By Sameep Mehta
Positive, negative, funny, serious – anything but neutral. That about sums up the typical sentiment around any political election, especially when considering what people write on Twitter. To find out, ABS CBN, a television news network in the Philippines, teamed up with IBM Global Business Services and IBM Research – India to analyze public Twitter chatter about candidates and related topics around the country’s recent mid-term Senatorial elections.
The social media tracker tool uses Natural Language Processing and Text Mining Technology (similar to how IBM Watson read and understood text) to comb through the millions of tweets about all manner of topic around the elections. ABS CBN then used the analysis, which filtered to them in near real-time to generate new articles for their website as part of their election coverage.
Uncovering election trends
Apart from the standard “velocity” and “volume” aspects associated with Big Data, the team also had to develop language models to understand Tagalog – including slang. Then, the tool had to accurately categorize the thousands of tweets pouring in every hour about everything from candidates to broken voting machines.
The tool caught a spike in negative tweets about malfunctioning precinct count optical scan (PCOS) machines for which ABS CBN quickly developed TV and web news stories.
“Twitter was peppered with words ‘defective’ and permutations of the key word ‘malfunctioning’ opposite the word ‘PCOS’. There was even a play on ‘hocus pocus’ and ‘PCOS’, thus the coinage ‘hocuspcos’,” ABS CBN reported from the 10,070 PCOS-related tweets from between 8:00 a.m. and 12:00 p.m. on Election Day.
Their ability to quickly cover trending topics even put them in the spotlight: “#Halalan2013 (their news agency name) [was] the only news coverage-based hashtag for the Philippine elections to have landed in the trending topics.”
Discovering election influencers
Getting talked about on Twitter did not necessarily translate to votes for candidates – even when popular celebrities tweet their support. In fact, while the most overall chatter about the candidates swirled around what 2.7 million-follower-big comedian Vice Ganda tweeted, he did not talk about or support the most-influential candidate in social media, Maria Lourdes Nancy Sombillo Binay, or – more importantly – the top vote-getter, candidate Mary Grace Poe Llamanzares.
ABS CBN reported that “despite the traffic generated by Vice Ganda … it was still Nancy Binay who was most-talked about. According to the keywords analyzed by the Tracker, many posts about the vice president’s daughter were about her birthday on May 12 (elections were held on May 11). Also being talked about was her tiff with Vice Ganda.”
And when put head-to-head, Binay had 78 percent share of election-related tweets (almost 450,000 tweets), compared to Poe’s 22 percent (less than 150,000 tweets). But as of this writing, the results have Poe leading Binay 16 million votes to 13 million. Perhaps the 14 percent of those 450K tweets with a negative sentiment toward Binay influenced some voters. Of Poe-related tweets, only 6 percent were negative.
In another example, the Tracker found that Senator Miriam Defensor-Santiago’s supportive “sex appeal” tweet of candidate Sonny Angara created a stir that did wonders for his social media attention, but perhaps mixed results at the polls:
“Stories by other news agencies, including ABS-CBNnews.com, were also reposted by netizens the same day. People online who agreed with Santiago gushed about Angara’s looks, and posted how they will vote for him because of his good looks and achievements.
“Others, however, criticized basing how much a person deserved a vote because of his looks,” ABS CBN reported.
ABS CBN continues to tease out the details. You can compare the candidates’ social media influence, versus where they stand in current election results on Halalan’s social media battlefield site.
Social sentiment from politics to retail
Since helping ABS CBN cover the Philippines’ elections, retailers want to use Social Tracker to find out what people across the globe think about their products. The tool’s ability to quickly derive insight about numerous related topics in any language could help do everything from find out why a product is popular (or unpopular), to improve online customer service.