federkasten

Awesome!

Updated.

* fix some stylesheet issues (support Chrome on Android)

Updated.

1. fix some bugs (e.g. "item.time_str")

2. support responsive design (currently optimized with iPhone)

Thanks.

You're right. That's nice idea! I will add that feature shortly.

It's nice idea. I will try it.

Yes. You are right.

I trained classifiers with hand annotations (about 1000 contents or so)

Thanks. I will support responsive web design. It maybe solve it.

Thanks. hnstream is really cool.

You're right. I have limited frequency of crawling due to HN's access thresholds.

hnstream will help me. I will try it.

Thanks. I will fix it shortly.

Thanks.

It crawls HN topics every one hour. So news in our application would delay up to one hour.

Yes. I will support responsive design in the near future.

I am planning to open source it in several months. (Our codes have not been well-commented and well-structured yet...

Our implementation and algorithm detail is followings.

Its categorizing process is written in Python.

Using nltk, it makes corpus with TFIDF model from HN topics and comments. And it generates classifiers from this corpus with SVM algorithm using scipy and numpy.

FYI, its web interface is written in Clojure and ClojureScript.

Using natural language processing and supervised learning, it automatically tagged news with 9 categories.