Global-Scale NLP at SEEK : Melbourne NLP Meetup

In this blog post, Saumya Pandey a Data Scientist in Artificial Intelligence & Platform Services (AIPS) at SEEK, Melbourne, writes about the highlights from the recent Melbourne NLP meetup hosted by SEEK.

Natural language processing (NLP) has traditionally involved enhancing the automatic semantic and syntactic understanding of human languages. In recent years, the emergence of large language models has revolutionised NLP (Timeline of new transformers models shown) by achieving state-of-the-art performance on many language tasks such as question-answering and sentiment analysis. It has also spurred new applications across several industries including the employment marketplace domain where SEEK is the leading employment marketplace in ANZ, South-East Asia and Latin America.

NLP Meetups play an integral role in providing the forum for professionals, students, enthusiasts and academics to come together to understand the industry trends, recent advances, and bring their knowledge to like-minded people sharing the same passion for NLP.

Timeline of new Transformers models | Source : Hugging Face

In August 2022, SEEK hosted the Melbourne Natural Language Processing (NLP) Meetup in the new Cremorne HQ. The objective of the meetup was to showcase the scale of NLP at SEEK. This would enable SEEK to attract world-class talent from industry and academia as well as to establish connections for future collaboration. The meetup had several presentations from the NLP practitioners at SEEK highlighting some of the exciting work in the employment domain. This work is aimed at harnessing the potential of cutting-edge research in NLP.

Eugene Svistunov, Director of Machine Learning, discussing NLP at SEEK

The Meetup opened with remarks from Eugene Svistunov, Director of Machine Learning, who introduced Artificial Intelligence Platform Services (AIPS) that is at the core of the employment marketplace. AIPS is organised in squads such as Job discovery, Candidate Quality, Recommendations and Pricing.

• Silvia Necsulescu (Marketplace Knowledge) gave the talk on Domain adaptation using fine-tuned multilingual models. She first described the challenges in understanding and creating meaningful structured data (for example, Job Title normalisation). Lexical and embeddings-based approaches for normalisation along with development of multilingual ​embeddings models were also discussed.
• Kees Varekamp (Candidate Quality) discussed the use of Pre-Trained language models on Databricks for Machine Translation at scale. He also described the process of building training pipelines to support multilingual models to serve candidates across Asia-Pacific.
• Trevor Yann (Search) explained how the squad’s purpose of helping job seekers discover content based on their query, drives experimentation & development within the squad. He highlighted different aspects involved in running a Search Engine on an employment marketplace such as SEEK. This involves Retrieval and Ranking of relevant jobs matching a job seeker’s information needs. Similarly, generative language models can be used for alternative query suggestions to the job seekers.
• Aditya Joshi (Recommendations) discussed how the NLP techniques provide a mechanism to power recommendations. This may be done by obtaining structured information from ubiquitous textual data on SEEK and the integration of textual semantics. He also highlighted the use-cases of a representation learning pipeline developed within his squad.

Representation Learning Pipeline | Source: SEEK

The NLP meetup was conducted as an in-person meeting at Cremorne along with online zoom event. It saw a huge turnout (both online and offline). It was fantastic to welcome the vibrant Melbourne NLP community to SEEK HQ and share food, drinks and thoughts about the present and future of NLP. There was a buzz in the room and we were excited to see that participants enjoyed the event.

It was great to hear about the work that the different squads at the AI Platform Services (AIPS) are doing: machine translation and transfer learning for multilingual semantic similarity, search, recommendations

- Damiano (Senior Lecturer, RMIT University)

This is the first hosting of the Melbourne NLP Meetup in the new SEEK HQ and represents SEEK’s ongoing plan to support and engage with the Australian Data Science and Artificial Intelligence community. Watch this space!

Interested in joining us for reshaping the employment marketplace with AI/NLP? Explore our open roles and discover your passion for AI/ML

Acknowledgements:

Organising an event like the NLP Meetup requires a lot of help! The space, technology, and food couldn’t have been organised without the wonderful support of Stephanie De Luca and the rest of SEEK’s Events and Catering teams. The idea and initiative for this event is entirely due to the Melbourne Natural Language Processing (NLP) Meetup organisers, Andy MacKinlay, Meladel Mistica, and Bahar Salehi.

We thank AIPS leaders Grant Wright and Eugene Svistunov for their support of the event. And huge thanks to Aditya Joshi, Pieter Barnard, Terrence Szymanski, Ruey-Cheng Chan, the speakers Silvia Necsulescu, Trevor Yann, Kees Varekamp, the entire AIPS NLP Reading Group, and many more for helping to prepare, coordinate, and host the event on the night!