Chatterbox: Machine learning meets our social media emotions

Brands think they know their audience and customers. But few brands profile their audience and truly understand what the conversation is about. When a crisis breaks its often too late to stem the flow of angry comments.

Chatterbox, a UK social intelligence start-up founded in November 2011, uses natural language processing across eight European languages, Mandarin Chinese and soon Arabic. It searches and crunches through social data to get more business relevance in real time.

stuart battersby
Stuart Battersby, CTO Chatterbox Credit: Chatterbox

Chatterbox is an enterprise solution that uses machine learning to provide deep relevant social intelligence to sales, service and marketing teams in large companies.

Machine learning and natural language processing is built into the solution. Its sentiment analysis is built to directly target social data sets.

In blog posts and pages, languages are very structured with easy to understand sentences. Social conversations on the other hand are peppered with slang, acronyms and short form words.

Machine learning learns the type of language people use in their day to day social conversations which enables more accurate and faster insight.

This insight goes beyond the normal positive and negative sentiment engines. Chatterbox aims to understand what angers people, what excites them, what worries them – and at what time this is happening.

The tool extracts different types of opinions. It looks at phrase analysis and topic mining. It then uses this knowledge to alert the organisation to why opinions and sentiment changed and what is driving the emotional response.

Phrase analysis automatically indexes the content of unstructured social messages to tell you consistent themes and phrases. It is used extensively for competitive intelligence and insight.

Chatterbox's topic mining analyses your social data set to automatically group similar messages together. Identification of strategic themes gives the relevant insights to the business.

The company can then use this automated insight to perform actions. If there is a reduction in sentiment it could potentially be narrowed down to either a fault in the product, a missed sales opportunity or an advertising campaign that has gone wrong.

Chatterbox also looks at the people who are talking about the product or brand. Influencers can be surfaced, competitor conversations tracked outside of the usual high profile brands and individuals that are driving opinion within the community.

The company is also looking at the conversations people have when they are about to buy a product. Its Path 2 Purchase technology filters through the deluge of social data to identify individuals exhibiting signals of an intended purchase.

Machine learning has the ability to identify and learn about differences used in different industry sectors such as finance or technology which each use different forms of industry-specific or regional language.

“We take this data and try to understand meaningful insights” Stuart Battersby, Chatterbox CTO

Chatterbox focuses on data science. It is backed by five years of academic research from Queen Mary University in London with input from Stanford, California.

Although it uses social media datasets, Chatterbox does not position itself as a social media company.

The engine can take input from any form of social data inside the organisation – wherever people have conversations.

If there is a dataset inside the organisation then Chatterbox can provide the social intelligence around that information.

“We take this data and try to understand meaningful insights” Stuart Battersby, Chatterbox CTO said. “We are trying to understand opinions, topics, who the people are within this data to provide an actionable insight across the enterprise”.

Sometimes TV shows intend to create disruption amongst viewers. “Sentiment analysis will not give you the full range of emotional arousal around the product. We look at the excitement in the messages by learning the types of language people use when they are expressing excitement and measure these across the social dataset” he added.

Credit: Chatterbox

Far from the ‘one size fits all’ solution each installation is bespoke with an isolated architecture for each enterprise.

Running on Linux virtual machines, Python is the main coding language. MongoDB databases are used for the data.

It uses multiple cloud providers, Amazon for its business critical information, Microsoft Azure in preview and the Linode Cloud.

Chatterbox has a long line of failover and disaster recovery options.

Amazon cloud services are used for geographic failover with the data copied to another cloud provider so that if one geography goes down it can keep supplying customers.

In 2011 the company collaborated with Intel which provided one of its supercomputers for use, free of charge.

It has attracted £385,000 investment so far from Telefonica, Queen Mary University of London, Technology Strategy Board and CreativeWorks London. in June 2012 it won a place on the global Wayra Academy for start-ups.

So will Chatterbox dig deeper than standard sentiment monitoring engines such as Onalytica, SM2, Radian6 and Social Media analytics by NetBase? Will machine learning ever be able to accurately prevent social disasters? Or will it help to minimise the effect for brands once the genie is out of the bottle and the damage has been done?

Whatever the solution, brands need to become smarter about customer conversations to avoid  potential social media meltdowns that can harm a brand.