Project Summary

Wordnerds proposes to build a new, agile neural network which has two unique features: it is specifically trained to minimise bias in the UK, and it is trained on passenger data (rail and aviation).

Project Achievements

Activities Reduction of gender, location, class and other bias on the Wordnerds platform through: • A sentiment model to better handle short content • Removal/neutralisation of words that can attract bias before sentiment processing, e.g. pronouns, names and entities These developments are now in full deployment live on the platform and in use by customers. Also the creation of a public transport theme bank offering operational, customer journey and segmentation methodologies to customers.

Conclusions

This project is allowing public transport providers to hear the voices of all their passengers with unprecedented clarity. Providers are able to quickly understand the issues facing passengers and take action to improve passenger experience, confident that they are making those decisions on data that contains minimal bias: • Bias reduced to around a quarter of original levels, from avg. 3.45% to avg. 0.93%. • Average platform set-up time for new transport customers reduced by 70%, from 35 to 10.5 hrs.

Next Steps

We intend to fully capitalise on the commercial opportunities provided by this project, by expanding further in the UK public transport market – starting at World Passenger Festival. We will continue to develop our methodologies work in adjacent sectors, and hope to explore other European languages next year. Our dedicated work on bias has already contributed to TOC contract renewals and led to a report commissioned by the Connected Places Catapult.