Retrieve-and-edit seq2seq modelling for review response generation

Supervisors: Tannon Kew/Martin Volk


Conditional text generation tasks, such as machine translation, summarisation and dialogue modelling are often built on a sequence-to-sequence framework. In the context of review-response generation (e.g. online customer reviews for restaurants, apps, etc.), target texts often follow a relatively standardised structure, involving a greeting, expressing thanks and/or apologising about something mentioned in the review, and a salutation. As a result, a suitable response to one review may be derived by editing a previously written response to a similar review.


The goal of this project is to apply the technique proposed by Hossain et al., (2020) and investigate retrieve-and-edit seq2seq methods for review response generation in the hospitality domain.


  • Programming knowledge in Python
  • Foundational knowledge in the field of machine learning/deep learning in the courses Introduction to Machine Translation and Machine Learning for NLP