Unlocking the Potential of the Cloud for IBM Power Systems
SM&WA_S1-2.pptx
1.
2. Content
• Provides an overview of social media analytics
• Discusses various analytics methods for social media data
• Illustration of social media analytics
• Challenges are discussed
3. 1. Social media analytics
• Social media analytics refers to gathering data from social media
platforms and analyzing the data to help decision-makers address
specific problems
• Automated social media analytics is inexpensive and fast compared to
traditional media analysis, via which data collection is oftentimes
manual, and the analysis is labor-intensive.
• The same survey identifies marketing as the primary area of social
media analytics
• Innovation & product development, followed by customer service,
operations, and strategy
4. 2.Categories of social media analytics
Consideration criteria
(hunt, 1991)
• Adequacy of phenomenon specification
• Adequacy of specification of classification characteristic
• Mutually exclusive categories
• Collectively exhaustive typology
• Usefulness of topology
5. Kohli and Jaworski (1990)
Timeliness
Real time
analysis
Non-real-
time analysis
market
orientation
customer
competitor
6. Table 1. A typology of enterprise social media
analytics
Timeliness
Real-Time Non-Real-Time
Market Orientation
Customer
Real-Time Customer Social
Media Analytics
Reactive marketing efforts
(e.g., keyword analysis,
location analysis, conversation
analysis, complaint detection,
and alert from online review
or comments)
Non-Real-Time Customer Social Media
Analytics
Proactive marketing efforts
(e.g., identification of profitable
customer groups, social network
analysis, influencer analysis, web
analytics, sentiment analysis)
Competitor
Real-Time Competitive Social
Media Analytics
Operational intelligence
(e.g., monitoring of prices and
promotions, news alert,
headlines, new product
announcement, merger and
acquisitions)
Non-Real-Time Competitive Social
Media Analytics
Strategic and tactical intelligence
(e.g., periodic trend analysis of
competitors’ pricing, new product
development, technology
development, customer services,
complaints, employee comments)
7. 2.1.Customer analytics
Supported by commercial service providers
Metrics :
Size of social media pages
Response time to complaints
Number of engagements (comments, likes, shares for each post)
Demographics of the people connected to a user
8. 2.2.Competitive analytics
Method to gather and analyze data about competitors and the business environment; this may utilize
news analytics methods and sentiment analysis.
Tasks of competitive analysis include:
• Sentiment tone
How positive, negative, or neutral the tone of the content is about the competitor
• Relevance
How relevant or substantive the content is for the competitor.
• Keywords analysis
What and how many different keywords are used in the content about the competitor
• Intensity analysis
How repetitively the keywords are used over time.
• Alert analysis
Tracking special announcements such as technology breakthrough, product recalls, change of upper
management, and financial performance
9. 3.Social media platforms
• online communities via which members seek and share common
interests, activities, experiences, and information
• provide archival data and real-time feeds as well as sophisticated
analytics tools
• Social media platforms enable consumers to be content creators as
well as content users (Hajli, 2015)
• Social networking sites
• Blogs
• Content sharing sites
• Opinion sharing sites
10. 6.Social media analytics
• Social media platforms contain a wide range of data types and data
volumes
• Depends upon the social media platform and type of analysis
required by managers
Sentiment analysis
Social network analysis
Statistical analysis
Image and video analysis
11. 6.1.Sentiment analysis
• sentiment analysis mainly uses two methods:
1. a machine-learning method
2. a lexical-based method
• Sentiment analysis divides into the following specific subtasks
(Batrinca & Treleaven, 2015)
1. Sentiment context
2. Sentiment level
3. Sentiment orientation/polarity
4. Sentiment strength
5. Sentiment subjectivity
12. 6.2.Social network analysis
• social network theory
explain how networks operate and analyze the complex set of
relationships within a network of individuals or organizations (Scott,
2012, Wasserman and Faust, 1994)
• Understanding the dynamics of interactions between users can assist
in identifying influencers to target in branding and ad campaigns
• Well-connected users are particularly important for social networking
sites, as these users can be highly relevant for the promotion of
brands, products, and viral marketing campaigns
13. 6.3.Statistical analysis
• statistical methods typically require transformation of the original
contents into a coded format suitable for statistical methods
• Markov chain Monte Carlo methods, regression models, logistic
regression, factor analysis, and cluster analysis
14. 6.4.Image and video analysis
• Image analysis organizes images into an archive that is the fully
searchable and analyzable
• Statistical analysis of tag data, demographic data, and download
frequency (e.g., Instagram account’s average engagement per photo,
keyword analysis for comments, most active followers, top locations)
• Video analysis typically involves quantitative metrics such as the
number of users, response rate, subject, and location.
• The emotional state of the user, and applying a behavioral model to
spoken words to determine the personality type of the user.
15. 7.Processes for social media analytics
Stage 1: Develop key social media metrics
Stage 2: Choose, monitor, and listen to social media platforms
Stage 3: Perform social media analytics
Stage 4: Build social media intelligence
16. 8.Challenges
1. Bias in social media data
2. Selection of good social media metrics
3. Noise in social media data
4. Unstructured social media data
17.
18. Introduction
• 75% of marketers consider finding the right influencers the most
challenging aspect of an influencer marketing strategy
• Majority illusion
A paradox within social networks that makes some ideas, behaviors, or
attributes appear widespread even when they are not
• Well-connected members within our network can skew our
perception of how common an idea or behavior
• Location within a network plays a role in their potential to create the
majority illusion
20. • Being well-connected or strategically positioned within a social
network may impact one’s ability to influence others more than the
size of one’s following
• The structure of those connections matters, too. So marketers should
focus on two other factors that measure influence in a social network:
“betweenness centrality”
closeness
21.
22.
23. Conclusion
• To find influencers with high betweenness centrality, look for people
followed by other influencers who also have followers across many
verticals
• To find niche leaders, look for people with narrowly-focused interests
whose followers are similar to them but have a low number of
followers
• Timing : coordinating a handful of influencers to push your message,
product, or content around the same time to give the impression that
“everyone” is talking about your brand
24. Limiting the Spread of
Misinformation in Social
Networks
Budak, C., Agrawal, D., & El Abbadi, A. (2011, March)
Proceedings of the 20th international conference on World wide web (pp. 665-674)
25. Content
• Introduction
• Information propagation in social media
• Influence limitation problem
• Theorems
prove NP-hardness and sub modularity of influence limitation
• Experiments
compare performance of greedy algorithm with heuristics
• Algorithm
Predicting missing data
Limiting the spread of misinformation
26. Introduction
• Competing campaigns in a social network
• Online social networks
Benefits
Disruptive effects
• Social networks as a reliable platform
Tools to limit the effect of misinformation
• Contribution of paper
• Minimizing the number of people adopting misinformation
27. Related work
• The identification of influential users in a social network
• Different models of communication incorporating aspects of real
social networks
• Existence of competing campaigns
• Independent Cascade Model to capture the existence of competing
campaigns in a network
• Addressing the problem of influence limitation as opposed to
maximization
28. Problem definition
• A social network: directed graph G = (N, E)
• Context: influence spread
• N: users of the social network
• A node w is a neighbor of a node v if and only if there is ev,w ∈ E, an edge from
v to w in G
• pv,w is assigned to each edge ev,w which is used to model the direct influence v
has on w
• Objective : limit the lifespan of the “bad” information campaign
: maximize the effect of the “good” campaign in the
presence of the “bad” campaign
29. Eventual influence limitation
EIL: focus on minimizing the number of nodes that end up adopting
campaign C when the information cascades from both campaigns are
over
THEOREM 4.1. EIL is NP-hard even with the high-effectiveness property
The optimization problem is NP-hard
THEOREM 4.2. EIL is submodular when the limiting campaign L has
high-effectiveness property
provide approximation guarantees for a greedy solution for various definitions
of this problem by proving that they are submodular
30. Evaluation
• problem is NP-hard and therefore an exact solution is infeasible for
large-scale social networks
• two variations of the problem:
• two different communication models
• submodular
• a greedy method is guaranteed to provide a 1/(1 − e) approximation
• greedy algorithm is a polynomial time algorithm, it is still too costly
for large scale social networks
31. Experiments
• performance of the greedy algorithm, comparing it with 3 different
heuristics
• in many cases, the performance of heuristics, even the simple degree
centrality heuristic, is comparable to the greedy algorithm
• different aspects of the problem explored
• the effect of starting the limiting campaign early/late
• the properties of the adversary
• how prone the population is to accept either one of the campaigns
32. Algorithms
• realistic problem of influence limitation in the presence of missing
information
• Predictive hill climbing approach
• first predicts the current state of all the nodes of the network
• uses the hill climbing approach to choose the set of “influential” using the
predicted data
• optimization algorithm
• choose the set of nodes that are most likely to have been infected by the “bad
campaign”
33. Conclusion
• for most cases, the predictive hill climbing approach provides good
performance, within 96-90% of the performance that would be
achieved with no missing information
• Although for small delays the performance is consistently within 96-
90%, for large delays
• the performance degrades to 75% when the number of missing
information increases dramatically, i.e. states of the nodes are known
with only a 0.01 probability