Social Network Analysis and Mining for Business Applications

Bonchi, F., Castillo, C., Gionis, A., & Jaimes, A. (2011). Social Network Analysis and Mining for Business Applications. ACM Transactions on Intelligent Systems and Technology, 2(3), 22:1—22:37. doi:10.1145/1961189.1961194

This article classifies business processes based on the APQC Process Classification Framework that serves as a high-level, industry-neutral enterprise process model, and discusses social network analysis and mining techniques relevant to these processes.

Business Processes and Applications

The following list outlines business processes and the corresponding applications. The processes 1-5 are considered operating processes, while 6-12 denote to management and support processes.

1. Vision & Strategy - Social Networking
2. Products & Services - Product recommendations, social search
3. Marketing & selling - Social CRM, Trend spotting, Social marketing, Loyalty programs, Advertising, Business intelligence, Churn prediction, Reputation monitoring
4. Delivery - Production scheduling, customer data mining
5. Customer Service: Customer support
6. Human Capital - Internal social networking, recruiting, expert routing
7. Information Technology - Resource allocation, Information sources, Content Management
8. Financial Resources - Customer & product strategies, Customer-product mix
9. Property Management - N.A.
10. Environmental issues - N.A.
11. External Relationships - Public relations programs, Social networking, Media monitoring
12. Knowledge Management - Knowledge sharing, Internal social networks

Social Network Data

Social network data is represented as a graph G=(V,E) with either explicit or implicit connections. Implicit connections can be discovered due to

1. follower, voting, or content (e.g. similar tags, comments, . ) relations in social networks
2. communication activities (e.g. calls, SMS, mails) in telecommunication networks,
3. proximity obtained from GPS location logs or RFID tags, or
4. user similarity.

Reputation and Trust

1. global trust (reputation), computed from the perspective of the whole network and yielding a single trust value, and
2. local trust, denoting to the trust from the perspective of another agent allowing for multiple trust values

PageRank is the most popular method of this family. PageRank can also be interpreted as a random surfer who follows links at random. The score of a node corresponds to the fraction of time spend at that node by the surfer in the limit.

HITS distinguishes between a hub and an authority score. Nodes that receive many incoming links from important hubs obtain a high authority score. Hubs with many incoming links from authoritative nodes get a high hub score. As all the other spectral ranking methods, the HITS value is determined by an eigenvector computation.

Incorporation of negative feedback.

The semantics of distrust propagation ("the enemy of my enemy is my friend") are in practice less effective than the semantics of trust propagation ("the friend of my friend is my friend").

Guha et al. (2004) show that a good heuristic for computing trust propagates

1. trust on a global and
2. distrust only on a local (single-step) level.

Network Dynamics

Research on the properties of social networks shows that new nodes connecting to social networks will preferential attach themselves to existing nodes with a high number of connections. This is regarded as an instance of the multiplicative process, also known as the Yule process or richer-get-richer process.

Research on mining evolving graphs identifies typical evolution patterns to obtain information on the underlying evolution mechanisms.

Business Applications

Social Network analysis allows partitioning customers according to their social network community. Applications of this kind of partitioning are

1. social recommendations (Amazon, Last.fm, etc.)
2. social search, that exploits context from a person's connections (e.g. Linux fans $$\rightarrow$$ a search for totem returns a media player)
3. marketing in offline settings (reward programs)
4. security (fraud and terrorism detection)

Propagation and Virality

The idea of word-of-mouth or viral marketing is targeting the most influential users within a network to start a chain-reaction that propagates the marketing message through the network. Research on information propagation has shown that finding optimal propagation strategies in social networks is NP-hard.

One research problem in this area is distinguishing between influence and similarity, i.e. do determine whether people follow a course of action because they have interests similar to the "influential person" or because they have been influenced by that person.

Some authors have measured the effects of adding viral features such as (a) personal referrals, and (b) automatic broadcasting into products.

Churn is the business term for the fraction of individuals that leave a customer base. Churn analysis targets on

1. determining the reasons for customers churn, and
2. predicting individual churn so that appropriate counter measures may be taken.

Social Networking Applications

1. promotion of products and services
2. trend monitoring
3. as a customer relation instrument
4. research of new product ideas
5. creation of customer user-groups
6. advertising
7. sponsoring of interactive content
8. creation and monitoring of online focus groups

Bibliography

Guha, R., Kumar, R., Raghavan, P. and Tomkins, A. (2004). Propagation and trust and distrust. In proceedings of the 13 th International Conference on World Wide Web (WWW'04). ACM Press, New York, 403-412

Ziegler, C.-N.and Lausen, G. (2005): Propagation models for trust and distrust in social networks. Information Systems Frontiers 7, 4-5, 337-358

Updated: April 14, 2013

Share on

You may also enjoy

Big, Linked Geospatial Data and Its Application in Earth Observation

less than 1 minute read

Integrating earth observation data with linked open data would pave the way for easy reuse and integration of these datasets. The article discusses how knowl.

Employment relations: a data driven analysis of job markets using online job boards and online professional networks

less than 1 minute read

Career websites contain valuable data on employees, their skill sets and, employment history. This article uses k-means clustering on keywords describing ski.

Suffix array

The suffix array is a memory-efficient alternative to the suffix tree which provides a sorted list of string indices indicating the string’s suffixes.

Dynamic feature scaling for online learning of binary classifiers

less than 1 minute read

This article describes and evaluates different online feature scaling approaches and their impact on the performance of binary classifiers. online feature.