PathMotion Product Update
The new, more accurate content categorisation model
After months of work, we’re excited to share with you our first ever machine learning model that we’ve built, taught and integrated into the platform. The model has greatly improved the categorisation of content on the platform…
Our aim was to better organise the existing content on the platform, ensuring that discussions are correctly allocated to topic categories. In turn, this would help attain our wider goal: to help candidates discover discussions on topics that interest them more quickly and easily. We are now able to show impactful content that is relevant to them, improving their candidate experience.
Those of you that have been with us a little longer might remember that in the past there were only four categories questions could fall into. These four categories were Life at the Company, Career Choices, Application Advice and General, and they would be selected by the candidate when asking the question.
We knew we needed to improve this process by creating more labels and increasing their accuracy: the platform hosts such a huge wealth of content that these labels were far too generic to cover them all. Due to the nature of the user selection, the consistency was also sometimes compromised.
The product team started to investigate and experiment. The Great Topics were devised: ten categories into which discussions were automatically allocated based on keywords. This automatic categorisation with multiple topics was a large improvement on the previous model. However, we knew there was more to be done: there was still a degree of accuracy missing when only picking out certain keywords and restricting a discussion to one category.
To further improve the accuracy of these discussion categories, we’ve moved towards machine learning.
We created a machine learning model in both English and French, trained it up, integrated it into the product and worked to improve its quality. This model takes a new discussion, runs it through its system, and, based on its training, allocates the discussion with a category from an extensive default list. We are now moving away from a user-allocated content categorisation process to an intelligent machine learning process.
Examples of these new categories include Role, Career tips, Everyday life and Eligibility advice:
The breakdown of the categories will appear as it used to, on the Dashboard of the back end, so that you can see which areas your candidates are most interested in.
This has been a process of constant improvement. Initially, the quality of the machine learning model was about 60%. The version launched this week now functions at over 70% accuracy rate. We will continue to improve the functionality of the model through a user feedback system:
As an administrator or insider of the platform, you are able to flag a discussion on the front end of the platform if you feel that it has been wrongly categorised. Once flagged, we will do the rest. This way, our model will be continuously evolving and improving, based on your feedback.
What does this update mean for your candidates?
- Candidates will find it easier to navigate around the site and quickly find the answer to their questions as headings are more specific.
- Candidates will see more discussions relevant to their interests. Consequently, the number of views on discussions will increase and the number of duplicate questions should decrease.
- In turn, this will mean that your candidates will be better informed on all aspects of your brand.
What does this update mean for you, as a client?
- We are able to give you a better understanding of the content available on your platform and on trending topics amongst your candidates. You can then use this information for campaigns you are running elsewhere in the business, for example on social media.
- You will gain a greater insight into what’s on your candidates’ minds and see what matters most to them.
- We have also been able to analyse the trends that this model has shown us across all of our clients, spanning multiple sectors, to provide you with these insights. For more information on this, check out our data-driven series here.
Whilst we have previously used some aspects of AI and machine learning in the chatbot and FAQ functions, we are really excited to reveal this platform update with you as it is the first machine learning model that we have built and taught ourselves and implemented on the platform.