AI is not yet a slam dunk with sentiment analytics – ZDNet

When we look at how big data analytics has enhanced Customer 360, one of the first disciplines that comes to mind is sentiment analytics. It provided the means for expanding the traditional CRM interaction view of the customer with statements and behaviors voiced on social networks.

And with advancements in natural language processing (NLP) and artificial intelligence (AI)/machine learning, one would think that this field is pretty mature: marketers should be able to decipher with ease what their customers are thinking by turning on their Facebook or Twitter feeds.

One would be wrong.

While sentiment analytics is one of the most established forms of big data analytics, there's still a fair share of art to it. Our take from this year's Sentiment Analytics Symposium held last week in New York is that there are still plenty of myths about how well AI and big data are adding clarity to analyzing what consumers think and feel.

Sentiment analytics descended from text analytics, which was all about pinning down the incidence of keywords to give an indicator of mood. That spawned the word clouds that at one time were quite ubiquitous across the web.

However, with languages like English, where words have double and sometimes triple meanings, keywords alone weren't adequate for the task. The myth emerged that if we assemble enough data, that we should be able to get a better handle on what people are thinking or feeling. By that rationale, advances in NLP and AI should've proven icing on the cake.

Not so fast, said Troy Janisch, who leads the social insights team at US Bank. NLP won't necessarily differentiate whether iPhone mentions represent buzz or customers looking for repairs. You'd think that AI could ferret out the context, yet none of the speakers indicated that it was yet up to the task. Janisch stated you'll still need human intuition to parse context by formulating the right Boolean queries.

The contribution of big data is that it frees analysts of the constraints of having to sample data, and so we take for granted that you can sample the entire Twitter firehose, if you need it. But for many marketers, big data is still intimidating.

Tom H.C. Anderson, founder of text analytics firm OdinText observed that many firms were blindly collecting data and throwing queries at it without a clear objective for making the results actionable. He pointed to the shortcomings of social media analytic technologies and methodologies providing reliable feedback loops with actual events or occurrences.

For that reason, said Anderson, social media analytics have fallen short in predicting future behavior. There's still plenty of human intuition rather than AI involved in connecting the dots and making reliable predictions.

Many firms are still overwhelmed by big data and being overly "reactive" to it, according to Kirsten Zapiec, co-founder of market research consulting firm bbb Mavens. Admittedly, big data has largely made sampling and reliance on focus groups or detailed surveys obsolete. But, warned Zapiec, as data sets get bigger, it becomes all too easy to lose the human context and story behind the data. That surprised us, as it has run counter to the party line of data science.

Zapiec made several calls to action that sounded all too familiar. First, validate the source, and then cross validate it with additional sources. For instance, a Twitter feed alone won't necessarily tell the full story. Then you need to pinpoint the roles of actors with social graphs to determine whether the voice is thought leader, follower, or bot.

Zapiec then made a pitch for data quality: companies should shift from data collection to data integration mode. We could have heard the same line of advice coming out of data warehousing conferences of the 1990s. Some things never change.

Of course, there is concern over whether social marketers are totally missing the signals from their customers where they live. For instance, the "camera company" Snapchat only provides APIs for advertising, not for listening. So could other sources or data elements make up the difference? Keisuke Inoue, VP of data science at Emogi, made the case that emojis are often far more expressive about sentiment than words.

But that depends on whether you can understand them in the first place.

Excerpt from:

AI is not yet a slam dunk with sentiment analytics - ZDNet