140616 - Xi Ma

Last Week

1.Extract training set from weibo and baiduzhidao using cross-entropy difference

2.Using the extracted text and seed corpus to train a new language model

3.Testing ppl and wer of the new language model using different test set

This Week

1.Mix the our model with jietong's model using good weight

2.Test the ppl and wer using the new test set.