Iwww

This paper may be of particular interest to DP Co-op Advertising Network and Link Vault users - Pagerank Increase under Different Collusion Topologies.

The paper is one of a number being presented in a workshop (WF06) titled 'Adversarial Information Retrieval on the Web' at the 14th International World Wide Web Conference in Chiba, Japan on May 10-14.

Unfortunately I can't make the IWWW conference this year but that little workshop WF06 is shaping up to be an interesting day with another paper (not published yet) by A. Benczur titled 'SpamRank - Fully Automatic Link Spam Detection'.

'SpamRank' - I love it already

- Michael

 

Thanks for posting that Michael, one has to be very bright to understand it, in the mean time can you explain it to us average folks

 

Fryman, if Michael does not come back to educate us get a hold of the guy in the photo

 

Never mind that guy -- he's deceased... he is no more... he is an ex-guy... he has passed on... he has cast off this mortal coil...

Get the woman in this picture instead...

Afterthought: I wonder if Albert Einstein felt like the cow in the second picture lot of the time...

 

Maybe that is what Michael & Shawn feel like Minstrel, they both sound like potential clients for you

 

Perhaps this DMOZ editor can shed some light...

 

I think that is the one hanging with TOOT over on his forum

 

Sure. It's not actually as complicated as it looks.

The authors are discussing "Collusion" and a possible method of detection. Collusion is taken to mean in this case any "deceptive linking practices to improve reputation algorithms". (A reputation algorithm of some kind will be present in every search engine, for example Google's is based on Pagerank and the snippet that fryman quoted above has its origins in an earlier paper by Page and Brin).

The method of detection they examine is measuring the ratio of the total Pagerank of a group of pages and the Pagerank they receive externally. The idea being that colluding groups will have a disproportionately high ratio compared with non-colluding groups (a fairly logical hypothesis!).

They proceed to look at some collusion strategies using this method of detection. They look at collusion by limiting the out-bound links of the group, collusion by having all pages in the group link to a single page (star topology) and collusion by each page linking to another in various sequences within a closed loop (ring topology).

The authors demonstrate that the method although quite good is not perfect because firstly there are some legitimate groups with disproportionately high ratios and secondly that not all collusion strategies result in a detectable group.

They then go on to suggest further possibilities for research in this area.

I suggested above that this paper might be of interest to DP Co-op Advertising Network and Link Vault users. That is because Co-op and Link Vault are essentially collusion strategies. Of course in these two cases it is not necessary to have a detection algorithm because they are well publicised and easily identified. However there will be tens of thousands of 'private' collusion strategies that cannot be so easily identified.

The underlying conclusion I think we can draw is that if a topic in search engine research like this one is actively pursued in public then you can bet that it is receiving more than a cursory glance in the research departments of the three main search engines.

I hope this helps.

- Michael

 

right, but an intergration into the algorithms could have an affect.

 

I am sure it would but of course there is no way of predicting the scale of the effect, which could be from very small to very large.

- Michael