Your transcript just got smarter

December 14, 2021

Two new features help therapists better search and review sessions.

At Lyssn, we are serious psychotherapy geeks, and we take feedback very seriously. Our feedback doesn’t rely on crude keyword searches but is meticulously trained by thousands of human rated sessions, and repeatedly assessed for accuracy. So, when we add new types of feedback, it’s kind of a big deal.

In 2014, the Lyssn founders were the first to demonstrate automatic identification of  evidence-based practices in counseling. By 2019, we had tested the first end-to-end software solution for providing automated feedback on these practices to therapists. Along the way we massively upgraded our speech pipeline and added the first ever system for evaluating cognitive behavioral therapy. Today, we are excited to announce two new types of feedback that we’ve been working on for quite a while!

What do people talk about in counseling?

As a psychologist, we like to talk (and argue) about complex theories of how psychotherapy works. Turns out though, that clients come to psychotherapy for help with problems – and these problems are what they want to talk about. Many of my friends not in the therapy world sometimes ask, “So what exactly do people talk to their therapist about?” While it seems simple—they talk about their problems, their relationships, what they value—it is actually a surprisingly complicated question.

How we quantify topics in a therapeutic session

To be able to define what exactly is being talked about in a session, we began by consulting existing problem lists from various standard assessment measures and then qualitatively examined hundreds of sessions for different topic themes. We then created a topics coding system that our internal team used to label thousands upon thousands of statements in sessions for things like relationship concerns, discussions of mood, anxiety, hobbies, work, school cultural issues, suicide, self-harm, and more. We trained a Machine Learning (ML) system to automatically recognize these topics in sessions and found that performance was comparable to human agreement.

Search by topic feature

Building on the above work, our latest addition to the system is a set of topic labels – available both on our platform and in our API.  In the platform you’ll now see a list of topic options at the top of each transcript. The topic you select will focus the transcript on the parts of the session where that topic occurred. It’s important to know that the presence of a topic in a session does not suggest a diagnosis or any assessment of risk. It just means that the Lyssn AI thinks that issue was discussed at some length by the client and therapist.

It is our hope that our customers will find this information useful for reviewing session content in supervision, writing notes (if you weren’t able to write them immediately after the session), and for learning more about the sorts of things being discussed in a clinical setting – helping supervisors suggest areas of training that could be useful for therapists.

Your most empathetic moment feature

We’ve been helping therapists reflect and learn with instant feedback on how treatment was going in a way that was rarely available before. Now we are hoping to make that feedback even more engaging. We’re piloting a new metric – ‘Most Empathic Moment’, which we think is going to be pretty cool.

For a long while, our internal coding team has been highlighting the best moments we see in sessions – those moments when we are just in awe of the impressive work a therapist is doing. As a result, we have created an extensive dataset of moments when a therapist was highly empathic. Recently our ML team started exploring whether we had enough data to automatically highlight some of these moments in a session – and I’ve got to admit – I’ve been super impressed. The system isn’t perfect (what coder is!?), but it’s fairly amazing how often it identifies moments in the session where a therapist makes a particularly insightful reflection or finds a creative way to connect with their client.

We thought it was time to share it with therapists on our platform. As of today, you may start to see the occasional session transcripts labeled “You had a most empathic moment!” We’ve designed the system to be pretty picky, so you might not see it all the time, but if you do, click on the label and see if you agree with what we picked out. Do you agree?

We can’t want to get feedback on what therapists think of these two new features! Meanwhile we’ll continue to work on expanding and improving the Lyssn platform, and in creating efficient AI-tools that can help therapists and supervisors better help their clients.

If interested in future updates to Lyssn’s metrics and features, you can sign up for blog updates.