Se ha denunciado esta presentación.
Utilizamos tu perfil de LinkedIn y tus datos de actividad para personalizar los anuncios y mostrarte publicidad más relevante. Puedes cambiar tus preferencias de publicidad en cualquier momento.
FINDING BALANCE IN SOCIAL NETWORKS
                                                                                       ...
Ha terminado este documento.
Descárguela y léala sin conexión.
Próximo SlideShare
Storyboard(printscreen)
Siguiente
Próximo SlideShare
Storyboard(printscreen)
Siguiente
Descargar para leer sin conexión y ver en pantalla completa.

0

Compartir

Games on Social Networks: Balancing Friendships

Descargar para leer sin conexión

Summer 2008 REU at Oberlin College.

Audiolibros relacionados

Gratis con una prueba de 30 días de Scribd

Ver todo
  • Sé el primero en recomendar esto

Games on Social Networks: Balancing Friendships

  1. 1. FINDING BALANCE IN SOCIAL NETWORKS Michael Brooks Katie Kuksenok Supervisor: Alexa M. Sharp, Computer Science Department BACKGROUND EXAMPLE Family Tension Balance Theory Social Networks Imagine a family with a mother, father, and son. The Three outcomes would bring balance: (1) the parents and Initially formalized by Fritz Heider in 1958 [2], balance These are collections of nodes and edges representing mother, father, and son are friendly with one another. son become estranged, (2) the son makes his parents theory is an idea from sociology that describes the people and the relationships between them. An change their minds about the bride, or (3) the parents attitudes of groups of people toward each other. ASSIGNMENT is a function mapping each edge to its The son announces that he is getting married. However, makes the son break the engagement. Relationships can either be positive (+) or negative (-), label: (+) or (-). A particular assignment on a social the mother and father do not get along with the bride. This denoting friendship or enmity. network is considered balanced when every triangle of situation is illustrated in Figure 6. Of these possible outcomes, the third only requires one nodes is in a balanced state, as described below. relationship change (Fig. 7), while the others require two. This is the optimal transition to stability. M Figure 6. The triad of a a a a M family members was Figure 7. The son has + + + - + + - - S B balanced, but the bride S B broken his engagement to has caused imbalance. the bride. F F b c b c b c b c + - - - Figure 1. Balanced triads Figure 2. Unbalanced triads These are BALANCED triads. Of the four possible These are UNBALANCED or FRUSTRATED triads. arrangements of positive and negative labels, these They tend to morph into either of the balanced states. are considered less stressful. The first is balanced In the first example, it is stressful for a to stay friends GOAL because of the principle “the friend of my friend is my with two people who are enemies of one another. In friend.” The second is balanced because “the enemy of the second example, it is better to ally with one of your my enemy is my friend.” enemies against the third than to remain enemies with Finding Balance Quickly both. Our objective was to find a way to balance an unbalanced We developed and analyzed several different algorithms network while flipping as few edges as possible. that attempt to solve or approximate this problem. Properties of Complete Networks A COMPLETE NETWORK is one where every person has a relationship with every other person. Figure 3 is an example of a complete network. NP-COMPLETENESS An important property of complete social networks is that, given a balanced assignment, the nodes can be divided into two groups called CLIQUES (Figure 4) with the Correlation Clustering following two properties: Figure 3. A complete network A version of the correlation clustering problem, called Finding the optimal division into k = 2 groups is known to 1. all relationships within each clique are friendly MinDisagree[k], is defined as follows: be NP-Complete. This means that the problem is 2. all relationships between the two cliques are Given a graph where each edge is labeled positive or probably not solvable in a polynomial number of steps. unfriendly negative, divide the nodes into k groups such that as few However, there are fast algorithms to find approximate positive edges are between groups and as few negative solutions. Given an unbalanced complete network, balance can be achieved by forcing the people into two groups and edges are within groups as possible. Since the problem of dividing up nodes into 2 groups flipping some labels to satisfy the two balance properties. This problem has been analyzed by Giotis and while minimizing “bad” edges is NP-complete, and this is Guruswami [1]. exactly our problem, we know that our problem is NP- complete. Figure 4. A balanced network divided into two cliques 2. Algebraic Aggregation 3. Add/Remove Method SOLUTIONS What is the best solution that can be achieved in What is the best solution that can be achieved on a single step of decision-making? specific kinds of social networks? 1. The Physics Simulation This algorithm is intended to translate the principles of the This algorithm is an attempt to improve on the following physical approach described above to a single, non- game-theoretic procedure: What is the best solution that can be generated Method iterative procedure. While the physical algorithm uses Repeatedly, every player chooses to join one clique or the We wrote a simulator that iteratively calculates the effects by allowing people to interact over time? repeated interaction between the 'people' in the network, other to minimize the number of edges he must flip. of the attractive and repulsive forces on the positions of algebraic aggregation is not iterative. Unlike the physical Given that balance theory is supposed to reflect changing This process has the following properties: the n nodes in an n-dimensional space. Points are pulled solution, this algorithm has a definite point of termination. degrees of attraction between people, it seems natural to 1. Guaranteed termination, together and pushed apart until they tend toward some define the input as a physical simulation. configuration. We allow the nodes to move in n Method 2. Quick (polynomial) running time, and 3. At most half of the edges must be changed (m / 2). The positive edges exert an attractive force on their two dimensions so that it was possible to start the nodes at Under this method, we say each person only cares about nodes and negative edges cause a repulsive force. The equidistant locations. The figures show the positions minimizing the number of relationships they have to To improve the quality of the solution, the Add/Remove idea is that over time the nodes diverge into two groups projected into a plane. (Fig. 2) change. There are four steps: algorithm attempts to identify the most troublesome that can be separated easily. Hopefully, making that 1. Every person makes a proposed division of all people nodes. It then uses this procedure on this subset of division stable would require a minimum or at least low Results into two groups, placing itself and its friends into one nodes, reducing the potential for bad outcomes. In many cases, this method results in a good grouping. In number of relationship changes. group and all others into the other group. graphs with little symmetry, the points tend into two 2. Each such proposal is given a weight based on the Results groups such that we are able to find an appropriate This algorithm terminates quickly and produces a solution quality of the proposal—the inverse of the total division between them. However, it can be unclear when that flips at most half of all edges (m / 2). If there are few SYMMETRY number of edges that must be flipped in order for that the simulation should terminate, making the quality of this malicious nodes, we will not change more than half of the partition to be adopted. Problems sometimes arise in networks where all or method difficult to analyze. edges incident to those nodes, so the bound, depending 3. Each edge is assigned a score—the sum of the many of the points are interchangeable, or symmetric. on the actual structure of the network, can be between weights of proposed partitions where it is positive. In these networks, the optimal grouping is often √(m) / 2 and m / 2. 4. Then, given the distribution of scores for each edge, arbitrary. For example, in the network that begins with we find a threshold such that all edges whose score is all negative edges, the optimal solution changes half below it become negative, and all others positive. of them to positive edges. These solutions might have difficulty deciding which nodes should make up each Results Figure 8. A network half. separated into groups This algorithm runs quickly (in polynomial time) and after the Add/Remove Symmetry poses the most problems in the algebraic terminates in a balanced solution. This solution, however, algorithm has balanced it aggregation approach; none at all, however, in the can be as bad as changing all the edges (m). Figure 5. An unbalanced network divided into two cliques. add/remove algorithm described in the next section. This division would require changing 11 edges. Resources Acknowledgements [1] I. Giotis and V. Guruswami. Correlation Clustering with a Fixed Number of Clusters. In Theory of Computing. Volume 2, 2006, pp 249-266. This project was funded by the Booth Ferris Foundation. Facilities provided by Oberlin College. [2] F. Heider. (1958). The psychology of interpersonal relations. New York: John Wiley & Sons.

Summer 2008 REU at Oberlin College.

Vistas

Total de vistas

614

En Slideshare

0

De embebidos

0

Número de embebidos

4

Acciones

Descargas

23

Compartidos

0

Comentarios

0

Me gusta

0

×