16 Aug Battling Conversational Noise With NLP
Every communication and collaboration system is inevitably noisy. No update is lost, everything is brought to your Inbox. Reliable message delivery is good but surely not helpful in this case.
An always connected digital age overestimates our ability to attend to everything that needs our attention. Computing power gets cheaper, bandwidth availability increases even on mobile devices and inventive minds conspire to bring more data to our senses. All the data one can eat, but our capability for information consumption has remained the same over millennia.
Bottom line, there is no Moore’s Law for human cognitive capability. We need help and it better be quick.
How Social Networks Deal with Too Much Data
Social networks like Facebook, Twitter etc deal with this problem of too much data in a couple of ways. You add friends or follow specific people to begin with. The popular “Like/Unlike” button, “Favorite” flag and the “Share/Retweet” buttons are all means to express your opinion. With even a 50 moderately active friends, the amount of information pushed at you becomes crazy.
Social networks use a “recommendation algorithm”, or conversely, a “noise filtering algorithm” to make your experience manageable. A piece of code determines whether an update is relevant to you based on your preferences, history of content that you have seen or engaged with. Each article is assessed for its likelihood of being liked or engaged with by you.
Whether its Netflix recommending movies to watch, FourSquare recommending nearby establishments to visit, or Amazon recommending Books you might like..there is a whole load of “recommending” or “filtering” going on. Which begs the question, why do our collaboration systems not do the same? Why does our Email client treat every “fyi” or that critical “order cancellation” mail with the same priority? Why does every update on our internal collaboration tool get treated the same?
Recommendation in Everyday Collaboration
So how could a recommendation algorithm work while collaborating with internal and extended teams? Imagine the following scenarios playing out in your email client or collaboration tool.
A senior partner is unable to make it to an important client meeting. A vendor refuses to reduce price for a critical component. Your key team member will not be in next week. You had promised to provide feedback on a proposal by monday, and happen to have forgot about it. Legal team wants confirmation on changes to a specific clause in a contract document.
In each of these cases the mail or update is a signal to adapt your plan. Perhaps look for an alternative senior partner to attend the meeting, or hunker down to finish that proposal review, or get a junior team member to stand-in next week or respond to the query and so on.
A few characteristics of these updates. They are unstructured text, “bad news” to a varying degree, can involve “time” expressions, could be queries that need responses and so on.
An effective collaboration tool would help you sift through a hundred updates and flag updates that matter. Which item should be acted upon, or responded to, immediately? The tool would act like a private secretary, bringing critical items to your attention, and not make a fuss about any routine updates.
CollabLayer – Smart Collaboration
So how do we approach this problem in CollabLayer? Every conversation within our application is analyzed for negative sentiment and flagged. Time expressions parsed, alerts automatically created and brought to your attention at the right time. Queries, or interrogative statements, are flagged too. Instead of trawling through many routine updates, we make it easy for you to focus on what matters.
CollabLayer is our vision of what a smart collaboration tool should be. Personalized, proactive and able to understand context. A lot more remains to be done. For example, our Outlook plug-in makes it super easy to view and act upon updates.
Initial feedback from early adopters has been encouraging. A bunch of natural language processing algorithms, along with a custom corpus trained classifiers, work behind the scenes in near-real time to make collaboration a little more easier. Technical details powering this capability is topic for a future post.
Point to note, currently our NLP features are tuned for English, support for other Latin languages is on our roadmap. If you have a specific language preference, drop us a line via connect at tataatsu dot com.
We could not be more excited to bring this recommendation/noise filtering capability. Just so you know, CollabLayer is free during beta. We encourage you to visit our product site or head to the signup page. We are eager to answer any questions you may have on CollabLayer.