Summary

Unsupervised learning is getting more and more popular in the NLP community, since large-scale un-annotated corpora are increasingly available at almost no cost and unsupervised learning approaches provide the capability to directly utilize such un-annotated corpora.

However, the results may not be optimal or may even be unsatisfactory if the learning procedure is not conducted properly.

In this tutorial, we will first identify some frequently unnoticed pitfalls that might prevent the unsupervised learning from getting good performance. The suggested strategies for avoiding such pitfalls are then proposed.

Duration

3 hours

Speakers

Jing-Shin Chang (National Chi-Nan University, Taipei)
Keh-Yih Su (Behavior Design Corporation, Taipei)