Link spam detection using smooth classification function
View Patent ↗A spam detection system is disclosed. The system includes a classifier training component that receives a first set of training pages labeled as normal pages and a second set of training pages labeled as spam pages. The training component trains a web page classifier based on both the first set of training pages and the second set of training pages. A spam detector then receives unlabeled web pages uses the web page classifier to classify the unlabeled web pages as spam pages or normal pages.
1. A computer-implemented method of detecting spam in a collection of web pages, the method comprising:
receiving a first subset of the collection of web pages marked by a trusted entity as spam;
receiving a second subset of the collection of web pages marked by the trusted entity as content;
identifying a classifier function by encouraging, with a cost function penalty, the web pages in the first subset to take a value close to a spam value and the web pages in the second subset to take a value close to a content value, and penalizing, with the cost function, differences in function values over nodes in the collection of web pages;
utilizing a computer processor to classify a set of unlabeled web pages as spam or content based on the classifier function values assigned; and
outputting an indication of the classification of the unlabeled web pages as spam or content for use by a user.
2. The method claim 1 , wherein identifying the classifier function includes performing a random walk over the web pages by:
choosing a starting point in a directed graph having nodes representing the web pages in the collection and edges representing links among the nodes; and
repeatedly moving to a new node in the directed graph by selecting uniformly, at random, whether to follow a link from a current node to another node in the directed graph or to randomly jump to another node in the directed graph.
3. The method of claim 2 , wherein assigning the classifier function comprises:
computing stationary probabilities and transition probabilities based on how often a link and node are selected.
4. The method of claim 3 , wherein selecting whether to follow a link is done according to a predefined random walk definition.
5. The method of claim 4 , wherein the predefined random walk definition comprises:
one of following outlinks from the current node, inlinks to the current node, and inlinks or outlinks regardless of direction relative to the current node.
6. The method of claim 2 and further comprising:
breaking the directed graph into strongly connected components, such that each node in any given strongly connected component has at least one path, through links in the given strongly connected component, to every other node in the given strongly connected component.
7. A computer-implemented method of detecting spam in a collection of web pages, the method comprising:
receiving a first subset of the collection of web pages identified as spam;
receiving a second subset of the collection of web pages identified as non-spam content;
identifying a classifier function based on values assigned to web pages in the first and second sebsets;
utilizing a computer processor to classify a set of non-classified web pages as spam or non-spam content based on the classifier function; and
outputting an indication of the classification of the non-classified web pages as spam or content for use by a user.
8. The method of claim 7 , wherein identifying the classifier function comprises identifying the classifier function based on application of a cost function penalty to pages in the collection web pages.
9. The method of claim 7 , wherein identifying the classifier function comprises identifying the classifier function based on application of a function penalty to the web pages in the first subset such that the web pages in the first subset take on a value close to a spam value.
10. The method of claim 9 , wherein identifying the classifier function comprises identifying the classifier function based on application of a function penalty to the web pages in the second subset such that the web pages in the first subset take on a value close to a non-spam content value.
11. The method of claim 7 , wherein identifying the classifier function comprises identifying the classifier function based on application of a function penalty to the web pages in the second subset such that the web pages in the first subset take on a value close to a non-spam content value.
12. The method of claim 7 , wherein identifying the classifier comprises identifying based on a random walk over the web pages in said collection.
13. The method of claim 12 , wherein the random walk is accomplished by a series of steps, comprising:
choosing a starting point in a directed graph having nodes representing the web pages in the collection and edges representing links among the nodes; and
repeatedly moving to a new node in the directed graph by selecting whether to follow a link from a current node to another node in the directed graph or to jump to another node in the directed graph.
14. The method of claim 7 , wherein assigning the classifier function comprises computing stationary probabilities and transition probabilities.