Power, Stories and Application of Social Media data

The Power of  Twitter Data

Social Data Stories

Government Social Media Use

I attend a social media data meet up at the Twitter HQ on 02/04/2015. The above topics were covers and my learnings are below.


 The Power of Twitter Data

@RCalmond from GNIP

What makes twitter unique and disruptive?

  • Conversational
  • Real Time
  • Public
  • Distributed Globally

“Twitter data has unlimited value and near limitless application”

A tweet may only be 140 characters, but … it has over 65 data elements (JSON file)

Devices types and Geo-locations

Using Device (Generator) and Geo locations (GPS) – Mapbox plots 18 months of twitter data. This represents London – red represents iphones and purple is blackberry. From looking at theses maps on cities around the world they found that iphones are in rich areas. Androids in less wealthy areas. The blackberries are the only phones in Indonesia (being shown Jataka sorry no picture). Produced by GNIP. This social geo layer has been used to help mobile networks.

IMG_8129[1]

Saving lives

Datagram gave data to academics to the University of Wollongong to help with Jataka floods. Residents were asked to tweet #flood and location data which helped command centre turn off power when the water became knee high.

Identify food borne illness

When analyzing the mass of tweets – associated words help to understand

Business problems

Enterprise can be applied to everything

IMG_8131[1]

Better campaigns – Example given on special “Mountain Dew” of which enthusiasts were found on twitter which increase response by 83%.

Product development

Example: Samsung mined all the complaints/weakness/issues on twitter about the iphone and used it to develop their new phone and focus the advertisement campaign on long battery life, resilience and cost.

Building a brand

Identifying unique opportunities

Example: Dove used the stat from twitter that 5 million negative comments are from women about their body. They used this to create the campaign #speakbeautiful. They used a company called “Lithium” to catch tweets on the negative topic to response to the individual with the speak beautiful hashtag.

All the above can not be achieved by the normal twitter API but Twitter are active in projects help people use twitter data for good.


 Social Data Stories – Research approaches to make the most of social media data

@abc3d (Pulsarplatform – social intelligent platform)

“Social Media data is not quantitative data, rather qualitative data on a quantitative scale”

1.  Safeguard Privacy

2. People are individuals

tumblr_inline_msboe9mhNe1qz4rgp

3. Social Media is not mining rather data surfacing

Look at connections between topics

4. Social Media data makes more sense in context not in a silo

Connect online and offline campaigns

IMG_8141[1]

5. The more focused the data the stronger the insights

Connect online and offline campaigns

6. Social Media is all about the Visuals

For example hotdog legs and that dress.

Instagram-is-down-just-describe-your-lunch-to-me

Imagga is a technology which can work out the topic of the image given itself its own accuracy rating.

IMG_8145[1]

5 ways to mine data

New ways to sample social data

  1. Topics – Track a set of keywords
  2. Audience – Follow a set of users
  3. Contents – Track a URL
  4. Location ( 1% geotagged can increase to 20% around attractions like sport events)

1. Topic tracking

Example : UN used to help make policies decisions

IMG_8146[1]

Example: Key word tracking – O2 – localized tweets to predicts network disruptions

Example: Key word tracking – Tesco – Mines what people forgot most before christmas and increase the stocks and display before christmas

2. Tracking Audiences

Mentions on organisation twitter account only represents 0.1% of the conversation that is a massive blindspot.

IMG_8151[1]

There is a need to connect brands in a 3D space on how your brand connects to the outside.

You need to identify communities and clusters of fans. Example is “News Night” discuss grouped by demographic inferences. The different colours represent difference demographics IMG_8152[1]

IMG_8153[1]

3. Content diffusion – URL tracking

How do video’s go viral?

Example is the Spaceman singing David Bowie – Green is Spaceman and the spread of tweets represented visually. From Pulsar tracking looking at Audiences Structures.

IMG_8147[1]

Great visualizations of viral spread on 4 topics from “Ryan Gosling won’t eat his cereal” to “Turkish Protest”

IMG_8149[1]

 4. Channel Performance

Syfy – power of social audience used to quantify audiences vs. other network

Looking at biggest Brand, average reach

5. Real Time Segments (dynamic segments)

Real time audience insights -For example when Margret Thatcher died.

IMG_8154[1]

50K readers were examined across the leading newspapers and ranked on response (negative, positive, neutral)

IMG_8155[1]

6. Hybrid – Social media in context

Ticket sales tracking (3 gigs”)  – Tweets lead to visiting website (Mentions lead to sales) Announcement 48 hours before helped.

Social media activity help predict ticket sales almost as strongly as visits to the live nation website.

IMG_8156[1]

Emotions links help spread content viral. High emotions lead to faster online spread

IMG_8157[1]


 Government Social Media Use for policy decision making by UK Gov

By Food Standards Agency

Allergies

Used thunderclap “the power of the crowd” to gain support and documented using storify.

DNP deaths

DNP deaths which is an illegal diet & muscle drug – used key twitter influencers to engage with the young audience by music artists. #StoppedSelling #StoppedBuying

Food safety week – “Don’t wash chicken”

Reach and mentions – actually behavioral change very difficult to measure

Food safety week – “Don’t wash chicken”

Reach and mentions – actually behavioral change very difficult to measure

Norovirus – “Winter illness”

The aim was to predict the outbreak using frequent illness keywords (removing unassociated words) group factor analysis. The sickness tweet peaks before the frequency of lab reports.

IMG_8132[1]

The predictive model was created where significant change in tweets helps predict lab reports.

IMG_8133[1]

How much data ?

  • Listening
  • Historical information
  • Predictive analytics (norovirus)
  • Mapping
  • Cost effectiveness and collaboration

Additional notes

Social media signals newsletter was recommended by Francesco

FunF- http://www.funf.org/

Used to extract data from smartphones  to examine phone activity in real time

5th May next meet up

Advertisements