Improving Meeting Focus with Conversation Relevance

Meetings are a big part of work life, but let's be honest: a lot of them are not very useful. You sit in a room (or on a video call), and the clock ticks while people talk about everything under the sun, except what they should be discussing. This leads to frustration and a lot of wasted time. What if we could fix that? What if we could keep Conversations on track with clear goals?

That’s where the idea of Topic-Conversation Relevance (TCR) comes in. It’s about figuring out if the conversation matches what the meeting was supposed to be about. To help with this, a large dataset was created to study how closely conversations relate to set topics.

#What’s in the TCR Dataset?

So, what does this dataset contain? It has tons of meeting transcripts-specifically, about 1,500 unique meetings with around 22 million words. That's more than enough to fill a library! These meetings cover various topics and styles, which is essential because not all meetings are created equal.

The dataset includes over 15,000 meeting topics, gathered from both new meeting data and existing public sources. Additionally, there's a bunch of scripts that help generate synthetic meetings. Think of these as practice meetings created to make the dataset more diverse and representative.

#Why Do We Care About this Data?

With the rise of online meetings-surprisingly, in-person meetings have dropped from 63% in 2019 to just 33% in 2021-keeping discussions focused has become even more important. Especially with people often multitasking at home on their couch (don’t pretend you haven’t done it).

Having a meeting facilitator can help maintain focus, but we can also use technology to assist. Measuring how relevant a conversation is to its intended topic helps ensure that discussions don’t wander off into the weeds. For example, if a conversation strays far from the main topic, it signals that the discussion might need some guidance.

#How Does It Work?

The dataset allows for evaluating conversations against their intended topics of discussion. For example, if a meeting’s agenda is to discuss the launch of a product, but the conversation veers off into personal stories about weekend trips, you can bet the relevance score for that conversation would be low.

To develop a clearer understanding of what works and what doesn’t, benchmarks were created using advanced AI tools-a bit like a digital assistant that reads the transcripts and decides how on-topic each part of the conversation is.

#What’s the Plan for the Future?

The goal is clear: we want to gather data on more types of meetings in different fields. However, this can be challenging because many business meetings involve sensitive information. To overcome this, experts from various industries are being invited to create meeting agendas and conduct meetings based on those plans.

Also, expanding the dataset to include other languages is on the to-do list. Because let’s face it, meetings aren’t just held in English, and it would be a shame to leave out all the non-English speaking folks who also need help staying on topic.

Finally, adding audio data to the dataset can help improve the understanding of conversations. So, combining both audio and transcripts could enhance the ability to evaluate meeting effectiveness even further.

#Some Fun Numbers

Let’s take a quick dive into the numbers to see how this all plays out.

• The TCR dataset contains around 1,506 unique meetings.
• The total word count in the transcripts is about 22 million. Just imagine reading that. You’d need a lot of coffee!
• There are roughly 15,000 meeting topics included.

These numbers are not just for bragging rights; they provide a solid foundation for testing how well topics match conversations.

#We Can’t Do It Alone

If you think collecting all this data and making sense of it is easy, think again! It takes a whole team to create, analyze, and improve Datasets like this. People need to work together, share insights, and refine the process to get to a point where we can effectively measure meeting relevance.

And let’s not forget those who participate in the meetings and lend their voices to the data. Without their consent, we’d be lost. Luckily, all participants in the data collection process signed consent forms, ensuring that everyone is on board with contributing to this important work.

#So, What’s Next?

As we move forward, the research will focus on improving the dataset, improving performance on relevance tasks, and better understanding meeting dynamics. Building on the existing knowledge and technology, we can take meetings from drone sessions to productive places filled with useful discussions.

In conclusion, the TCR dataset is all about making meetings better. With clear topics, focused conversations, and smart use of technology, we can make sure our work meetings are more effective and less of a time sink. The data and insights gained from this work can help shape the future of how we meet and communicate in the workplace.

Now, wouldn’t it be great if someone invented a way to automate that pesky coffee-making task for those marathon meetings? Until then, let’s hope this dataset leads us to more productive discussions. After all, no one ever thought: “Wow, I really enjoyed that meeting where we talked about everything except what we were supposed to.”