Detail of the student project

List
Topic:Sentence pair similarity.
Supervisor:Jan Sedivy
Announce as:DP,BP,PTO
Description:People produce a lot of text every day on the Internet. We need to teach machines to understand the natural language. The computers will help us search or ask questions about the content. One way of doing it is to tell the machine meaning of a sentence and then train a model recognizing similar sentences. Respectively the task is to create a model measuring the semantic similarity of sentences. We can such model not only for recognizing similar phrases but also for finding the next utterance for dialogue or finding the correct answer to the given question.
Instruction:Instructions
Select a domain
From the selected domain generate pairs of similar sentence, for example question answer or event and response in a dialog
Review the latest LSTM, GRU neural networks architectures
Train a first model
Optimize the architecture and meta parameters to achieve the lowest error rate
Bibliography:https://arxiv.org/abs/1603.06127
Date:22.02.2017
Max.number of students:0
Responsible person: Petr Pošík