Made a similar news recommendation system project, using a mixture of content-based and collaborative filtering recommendation algorithm, just tell me about how I do, a lot of journal also have specific algorithm implementation,
Data is crawl the sohu news, without considering the classification at the beginning, later found it when you are going to use it before too much is not marked data classification, and a lot of news on the original site classification is rather vague, many hot news, news and other web sites, there is no way to system of classification,
See a lot of posts, and discovered the sogou laboratory in 2012 to sort out the sohu news data sets, there is a url, can treat as category information, training a classifier, the final classification of F1 value at around 0.87, 11 classes, also almost is the sohu news categories, (whole training the classifier for four times, every time is a little small error, ah)
Use of tf - idf values and cosine similarity, classified information at the same time, also there is the news keywords,