Big Data: How parties mine votes in mountains of tweets
Source: Glen McGregor
Social media concept torn newspaper headlines reading marketing, networking, community, internet etc
A little-known branch of computer science could give political parties an edge in the next federal election by helping form strategy based on the millions of messages Canadians post on Twitter and Facebook each day.
Data analysts who help parties track voters electronically are increasingly using the techniques of “sentiment analysis” to extract meaning from the tsunami of opinions volunteered on social media, and target campaign resources at those most likely to be supportive.
The challenge for “Big Data” scientists working for the parties is to match social media accounts to the real people on the list of eligible voters and then accurately interpret their tweets or posts to figure out their level of political engagement and party preference.
When an individual voter’s sentiments are known, a campaign can make better decisions about whether to spend time and money trying to convince him or her to vote. The Liberals, for example, wouldn’t want to waste money sending direct mail or canvassers to the door of someone who regularly tweets support for Prime Minister Stephen Harper. But someone who posts an enthusiastic message about leader Justin Trudeau can be flagged in the Liberal database and sent an emailed reminder to vote on election day.
“Lots of people are throwing their opinions at us, and being able to connect online personalities to offline is the big nut we’re working on and trying to crack,” says Andrew Drechsler, who did analytics work on past Barack Obama presidential campaigns and is now vice-president of HaystaqDNA, a Washington-DC data firm.
The Obama campaigns, with their heavy focus on data, showed the power of statistical analysis to identify and rally potential supporters to give money and vote. In Canada, the Conservatives, New Democrats and Liberals are building up their own analytics capacity in advance of the next federal election, scheduled for October 2015.
With much smaller spending limits, the ability to focus campaign resources on particular voters is potentially more useful in Canada than in the U.S., where presidential campaigns enjoy enormous budgets set aside for voter-contact calls.
As only a fraction of eligible voters can be contacted by live telephone calls, it can be easier to solicit the views of some people through their social media presence, Drechsler says.
“I may never be reached on the phone but if you are able to connect me (to a social media or voter profile), you can say I’m somebody we know and who is a supporter or non supporter.”
Indeed, with more people abandoning land lines, conventional telephone polling fishes for opinion samples in a shrinking pool. But through social media, and Twitter in particular, more are expressing their opinions about politics in plain view. Not all Facebook messages are public and so can be harder to analyze, but can also be key to rounding out a voter’s profile in a party’s database.
In the U.S., data scientists also look closely at the networks that connect people on social media, as research suggests our views about politics may be shaped more by what people or our friends say than by advertising.
Using computer algorithms to glean the true meaning from these massive quantities of text written by humans is an inexact but improving science. Sarcasm and irony, found in abundance in social media, are particularly confounding, Drechsler says.
It can be difficult to write a computer program to understand the meaning of a tweet such as “I love Barack Obama as much as a root canal,” he says.
“For a human, it’s easier to look at a message and say this is positive or not, rather than go and write a program that does the same thing,” says Diana Inkpen, a computer science professor at the University of Ottawa who studies “text mining” techniques.
The science of sentiment analysis is improving as natural-language processing software becomes more sophisticated.
One blunt-force approach uses a dictionary of key words that are flagged either as positive or negative and counts how often they occur with certain keywords or hashtags, such as “Harper,” “Trudeau” or “#cdnpoli” �C the Twitter shorthand for discussions about Canadian politics.
A more sophisticated analysis involves sampling a small number of tweets or posts on a particular topic and having humans read them to discern meaning, tagging each as positive or negative. The computer is then “trained” to recognize patterns of words found in the tagged samples, then classify new text accordingly. Inkpen says this method can achieve about an 80-per-cent accuracy, if done properly.
Corporations are already using these sentiment analysis techniques to monitor costumer feedback and the vast quantity of product reviews posted online.
gmcgregor@ottawacitizen.com
Twitter.com/glen_mcgregor
The ‘Big Data’ Electon: About this series
The federal election scheduled for 2015 promises to be the “Big Data Election,” in which the ability to use sophisticated analysis of records to identify supporters and get them to the ballot box could prove decisive.
Saturday: How political analysts fill in your profile; Canadian parties build detailed voter-information banks.
Monday: How American political consultants perfected the use of voter data.
Tuesday: Just how accurate is the data Elections Canada uses on election day?
Wednesday: Looking for clues to your intentions in 140 characters or less.
Thursday: Cable boxes could soon be serving up customized political ads on television, similar to the targeted ads that follow you on your web browser.
| }
|