusters of friends rather than isolated individuals. Furthermore, organizations seeking to foster growth should monitor high internal triadic closure, which may signal insular communities that struggle to attract new members. Researchers should incorporate friend-to-friend connectedness and asynchronous timing delays into theoretical network diffusion models to better reflect empirical dynamics.

The conclusions carry high statistical confidence across the studied environments due to sample sizes spanning tens of millions of data points. However, limitations exist because the analyzed datasets represent specific environments—an early social blogging site and academic co-authorship networks—which may possess distinct behavioral incentives compared to modern corporate communication tools or alternative consumer platforms. Decision-makers should validate these structural patterns within their specific platform contexts before deploying automated growth or recommendation systems.

blogging coauthorship communication Community Social Communities

Backstrom et al (2006).pdf

Paper: Graphs over time: densification laws, shrinking diameters and possible explanations, J. Leskovec et al. (2005). This foundational paper establishes the empirical densification laws and temporal graph evolution dynamics that frame the source paper's study of group growth.
Paper: The link prediction problem for social networks, David Liben-Nowell et al. (2003). This work introduces topological proximity and link prediction methods that the source adapts to model group membership and triadic closure.
Paper: Maximizing the spread of influence through a social network, David Kempe et al. (2003). This paper establishes the formal algorithmic foundations of social diffusion and influence propagation that the source evaluates empirically at group scale.
Paper: An Information Flow Model for Conflict and Fission in Small Groups, Wayne W. Zachary (1977). This classic study offers the foundational network-flow paradigm for community cohesion and factional split that precedes modern large-scale group formation models.

Paper: Defining and evaluating network communities based on ground-truth, Jaewon Yang et al. (2012). This study builds directly on explicit group affiliations in large social networks to benchmark and define ground-truth community evaluation metrics.
Paper: Learning to Discover Social Circles in Ego Networks, Julian McAuley et al. (2012). This work extends the understanding of localized network topology and ego-network connections to automatically discover fine-grained social circles.
Paper: Community detection in graphs, Santo Fortunato (2009). This comprehensive survey contextualizes empirical group dynamics within the broader algorithmic landscape of graph community detection.
Paper: Temporal Networks, Petter Holme et al. (2011). This review generalizes the source's findings on bursty, time-resolved group evolution into a comprehensive framework for temporal networks.
Paper: Stochastic blockmodels and community structure in networks, Brian Karrer et al. (2010). This paper develops degree-corrected generative models to address real-world community structures that deviate from uniform connectivity assumptions.
Paper: Community detection in networks: A user guide, Santo Fortunato et al. (2016). This guide evaluates modern community detection methods against ground-truth metadata, building on the structural principles observed in evolving groups.

1. INTRODUCTION
2. COMMUNITYMEMBERSHIP
2.1 Dependence on number of friends
2.2 A broader range of features
2.3 Results and Discussion
3. COMMUNITYGROWTH
3.1 Results
3.2 Discussion of Results
4. MOVEMENTBETWEENCOMMUNITIES
4.1 Time Series and Detected Bursts
4.2 Papers Contributing to Movement Bursts
4.3 Alignment between Different Conferences
5. CONCLUSIONSANDFURTHERDIRECTIONS
6. REFERENCES

Knowls

Knowl 1 — Sublinear Joining Probability and Marginal Benefit of Initial Friends
empirical result
In large-scale social networks, the probability $P(k)$ that an individual joins a community given that they already have $k$ friends in that community follows a diminishing-returns curve well-approximated by:

$P(k) = a \log k + b$

for empirical constants $a$ and $b$ .

This relationship was measured empirically across two distinct domains:
1. LiveJournal Social Network: Evaluated on approximately 500 million triples $(u, C, k)$ where user $u$ was not a member of community $C$ and had $k$ friends in $C$ at an initial snapshot. $P(k)$ represents the fraction of such users who joined $C$ within a one-month window.
2. DBLP Co-authorship Network: Evaluated on 7.8 million instances where an author had $k$ co-authors who had previously published in conference community $C$ . $P(k)$ represents the fraction who published in $C$ in the subsequent year.
Both datasets exhibit a distinct deviation from pure sublinear growth at low friend counts:

$P(2) > 2 P(1)$

indicating that the marginal probability boost of acquiring a second friend in the community is disproportionately high. For $k \ge 2$ , the curve increases strictly sublinearly with diminishing returns. This empirical observation contrasts with classical theoretical models of innovation diffusion that posit an S-shaped logistic adoption curve with slow initial adoption followed by a rapid critical-mass inflection point.
Knowl 2 — Effect of Internal Connectedness Among Embedded Friends on Joining Probability
empirical result
For an individual $u$ who is in the fringe of a community $C$ and has a set $S$ of $k$ friends within $C$ , the probability of $u$ joining $C$ depends crucially on the internal connectedness of $S$ .

Let $E_C$ denote the edge set of the subgraph induced by community $C$ . Let $e(S) = |\{(v, w) \in E_C : v, w \in S\}|$ be the number of edges connecting pairs of friends within $S$ , and let the internal link density be defined as:

$\phi(S) = \frac{e(S)}{\binom{|S|}{2}}$

Controlling for the total number of friends $k$ (evaluated for $k=3, 4, 5$ ), the empirical probability that $u$ joins $C$ increases monotonically with $\phi(S)$ . An individual with $k$ friends in a community is substantially more likely to join if those $k$ friends are interconnected mutual friends rather than disconnected from each other.

This finding indicates that within online community membership decisions, the trust, local reinforcement, and coordination benefits associated with closed social capital outweigh the informational diversity benefits predicted by structural hole and weak-tie theories.
Knowl 3 — Decision-Tree Methodology for Individual Community Membership Prediction
model/method
To predict whether an individual $u$ in the fringe of a community $C$ (defined as non-members having at least one friend in $C$ ) will join $C$ within a fixed time window, an ensemble decision-tree learning framework is used based on structural and behavioral features.

Feature Representation: For each pair $(u, C)$ , extracted features include:
1. Community Structural Features: Community size $|C|$ , fringe size, number of boundary edges between $C$ and fringe, internal edge count $|E_C|$ , open triads $|\{(v,w,x) : (v,w),(w,x) \in E_C \wedge (v,x) \notin E_C\}|$ , closed triads $|\{(v,w,x) : (v,w),(w,x),(v,x) \in E_C\}|$ , ratio of closed to open triads, and fringe degree distribution (fraction of fringe nodes with $\ge k$ friends in $C$ for $2 \le k \le 19$ ).
2. Individual-to-Community Structural Features: Number of friends in community $|S|$ , number of adjacent friend pairs $e(S)$ , number of friend pairs connected by paths in $E_C$ , average path distance between friends in $E_C$ , total members reachable from $S$ in $E_C$ , and average distance from $S$ to reachable members.
3. Activity Features (for blogging platforms): Total posts and responses by community members, fraction of members with $\ge 1$ post/response, responses per post, total posts/responses by friends in $S$ , and number of individuals in $S$ with $\ge 1$ post/response.
Ensemble Induction and Model Averaging:
1. 20 binary decision trees are grown.
2. For each tree, communities are randomly assigned to the training set with independent probability $0.5$ , and all fringe members of chosen communities are included.
3. Tree splits are selected greedily to maximize entropy decrease (information gain) across all features and binary thresholds.
4. Tree growth terminates when a node contains fewer than 100 positive instances, producing a leaf that predicts the empirical ratio of positive instances to total instances in that node.
5. Predictions for test instances are obtained by averaging predictions across all trees in which the instance's community was excluded from training.

Knowl 4 — Performance of Individual Community Membership Prediction Models

data/table

The performance of the decision-tree prediction model for single individuals joining communities was evaluated on LiveJournal (17,076,344 fringe pairs $(u, C)$ , 14,488 positive joins, baseline rate $8.48 \times 10^{-4}$ ) over a 1-month window, and on DBLP (7,651,013 author-conference pairs, 71,618 positive joins) over a 1-year window. Performance was evaluated using Area Under the ROC Curve (ROCA), Average Precision (APR), and Cross Entropy (CXE).

LiveJournal Dataset	ROCA	APR	CXE
Number of Friends Only	0.69244	0.00301	0.00934
Post Activity (+ basic features)	0.73421	0.00316	0.00934
All Features (Structural + Activity)	0.75642	0.00380	0.00923
DBLP Dataset	ROCA	APR	CXE
Number of Friends Only	0.64560	0.01236	0.06123
All Structural Features	0.74114	0.02562	0.05808

Incorporating topological features beyond friend count—most notably the internal connectedness and triad structure of friends in the community—substantially improves prediction accuracy and calibration over single-variable baselines.

Knowl 5 — Structural Predictors of Community Growth and the Negative Impact of Triad Closure
empirical result
In predicting whether a community with initial size $|C| \ge 100$ will grow significantly over a 4-month period (formulated as binary classification between high growth $> 18\%$ vs. low growth $< 9\%$ , with mean growth $18.6\%$ and median $12.7\%$ ), decision tree analysis reveals two primary structural determinants:
1. High-Degree Fringe Concentration: The most informative top-level split in the decision tree is the proportion of fringe individuals who have at least 13 friends in the community. Communities surrounded by fringe nodes with high friend counts are significantly more likely to experience rapid growth. In communities with fewer such fringe members, the subsequent split relies on fringe individuals with $\ge 7$ friends.
2. Triad Closure Ratio (Triangle Density): The ratio of closed triads to open triads within the community subgraph $E_C$ :
$\text{Triad Ratio} = \frac{|\{(u, v, w) : (u, v), (v, w), (u, w) \in E_C\}|}{|\{(u, v, w) : (u, v), (v, w) \in E_C \wedge (u, w) \notin E_C \wedge u \neq w\}|}$

is strongly negatively correlated with future community growth. Communities with high closed-to-open triad ratios exhibit significantly lower growth rates. This structural pattern indicates that excessive local cliquishness reduces openness to new members, or alternatively reflects mature communities that have ceased expanding and shifted toward internal densification.

Knowl 6 — Classification Performance for Fast vs. Slow Community Growth

data/table

Community growth prediction was evaluated as a balanced binary classification task on 13,570 LiveJournal communities (initial size $\ge 100$ ), classifying whether a community grew by $>18\%$ (Class 1, 49.4% of instances) or $<9\%$ (Class 0) over a 4-month interval. An ensemble of 100 decision trees (minimum 50 data points per leaf) was trained on structural features and evaluated against baseline feature subsets using Area Under the ROC Curve (ROCA), Average Precision (APR), Cross Entropy (CXE), and Accuracy (ACC).

Features Used	ROCA	APR	CXE	ACC
Fringe Size	0.55874	0.53560	1.01565	0.54451
Community Size	0.52096	0.52009	1.01220	0.51179
Ratio of Fringe to Size	0.56192	0.56619	1.01113	0.54702
Combination of Above 3	0.60133	0.60463	0.98303	0.57178
All Structural Features	0.77070	0.77442	0.82008	0.70035

Simple size and fringe baselines perform only marginally better than random guessing (ROCA $0.52\text{--}0.60$ ). In contrast, incorporating detailed subgraph topology (including triad closure metrics and fringe degree distributions) boosts ROCA to 0.77070 and accuracy to 70.0% (reaching 80% accuracy on the fastest-growing communities).

Knowl 7 — Definitions of Term Bursts and Inter-Community Author Movement Bursts
definition
To analyze the co-evolution of topical content and community membership over time across conference communities $C$ and calendar years $y$ :
1. Conference Membership: An author $a$ is defined as a member of conference $C$ in year $y$ if $a$ published at least one paper in $C$ during the 5-year window $\{y-5, y-4, y-3, y-2, y-1\}$ .
2. Author Movement: Author $a$ moves from conference $B$ into conference $C$ in year $y$ if $a$ publishes a paper in $C$ in year $y$ and was a member of $B$ in year $y-1$ .
3. Movement Time Series and Movement Burst: Let $M_{B,C}(y)$ denote the fraction of all authors at conference $C$ in year $y$ who moved from $B$ into $C$ . A $B \to C$ movement burst is an interval of years during which $M_{B,C}(y)$ exceeds the historical mean of $M_{B,C}$ by an additive difference of at least $0.10$ .
4. Term Time Series and Term Burst: For a title word $w$ and conference $C$ , let $T_{w,C}(y)$ be the fraction of paper titles at $C$ in year $y$ containing $w$ . A word $w$ is defined as hot (in a term burst) at $C$ in year $y$ if $y$ falls within an interval where $T_{w,C}(y)$ can be modeled with a burst rate equal to twice its average rate using a two-state burst-detection automaton.
5. Topical Alignment: Conferences $B$ and $C$ are topically aligned in year $y$ via word $w$ if $w$ is hot at both $B$ and $C$ in year $y$ .

Knowl 8 — Topical Characteristics of Papers Associated with Author Movement Bursts

empirical result

In computer science conferences (analyzed across 87 conferences over $\ge 15$ years in DBLP), papers published by authors who are part of an active $B \to C$ author movement burst exhibit significantly altered usage rates of hot, expired, and future topical terms compared to general papers:

A paper uses a future hot term if it contains a word that experiences a burst at $C$ starting in year $> y$ , and an expired hot term if it contains a word whose burst at $C$ ended in year $< y$ .

Papers contributing to movement bursts show statistically significant elevation in currently hot terms ( $43.91\%$ vs. $38.59\%$ , $p < 10^{-15}$ under a Chernoff-Hoeffding bound) and expired hot terms ( $31.02\%$ vs. $26.37\%$ ), but a marked deficit in future hot terms ( $11.53\%$ vs. $17.40\%$ ). This demonstrates that author migration into a community is drawn by currently or recently active topics rather than serving to introduce or germinate future topical bursts.

Knowl 9 — Temporal Alignment Patterns Between Author Movement and Topical Convergence

empirical result

To determine whether topics follow people or people follow topics between conferences $B$ and $C$ , the temporal onset of three events is evaluated for all instances where $B$ and $C$ are topically aligned via word $w$ during an active $B \to C$ author movement burst:

$t_B$ : Start year of the term burst for $w$ at conference $B$ .
$t_C$ : Start year of the term burst for $w$ at conference $C$ .
$t_{\text{move}}$ : Start year of the $B \to C$ movement burst.

Across 322 non-simultaneous instances in DBLP, four distinct temporal ordering patterns occur:

Pattern Name	Temporal Ordering	Count	Percentage
Shared Interest	$t_B < t_{\text{move}}$ and $t_C < t_{\text{move}}$	194	60.25%
Shared Membership	$t_{\text{move}} < t_B$ and $t_{\text{move}} < t_C$	61	18.94%
Exploration	$t_C < t_{\text{move}} < t_B$	35	10.87%
Colonization	$t_B < t_{\text{move}} < t_C$	32	9.94%

Shared Interest (where the topic bursts in both conferences prior to the onset of the author migration burst) is the dominant pattern, accounting for more cases than the other three patterns combined (60.25% vs 39.75%). The Colonization hypothesis—where authors from $B$ transplant an established topic $w$ from $B$ into $C$ to trigger a burst there—occurs in under 10% of cases. Thus, topical convergence between communities predominantly precedes author migration rather than being caused by it.

Coverage note — Qualitative 2D Latent Semantic Indexing (LSI) visualizations of conference trajectory over time were omitted as exploratory illustration rather than a quantitative empirical result.

References

1.Lada A. Adamic, Orkut Buyukkokten, and Eytan Adar, A Social Network Caught in the Web First Monday, 8(6), 2003.
2.R. Baeza-Yates, B. Ribeiro. Modern Information Retrieval. Addison Wesley, 1998.
3.S.Boorman, P.Levitt. The genetics of altruism. Acad. Pr. 1980.
4.C. Borgs, J. Chayes, M. Mahdian and A. Saberi. Exploring the community structure of newsgroups Proc. 10th ACM SIGKDD Intl. Conf. Knowledge Discovery and Data Mining, 2004.
5.Leo Breiman. Bagging predictors. Machine Learning, 24(2):123–140, 1996.
6.R. Burt. Structural Holes: The Social Structure of Competition. Harvard, 1992.
7.D. Centola, M. Macy, V. Eguiluz. Cascade Dynamics of Multiplex Propagation. Physica A, to appear.
8.J. Coleman Social Capital in the Creation of Human Capital, American Journal of Sociology, 94(Supplement):1988.
9.J. Coleman. Foundations of Social Theory. Harvard, 1990.
10.S. Deerwester, S. Dumais, T. Landauer, G. Furnas, R. Harshman. Indexing by latent semantic analysis. J. Amer. Soc. for Information Science, 41(6), 1990.
11.S. Dill, R. Kumar, K. McCurley, S. Rajagopalan, D. Sivakumar, A. Tomkins. Self-similarity in the Web. 27th International Conference on Very Large Data Bases, 2001.
12.P.S. Dodds, D.J. Watts. Universal behavior in a generalized model of contagion. Phys. Rev. Lett., 92:218701, 2004.
13.P. Domingos, M. Richardson. Mining the Network Value of Customers. Proc. 7th Intl. Conf. Knowledge Discovery and Data Mining,
14.Gary Flake, Steve Lawrence, C. Lee Giles, Frans Coetzee. Self-Organization and Identification of Web Communities. IEEE Computer, 35:3, March 2002.
15.G. W. Flake, R. E. Tarjan, and K. Tsioutsiouliklis. Graph Clustering and Minimum Cut Trees. Internet Math. 1(2004).
16.M. Girvan, M. E. J. Newman. Community structure in social and biological networks. Proc. Natl. Acad. Sci. 99(2002).
17.M. Granovetter. The strength of weak ties. American Journal of Sociology, 78(6):1360-1380, 1973.
18.P. Hoff, A. Raftery, M. Handcock. Latent space approaches to social network analysis. Journal of the American Statistical Association, 97(2002).
19.P. Holme, M. Newman. Nonequilibrium phase transition in the coevolution of networks and opinions. arXiv physics/0603023, March 2006.
20.J. Hopcroft, O. Khan, B. Kulis, B. Selman. Natural communities in large linked networks. Proc. 9th Intl. Conf. on Knowledge Discovery and Data Mining, 2003.
21.D. Kempe, J. Kleinberg, E. Tardos. Maximizing the Spread of Influence through a Social Network. Proc. 9th ACM SIGKDD Intl. Conf. on Knowledge Discovery and Data Mining, 2003.
22.J. Kleinberg. Bursty and Hierarchical Structure in Streams. Proc. 8th ACM SIGKDD Intl. Conf. on Knowledge Discovery and Data Mining, 2002.
23.G. Kossinets, D. Watts. Empirical analysis of an evolving social network. Science, 311:88 – 90, 2006.
24.R. Kumar, J. Novak, P. Raghavan, A. Tomkins. Structure and evolution of blogspace. Comm. ACM 47(2004)
25.J. Leskovec, L. Adamic, B. Huberman. The Dynamics of Viral Marketing. Proc. 7th ACM Conf. on Electronic Commerce, 2006.
26.D. Liben-Nowell, J. Novak, R. Kumar, P. Raghavan, A. Tomkins. Geographic routing in social networks. Proc. Natl. Acad. Sci. USA, 102(Aug 2005).
27.M. Macy, personal communication, June 2006.
28.M. E. J. Newman. Detecting community structure in networks. Eur. Phys. J. B 38, 321-330 (2004).
29.J.R. Quinlan. Induction of decision trees. Machine Learning, 1(1):81–106, 1986.
30.M. Richardson, P. Domingos. Mining Knowledge-Sharing Sites for Viral Marketing. Proc. 8th Intl. Conf. on Knowledge Discovery and Data Mining. 2002.
31.E. Rogers. Diffusion of innovations Free Press, 1995.
32.P. Sarkar, A. Moore. Dynamic Social Network Analysis using Latent Space Models. SIGKDD Explorations: Special Edition on Link Mining, 2005.
33.D. Strang, S.. Soule. Diffusion in Organizations and Social Movements: From Hybrid Corn to Poison Pills. Annual Review of Sociology 24(1998).
34.T. Valente. Network Models of the Diffusion of Innovations. Hampton Press, 1995.
35.F. Viegas and M. Smith. Newsgroup Crowds and AuthorLines. Hawaii Intl. Conf. Sys. Sci. 2004.
36.X. Wang, A, McCallum. Topics over Time: A Non-Markov Continuous-Time Model of Topical Trends, Proc. 12th ACM SIGKDD Intl. Conf. on Knowledge Discovery and Data Mining, 2006.
37.S. Wasserman and K. Faust. Social Network Analysis. Cambridge University Press, 1994.

Citation

MLA

Backstrom, L., et al. “Group Formation in Large Social Networks”. Proceedings of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2006, pp. 44–54, https://doi.org/10.1145/1150402.1150412.

APA

Backstrom, L., Huttenlocher, D., Kleinberg, J., & Lan, X. (2006). Group formation in large social networks. Proceedings of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 44–54. https://doi.org/10.1145/1150402.1150412

Chicago

Backstrom, L., D. Huttenlocher, J. Kleinberg, and X. Lan. 2006. “Group Formation in Large Social Networks”. Proceedings of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 44–54. https://doi.org/10.1145/1150402.1150412.

Harvard

Backstrom, L. et al. (2006) “Group formation in large social networks”, Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, pp. 44–54. Available at: https://doi.org/10.1145/1150402.1150412.

Vancouver

1. Backstrom L, Huttenlocher D, Kleinberg J, Lan X (2006) Group formation in large social networks. In: Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, pp 44–54

BibTeX

@inproceedings{Backstrom_2006, series={KDD06}, title={Group formation in large social networks: membership, growth, and evolution}, url={http://dx.doi.org/10.1145/1150402.1150412}, DOI={10.1145/1150402.1150412}, booktitle={Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining}, publisher={ACM}, author={Backstrom, Lars and Huttenlocher, Dan and Kleinberg, Jon and Lan, Xiangyang}, year={2006}, month=Aug, pages={44–54}, collection={KDD06} }

Metadata:Crossref

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF

Metric	All Papers	Papers Contrib. to Movement
Number of Papers	99,774	10,799
Fraction with Currently Hot Terms	0.3859	0.4391
Fraction with Future Hot Terms	0.1740	0.1153
Fraction with Expired Hot Terms	0.2637	0.3102

Group formation in large social networks: membership, growth, and evolution

Topics

Keywords

Table of Contents

Knowls

Knowl 1 — Sublinear Joining Probability and Marginal Benefit of Initial Friends

Knowl 2 — Effect of Internal Connectedness Among Embedded Friends on Joining Probability

Knowl 3 — Decision-Tree Methodology for Individual Community Membership Prediction