[IR] Link Analysis

网络信息的特点在于:

Query: "IBM" --> "Computer" --> documentIDs.


 

In degree i 正比于 1/iα ,  例如: α = 2.1

即:i越大,量越少。

 

Query processing

§  First retrieve all pages meeting the text query (say venture capital).
§  Order these by their link popularity (either variant on the previous slide).
§  More nuanced – use link counts as a measure of static goodness (Lecture 7), combined with text match score.

 

link多,但不一定意味着都是重要的推荐(link).

可以让PageRank Scoring通过"Flow" Model来获得,即访问量。

 

 

  • 方法一: 

 

解方程得:

y+a+m = 1
y = 2/5, a = 2/5, m = 1/5

 

Gaussian elimina*on method works for small examples, but we need a better method for large graphs.

 

  • 方法二:

 利用Markov chainsx= xPi

初始值,可以假设是uniform distribution,最后也将达到稳定状态。

 


 

若干可能的问题:

  • Spider traps 

 

Sol: Random teleports - 随机瞬间移动,防止掉入死胡同

  可见,1变为了7/11,但貌似并不是效果特别满意。

 

  • Dead Ends

  

 

§  Follow random teleport links with probability 1.0 from dead-ends
§  Adjust matrix accordingly. How?

 

Sol:

0.8 * [0,0,0] 这里是触发条件。

发现,0.2*[1/3, 1/3, 1/3]这么下去,趋势必然为0。

那就,0.2这个随机处理去掉好了。[1/15,1/15,1/15] --> [1/3, 1/3, 1/3]

 

posted @ 2016-11-08 09:05  郝壹贰叁  阅读(823)  评论(0编辑  收藏  举报