Opinion mining
AI is good at precision tasks, not general teaching or training. With that mantra in mind, 'opinion mining' is one such precision technique. We can now mine social learning data for actionable insights.
With the rise of LXP systems the evaluation of deliverables and data jumps from the impoverished world of happy sheets and SCORM completion to a completely different level. We literally go from 10 to 500 miles an hour.
AI is good at precision tasks, not general teaching or training. With that mantra in mind, 'opinion mining' is one such precision technique. We can now mine social learning data for actionable insights.
With the rise of LXP systems the evaluation of deliverables and data jumps from the impoverished world of happy sheets and SCORM completion to a completely different level. We literally go from 10 to 500 miles an hour.
We can also evaluate relatively unstructured data, getting rid of surveys and time consuming questionnaires. AI NLP (Natural Language Processing) techniques are far more suitable, as you will harvest data and opinions that are much richer than traditional survey data, completion rates and scores. You will be delivering dynamic learning that needs dynamic tracking. and evaluation.
How do we do this?
At WildFire we have been working on this for some time, looking for the most suitable techniques to use in learning. The level of granularity is important here. Let’s say you want insights on the content, teaching, time taken, relevance, or more generally anecdotal and miscellaneous opinions, you may need to be very granular, down to the word level. There is a pay-off here as the more granular you get the less context you have.
So there’s a whole number of separate steps, where one has to use different forms of software to get rid of extraneous words (such as pronouns), tag with aspects, set sentiment scores to opinion words, watch for the presence of negation words, and so on. Some of this software will be pre-trained using large corpuses of text. It is useful to know whether, for example, Wikipedia was used, as it is more relevant to knowledge based learning courses. Or one can build your own corpus of data to make this more targeted over time.
Actionable insights
You run your social data through this process, which can take a while (optimization is usually necessary) and out pops positive and negative sentiments on the aspects you identified.
Note that this can be applied to all of your social data, just one module a subset of your learners or individual learners. You have to be clear about what you want to get out of this in terms of decision making. Actionable insights can be selected in terms of their ranked importance; positive, negative, miscellaneous and so on.
We are on the verge of a revolution here but it needs careful handling and expectation setting. This mat, at last be the door through which learning professionals can walk, with data, techniques and insights relevant to learners, teachers, senior managers and employers. Actionable insights that allow decision making to happen, rather than the reactive delivery of yet another course, without really knowing what is actually going on.
Conclusion
Technology is always ahead of sociology, which is always ahead of pedagogy. This allows us to more align all three, as the technology feeds actionable insights to real people, quickly. This is not turning humans and learners into data points, it is identifying all that is human about their learning journeys. If you’re interested in using this service, contact us here… WildFire
