Social Computing (one)

Social computing and sequencing

0. Concepts, definitions and symbols

The properties of a large-scale network

1. Scale-free distribution (scale-free distribution)

also called Power law distribution (Power laws distribution). In large-scale networks, most nodes have a small degree, while a few nodes have a large degree. Large-scale networks at the Log-log scale exhibit similar patterns: a straight line or an approximate line . This pattern is called the Power law distribution or the scale-free distribution, and the self-similarity is independent of the scale. A network of nodes with a power-law distribution is called a scale-free network .

Friendship Network in YouTube:

Figure 1. Long tail distribution, scale-free distribution Figure 2. Straight line if plot in a log-log scale

2. Small World effect (the Small-world effect)

The longest shortest path in the network is its diameter (diameter), a small path that can be observed from a large, real-world network. such as the famous six-degree separation theory .

3. Strong community structure (strong community structure)

People are more inclined to connect with people in a circle, and people outside the circle tend to have relatively few connections. Friend's friend is easy to be friends, this transitivity can be measured by clustering coefficients (clustering coefficient), there is a relationship between friends (connection) number and the proportion of all contact quantity. Assuming that node vi has dI neighbors, these neighbors have ki edges, then the cluster factor Ci It is:

The clustering factor (clustering coefficient) measures the density of connections between a person's friends. A community-owned network is more likely to have a higher average clustering coefficient than a random network.

Figure 3 asocial network with 9 users and 14 contacts . diameter is 5.

The clustering coefficients of the points are c1= 2/3, c2=1, C3=2/3, C4=1/3, C5=2/3, C6=2/3, C7=1/2, c8=1, c9=0 average clustering coefficients c = (C1 + C2 + ... + C9)/9 = 0.61; and a random network with 9 users and 14 contacts The expected value of the cluster coefficients is 14/(9*8/2) = 0.19.

Ii. new challenges in social media mining

• Scalability (scalability) traditional social network analysis can only handle hundreds of objects or less. The network in social media is huge, and the direct application of traditional network analysis is not feasible.

• hybrid type (heterogeneity) There are many relationships between individuals. Two people may be friends and colleagues at the same time, so there are a variety of interactions among the same group of individuals in a network. Analyzing these promiscuous networks involves mixed entities and mixed interactions, and new theories and tools are needed.

• Evolution (Evolution) social media emphasizes timeliness. For example, in content sharing sites and blog spaces, people quickly lose interest in many shared content and blog posts. This is very different from the traditional web disintegration, where new users are added, new connections are established between existing members, and older users become inactive or simply leave. How should the dynamics of individual networks be acquired? How do we find the hard core members who are the backbone of the network? Can they determine the rise and fall of the community?

• Collective Wisdom (collective intelligence) in social media, people tend to share their connections. In the form of labels, comments, comments, and rankings, the wisdom of the group is usually gained. Meta-information, which is intertwined with users, is useful for many applications. How to use social connection information and collective wisdom effectively to build social computing applications is still a challenge.

• Reviews (evaluation) In traditional data mining, the training-test evaluation model can be used, but it is not the same in social media. Because most social media sites need to protect user privacy, only a very small number of benchmark data (benchmark) can be obtained. In addition, the frequently encountered problems are the lack of real backgrounds in many social computing tasks, which further hinder comparative studies of different jobs. Without a real background, it is difficult to make fair comparisons and evaluations.

Iii. Tasks of Social computing

? Central analysis and impact Modeling social Computing (II.)

? Social Computing for Community Discovery (III.)

? Classification and recommended social computing (establishments)

? privacy, spam and security Social Computing (WU)

Social Computing (one)