If there is a news APP, there are some people use, which will produce some user action logs. I want to build a system based on these user logs portrait, portraits and news users mainly portraits, portraits of these good statistics, good foundation for personalized recommendation later. Some common features news portraits and portraits of the user as follows:
一.news profile
Here can be used json hierarchical representation of a news, real-time news crawl parse extract. Extracting information from the tag should be coarse to fine, for example from: a classification -> two classifications -> specific content key words.
二.user profile
News and portraits, as can using json to store user portrait format. Users can collect basic information from registration; recording user behavior can be optimized in preparation for later; pre-label relies on its users to do the recommendation, there will be the user's preferences into long label in short, each preference is divided into three layers system tag (tag information with the same system), each corresponds to a respective label and their weights.