Tag: Sentiment Analysis

Your next firm should be an IT firm.

I’ve interviewed with them all. The big firms, the small firms. Both in Brussels and in New York.

After working in the European Commission and Parliament, I wanted to go to the private side of Communications and Policy. The problem? I had been running data and insights. Using terms like NLP, Sentiment analysis, cognitive science and connotation mapping to describe what I did not work well. On the other hand dumbing down would leave me looking like another 27 year old who does “the social media” and or “the internets”. Most firms doing the interview for analytics positions wanted to hear “pivot table”, “engagement”, “influence” and maybe SPSS. Upping the hierarchy on those terms to explain why “something was” appeared unnecessary and impractical since explaining this to a clients communications director is another task with in itself.

So to hell with the PA/PR firms, I joined an IT firm that also does communications – Intrasoft International, and could not be happier.  The people’s skill sets are well defined and paid, which gives them a certain confidence in contrast. They like things such as analytics and completely understand them. Their only fault is the soft side of Communications – which is now my job to merge.

IT is going to take over communications sooner than later. Current  PR firms will be left scrambling. The lack of investment in the deeper meaning (abstract knowledge such as transference, retention and pragmatics) will start to show as data/connotation mining becomes a standard practice. Most IT firms already have this infrastructure in place via their AI departments.

There will be a point where dropping shit jargon is irrelevant and companies Comm Director, who should be more of a CIO/CTO in the nearer future, will see right though it.

My 2 cents. Also check out my presentation at chandlerthomas.com

CT

Analytics and the Global Political Environment

In the global political environment a holistic view is needed.

The standard for message delivery is high in the modern media environments. It’s vital to create communications that people identify and empathize with. The threshold to get attention, and to get people to retain your information, is even higher. Politics now competes with Pepsi, Nike and Apple. No one will take time to care about your views or propositions. Firms must be proactive in both controlling and framing language, as well as marginalizing their communication and political strategy for efficiency.

The problem: There are many variables to consider when developing productive communication and political strategies. Further, basing decisions on just educated guessing, even for the most highly skilled and experienced  professional, does not meet the standards of modern business practices, which use analytics to make decisions.

The Solution: Online media provides an immense amount of information which can be monitored. The data can be used  to track political and policy instances qualitatively, as well as forecast.

To do this, we must synthesize research in cognitive linguistics and Natural Language Processing (NLP). In the last few years these technologies have evolved to the point where pragmatics can be quantified accurately. And further, because of the amount of data an average person creates in a day, we have and endless amount of information to drill down into for examining political and cultural phenomenon.

I’m currently looking at the French elections. There will be more on that and this in a bit.

Active Communications

The idea of “Active Communications”  is based heavily on media monitoring listening data AND proper framing, which is a bit of an art (unless you have first class AI/NPL skills) . To work efficiently, a multi-channel infrastructure that allows for real-time content is paramount.

Important concepts to take away:

  • Timing and message coherency, to create a critical mass.
  • Trust the data for real time decision making.
  • Location based targeting with Fundamental,Technical and Sentiment analysis, per channel/medium.

Perhaps the biggest aspect is willingness to abandon prior methods after the data is received. You might be more comfortable with Twitter or another channel, but for the specific campaign it could be a  waste of time. Don’t fight up stream.

NLP: Value Greater than just Positive and Negative Sentiment.

Positive (P) and negative (N) sentiment are just transmitting two results based on many other variables. This in itself contains little knowledge. For creating strategy, it can only bench mark after the fact.

Inherently the use of data is to make better decisions for the future. Production is expensive and time consuming. So is deductive reasoning.  The main goals should be to move away from adjusting to the end reaction and migrate to a predictive model. Look at the context in higher resolution i.e. control for  variables which lead to PN sentiment.

Here are some suggestions

  • Time of day
  • Medium (Twitter,Facebook ,Blogs, Mainstream news)
  • Comments PN sentiment
  • New dissemination to comment count (time, PN)

If you are advanced, use NLP tools that allow for custom taxonomies – within the NLP,  to create rules on varibels like types of framing. On a medium level this is very good at prediction, but more on that later.

Ciao

Heuristics, Algorithms and Communications: Quick Thoughts

The use of heuristic and taxonomy algorithms for identifying cognitive bias in  communications is a game changer.  According to SAS,  companies that invest in analytics out perform the S&P500 by 64%. I imagine the margin for communication/marketing firms is greater. The market is not anywhere close to mature, and is lopsided to a few in the know. The rest toil in mediocracy via old connections/business that will surely run ground soon.
Today’s great business’s – for the most part, are data driven. Communications is perceived  to be educated guessing.. The use of programming for bias is important because bias cannot be out run. Think of this much like “Inception” and Moneyball combined.

Thanks to social networks we have the largest focus group in history. Analysts can see what people are using to explain how they feel and think – during  specific situations or events. This can also tell  how people interact within context of syntax or an instances. Very powerful and cheap information in contrast to polling or focus groups.

There are patterns/norms with how people engage on each medium. In addition to what type of syntax they use to convey a certain type of  thought. This is all programmed in the mind for the most part. Linguist like Lakoff and Chomsky suggest that brains are hard wired to favor certain patterns that convey meaning. This plays out accurately on-line consistently.

Just a random Sunday thought…Now looking forward to watching the NFL playoffs.

Ciao

Market Shares and Topic Correlations: European Parliament Party Leadership

The chart above shows how much associated EP Group  leaders have on a variety of EU topics market share of the leaders on different subjects, including their own party. It’s possible that all leaders can be in more than one article so the totals can be over 100%.  The Green line rank the subjects of all the leaders combined. The EU ranks 1st followed by the Parliament – which should be expected, and then the Euro. Joseph Daul (JD) has the most association for the EPP Group at 81%, which is a good indicator for further party branding but not for pan EU leadership , where he lags behind all others.

On the far right, the chart shows the average and over/under performance to the market share compared to Group Parliament seats.

  • Guy Verhoftstadt (GV) (ALDE) is +7,
  • Martin Schulz (MS) (S&D) is + 20
  • Joseph Daul  (EPP) is at -20.

This does not fare well for the EPP Group. When looking at the market shares of EP leaders in context to one another, JD is consistently in last place. Why does the EPP vastly underperform while the S&D and ALDE over perform? This can be due to a number of reasons.

  • JD does not speak English
  • The EPP being the largest group cannot utilize polarizing and thus mobilizing language without alienating many of their MEP’s.
  •  JD has chosen to stay out of the spotlight given that Jose Barroso and Herman Van Rompuy[1] are EPP.
  • MS is from Germany which is heavily discussed at the moment, and will also become the new President of the European Parliament replacing Jerzey Buzek[2].

Robert Fitzhenery (Head of EPP Group Press and who controls the communications and outreach budget) explained to me that the EPP Group cannot be too polarizing on political issues because of it’s size, and cannot risk alienating some MEPs. On the other end some in the Group want to be more polarizing to mobilize debate and heighten the Groups profile.

During my work with the EPP, ALDE and GV typically lead market shares on media despite only having an 11% (88 of 754 MEP seats) market share of the Parliament[3]. I talked with Neil Corlett, Head of Press and Communication for ALDE. Neil explained since ALDE was a smaller group, they decided to follow whatever GV wanted and not deviate from a few main points. In short their message is consistent via both GV and ALDE’s MEPs. It paid off. ALDE is outperforming the EPP and S&D. And both have more money. It’s only in the last two months that MS has been generating so much sentiment. It will be interesting to see if this approach pays off in the 2014 EP elections. It will also be a good indicator of how mature the on-line EU landscape is.

Communications as Productivity

By looking at Communications in terms of production and variables, you will save millions. By not understanding variables in communications, you will loose millions and time. Gone where communication firms can pass  Bullshit. Now it’s far from an art form with metrics. Clients should expect decision making based on sound research, in addition to  web analytics and online monitoring data.

The goal: Create the most productive syntax, which could be words, photos,video or interactive digital content. Not the most view and clicks

Resonance: 

Output time/date to channel dissemination, and how much time  that takes. It shows language adoption, which is how you win. There’s a saying politics, “Win the language battle, win the war”.  Resonance is a good KPI for knowing if the framing has retention, thus productivity.

Its can also be useful to reverse engineer a past event’s content and framing with machine learning to compare and contrast. Again don’t focus too much on “click and looks”. You might see exponential gain which perhaps are linked to offline events. That way you can also map correlation. I tend to see this pattern a lot working in politics. This reiterates the massive amount of information available on-line by listening.

More Tricks:

Use best practices in cognitive ICT, psychology,linguistics and behavior economics. A real expert will know captology, sentiment analysis and gamification, and not neglect fundamental and technical analysis.

When you understand the basic variables, you see that  each enhances or destroys productivity.

If the person you hired  doesn’t understand what I’m talking about, the strategy will loose productivity. Ask yourself, why we are paying 200-300 plus per hour? Educated guessing is over. The data doesn’t lie.

Abandon Your Communication Strategy

You or your organization is not special to the news cycle. It’s goal should be smarter, faster, cheaper. While it might not seem like it, faster -despite large investment coast, are 90% of the time cheaper. They save on opportunity loss/cost.

Many organizations and political groups write a long, dull strategy for their communications. The idea is out dated.  Today’s modern communication environment is fast. It’s real-time, and it doesn’t care about you. It’s not an option to rely on un-adaptable 10,000 word papers if the goal is to stay relevant.

Don’t worry.

If you are good at data science and have monitoring tools, you’ll mitigate risk.  Build a COMM infrastructure that can handle “real-time”. The main things is to trust in the Data (it doesn’t lie), and remember  the strategic advantage of real-time outweighs the majority of mistakes you could  make.

Framing

In politics controlling language is vital. George Lakeoff (Linguist -Cal Berkley) confirms brain will generate a physical cognitive bias if you win the language war. Because of the frame being repeated, the attitude towards the subject changes.

For example the party who introduced the term wins solely on sheer mass. This can be determined by the total count of the terms.

For example: ”Death Tax” or “Estate Tax”. We also know the terms that we introduced to the discussion. Hence the campaign works when the Language has been adopted. It might not seem like much but these are long term goals and or policy matters to keep or change a status quo.

Response to Conversation Impact

CI can be found here 

In looking at Conversation Impact I think – it’s simple and clear which is important for creating a KPI. More importantly CI can be applied

easily across numerous campaigns. I think the main thing that I would add to CI model is accounting for the message framing and looking at mediums as separate instruments. Much of my political work involves this since it can shape perception and offer insight to what type of conversation is going to have the best ROI per medium and influencer. Each tends to have a different pattern that can make it possible to measure CI more accurately. In short there’s a lot of information in each medium interaction and this needs to be clearly separated.

For Figure 1:

The survey instrument – do  surveys correlate to an accurate representation of sales and behavioural change? I’m always sceptical of asking for an opinion i.e. “would you now do x?” or “purchase intent”.

To gain insight I’ll look at online and offline (attending a conference or event) actions. The more complex the task the constituent is willing to go through the more important I’ll view the issue/policy (in a political context). Also If someone posts a comment – I value this more than a “like” or “Vote” or mentions put forth by media as it shows more effort. The question for us all is if this translates into a vote or purchase.

I tend not to rely on automated sentiment as much – in political post (generally negative) it makes it very hard. To counter this specifically in Radian 6 I’ll Boolean for negative terms. As long as it is consistent across everything (like in CI) it can none the less be a good measuring stick and I have done this before.

Figure 2

The graph which illustrates Influencers led to a 6.3:1 rise in network influence. The study that I read was about a 5:1 ratio – so in your case it was a bit higher. Was this sustainable past March 09? Also was the campaign coordinated on traditional mediums and perhaps cross referenced with something like Cision Point which accounts for traditional news? The data from this would be interesting. On and offline merging is what I tend to measure/compare the most.

Also in my work I tend to see exponential gains in sentiment/mentioned but it’s usually built off an off-line instance like a political speech/conference or elections (shown in my Political Intelligence presentation pg 20).
If they were measured what was the sentiment/mentions before after a main event (if there was) for “X” to happen – like a conference, election or product releases? What was the ratio in the gains?

I think the figuring out the market shares or volume is right on. I’ve looked at this extensively with political party mentions by share of seats in parliament across all mediums to measure under/over performance.

Some Thoughts:
I’ve always wanted to “gamify” a campaign to measure what options people choose with in a situation. As we funnel down the process/engagement from a click of an online advertisement, to downloads, to checking in to a place via foursquare etc. and or using an off/on line app like Layar; This could provide a good way to measure campaigns to see what way behaviour can be modified from the digital realm and how best to do it. This is why I love the study of Captology.

Ultimately while it’s easy to make things more accurate, but it also gets complex and becomes hard measure for a multitude of brands/products. I think Conversation impact does a good job of measuring specifically what it’s supposed to – conversation. My only suggestion would be to find a way to account for framing techniques on a per medium basis and what technique generates the most engagement (comments, likes, votes). If we as influence marketers can crack the cognitive and linguistic code would be the Holy Grail and I think looking at data is the first step.

Please feel free to ask any questions.