Capstone Project

Emrick

What’s about

  • predicting next word given an incomplete sentence.

  • Online app with friendly user interface.

Algorithm

  • n-gram model

    • n-gram meaning n words as a single combination

    • words which are far away from each other unlikely to be related

    • In the app, 5-gram was used.

Algorithm (cont’d)

  • On inference

    • Try to fit last 4 words, predict most commont fifth word from the 5 gram data.

    • When unable to fit back off to fit 3 words, etc.

    • Only using 5-gram data, as lower gram just part of it. Therefore less space is occupied.

In Action