Building Effective Frameworks for Social Media Analysis

Building Effective Frameworks for

Social Media Analysis

Agenda

• Social Media: An INT perspective• Common Analytic Pitfalls• An Analytic Framework• Case Study: Brand Management

– Problem Definition– Source Selection– Data Capture– Data Reporting– Data Analysis

• Ways Forward, Future Analysis• Questions?

Intelligence

• Intelligence is information that has been transformed to meet an operational need

Data Intelligence

Operational Lens

Intelligence CycleNo matter what method you use…

…analysis is an iterative process

Social Media: The INT Perspective

HUMINT

OSINT SIGINT

Social Media gets the best and worst of three disciplines:

– HUMINT• Pros: Reveals intentions• Cons: Can be unreliable

– OSINT• Pros: Fast, Accessible• Cons: Noise

– SIGINT• Pros: Network, High Volume• Cons: Noise

Social Media Analysis Goals

• Need to have an end-goal with value to the organization (operational lens)

• Need to ensure cyclical feedback occurs from collection, processing, analysis, and consumption

• Need to make sure that a particular network is the right source for the task

Common Misconceptions

• Social media is not a panacea– Not everyone uses social media– Users of social media use it unevenly– User behavior changes based on situations

• Just because people can talk about anything does not mean they talk about everything all the time.

Common Pitfalls

• The important thing is often not what people are saying… but why they are saying it.

• Reporting tools rarely help dig into the why.• Many common tools, reports, and metrics are

actually misleading:– Word clouds atomize message context– Sentiment metrics are often highly inaccurate– Information in aggregate hides more than it reveals

Dangers of Disintegration

Source: Matthew Auer, Policy Studies Journal, Volume 39, Issue 4, pages 709–736, Nov 2011

Analytic Framework

• Data Capture (DC)• Data Reporting (DR)• Data Analysis (DA)

– 1. What to measure– 2. What the data is saying– 3. What should be done based on the data

Source: Avinash Kaushik, Occam’s Razor Blog http://www.kaushik.net/avinash/web-analytics-consulting-framework-smarter-decisions/

Analytic Framework

Capture

Reporting

Analysis

Choosing a Platform

• Social media is still new, evolving; and so is how we use it.– Static approaches to social media are flawed

from the outset– No one metric or set of metrics will always let

you know what is happening

• Need an adaptive platform to facilitate data capture, reporting, and analysis

Case Study: Brand Management

• Industry: Gaming– Experiencing 10% growth annually– Overall revenue expected to exceed $80

billion by 2014

• In May, Zenimax Online Studios announced Elder Scrolls Online– Elder Scrolls V: Skyrim 2nd largest game of

2011

Problem Definition

• As a brand manager, how can I use social media to track and understand public attitudes toward my product?

• Challenge is getting relevant information– Query too large = false positives– Query too small = miss potential information

Source: Twitter

• Twitter has some of the best analytic potential– High volume traffic– High volume user-base– Open API

• Not without limitations:– 140 characters– Limited historical / lookback

Platform: Infinit.e

Collecting Storing Enriching

Retrieving Analyzing Visualizing

Unstructured documents &

Structured records

Infinit.e is a scalable

framework for

Platform: Infinit.e

• Infinit.e supports the extraction of entities and creation of associations using a combination of built in enrichment libraries and 3rd party NLP APIs.

Data Capture – Initial Query

• Twitter search for “Elder Scrolls Online”– Simplest possible way to access information– RSS feed for 10 days (Jun 27 – July 6 2012)

Data Capture - Tagging{ "_id": "4fea6ddce4b0fa6316c7e07a", "communityIds": ["4fce07a1e4b06dc8a9107f3b"], "created": "Jun 26, 2012 10:20:12 PM", "description": "Twitter search for \"Elder Scrolls Online\" - started 6/26/2012", "extractType": "Feed", "tags": [ "games", "social", "entertainment" ], "title": "Elder Scrolls Online - Twitter“ "url": "http://search.twitter.com/search.rss?q=Elder%20Scrolls%20Online", "useExtractor": "AlchemyAPI-metadata", "useTextExtractor": "none“ ...}

Data Capture – Entity Map

Hashtag TwitterHandle URL

Unstructured Keywords

Time / Date Stamp

WhoTwitterHandle

WhatHashtags, Keywords, URLs

WhenTime, Date

WhereGeo (if Available)

Data Reporting

• Used Infinit.e’s Flash U/I Widget Framework– Document Browser (Individual Tweets)– Entity Significance (Top Entities)– Sentiment (Top Entities w/ Sentiment)– Query Metrics (Breakdowns of Query Results)

• Framework allows for additional visualizations to be constructed as needed

• Export options also available for manual review (e.g. graphml, excel, pdf)

Data Reporting

Data Reporting

Data Reporting

Data Analysis

• Analysis needs to be rooted in the operational need:

“How can I use social media to track and understand public attitudes toward my

product”• Emphasis on hypothesis generation,

testing, and experimentation

Data Analysis -> Capture

• Hash tags from an initial subset of Tweets fed back into the initial query

Twitter

Initial Query Results

Expanded Query Results

Data Analysis - Hashtags

• Top hashtags were almost all generic / more abstract– Undermines tracking and

understanding– Top hashtags tied to

franchise, not to the game

Data Analysis - Sentiment

• Converted URLs into derivative sources• 35% additional sources• Larger text sources offer potential value with

sentiment analysis that tweets alone cannot offer

Data Analysis - Sentiment

• Top negative and positive scores provided glimpses into aggregate attitudes

• Provide starting points for additional analysis

Data Analysis - Recommendations

• Actionable recommendations allow decision makers to make changes

Future Data Analysis

• Initial conclusions should be starting points for new analysis

• Broad entity capture allows for:– Key influencer identification– Clustering of tweets for segmentation– Map / Reduce for aggregate functions

Infinit.e’s Hadoop Integration

Expandable Model

• Identify key influencers on specific topics• Look at relationships between websites /

blogs and Twitter use (cross-network analysis)

Counting and Summing

• “Traditional” business intelligence analytics problems solved using aggregate functions:– Sum– Count– Average– Min– Max– Etc.

Clustering - Topic

• Topic Extraction– Key words -> Categories– Categories -> Related Categories

Key Value

graphics gameplay.pdf

story gameplay.pdf

company corporate.txt

… …

… …

Keyword Topic

graphics graphics

screenshots graphics

resolution graphics

quests story

zenimax company

… …

Clustering - Geo

Take-Aways

• All data providers can and do change their formats; users flock to and abandon platforms – what works today may not work tomorrow.

• Whatever platform you choose to do analysis, make sure it’s open and adaptable or your investment may degrade over time.

Take Aways (Things to Avoid)

• Data puking (less is more)• Metrics that cannot be tied to actions• Visualizations / reports that remove

context• Taking dashboards at face value

Take Aways (Things to Do)

• Segment data rather than work in aggregate• Look for the why behind the message• Always return to the source material• Explore alternative explanations• Always consider the ultimate goal

Thank You!

github.com/ikanow/Infinit.e

Andrew Strite

www.ikanow.com

[email protected]