Finding the human/AI balance in text analytics at KMWorld Connect 2020
At KMWorld Connect 2020, AI in action was examined from different perspectives. Alison Bushell, director, analytics, Confirmit, and Fang Chen, principal analytics consultant, Confirmit, presented a talk, titled "Making Experience Text Analytics Actionable: The Human/AI Balance."
Confirmit provides solutions and technologies for organizations running customer experience and employee engagement programs, and for market research agencies that provide services to their customers.
Bushell and Chen were joined in the session by Art Murray, CEO of the consultancy Applied Knowledge Services, who presented a talk on how to improve project management with text analytics.
Bushell and Chen have made their slide deck available on the KMWorld Connect 2020 website and a replay of their talk will be available in the next several days.
According to Bushell and Chen, if you're not analyzing your customer speech and text, you're getting an incomplete picture, which could be misleading your decisions. Speech and text analytics can be a valuable source of opinion, solicited and unsolicited. You risk making decisions on incomplete information if you don't take these resources into account.
Furthermore, it takes strong human intervention to align text topic categories to your business decision areas. Additionally, a great text analysis should be living/breathing, ensuring it stays relevant as your business and markets change. In this presentation, they helped attendees understand what steps are needed to make text analytics truly actionable.
Today, there is a data overload with volume and variety
- Structured v unstructured data
- Semi-structured text
- Acoustic + Linguistic Text Analysis together = tonal/inflection + content analysis
- Amount of untapped info here is HUGE
- Image recognition/mining
- Metadata is critical
Use AI for assistance, not to replace, Bushell and Chen advised.
Categorizing your data can be a daunting task and as a result, organizations are turning to AI-powered technologies to speed up the process. However, they cautioned, AI is not a solution on its own.
An AI based approach saves time and may identify previously unknown topics BUT disadvantages are low precision and recall, results are not actionable/not business-aligned, and it is not easy to explain or troubleshoot. Rule-based approaches are high precision, actionable/business-aligned, and are easy to explain and troubleshoot BUT they require human expertise, are not as quick, and there is a risk of overfitting and potentially missing out on topics.
Bushell and Chen concluded with a checklist for optimizing text analytics.
Among the key points to remember, they said, is that you don't need to reinvent the wheel. Hard lessons are expensive and competencies are available to help you succeed. Do your due diligence and understand journey maps, org charts, strategic and operational plans, KPIs and company metrics, what executives are on the hook for, etc., and keep in mind the long-term plans. Draft a framework but you don't need just one. It makes sense to have multiple categorization frameworks for different outcomes. And, finally, after you build out a text analytics model, keep checking it and refining it.
In a talk at KMWorld Connect 2020, Art Murray, CEO of the consultancy Applied Knowledge Services, looked at how to improve project management which often tends to be rearview mirror-oriented
19 Nov 2020