Hey team,

I use the latest enwiki and train a model with the following cmd:

wikipedia2vec train enwiki-20190701-pages-articles.xml.bz2 enwiki-20190701-300d \
--dim-size=300 \
--no-lowercase \
--min-word-count=30 \
--min-entity-count=10

To my understanding, by default, the category flag is False. Therefore, it should filter all wiki category pages.

However, by examine the titles in the wikipedia2vec model, I found the following titles:

:Category:American actors
:Category:American architects
:Category:American film actors
...

categories are not fully filtered.

We could change the code here by adding more filters and check if the title starts with :Category:.

In a similar fashion, you might also want to filter non-entity with title that starts with:

:wikt: 
:Category: 
:Image: 
:category: 
:commons: 
:Template: 
:File:
...

There are quite some suspicious non-entity title that starts with :