xlvector – Recommender System

当用户浏览一个网页的时候，他将面对很多信息，有些他选择点击观看了，有些则被他忽视了。传统的日志系统只会记录他点击了什么，而不会记录他忽视了什么。但忽视也是一种信息，而对这种信息的利用无疑可以改善推荐的性能。

当然有人会争辩，忽视的原因是多样的，不一定代表用户不感兴趣。比如用户可能真的就没看见。确实这是个问题，不过通过日志系统，我们可以拿到整个网站的click graph，那么如果从页面A到B有很多人都点击了，而user u没有点击，也就是说在页面A上，u忽视了B页面的链接，那这就说明u很可能是不感兴趣，而不是没有见到B。

当然，上面提到的方法还很粗糙，如果深入的做一下，还是可以获得不少启发。

When a user view a web page, she will see many links, and choose some to view while ignore others. Traditional log system will only record which links she clicks and will not record which links she ignores. However, ignore behavior is also a type of information and it can be used to boost recommendation quality.

Someone may argue that there are many reason of ignore. For example, the use may not see the link, so she ignore the link do not mean she dislike the link. However, there is a simple method to solve this problem. For example, if manys user visit page B when they visit page A, this means link of page B is easy to find in page A. In this situation, if a user ignore link of page B in page A, she may not interest in page B.