Expert Pages

What is an expert page?

According to the Hilltop Paper, an expert page must be objective and diverse. How is this determined?

1. Expert pages must meet an "out-degree" threshold. What is an out-degree threshold? In the paper, there isn't a definitave answer for this question. But it seems to indicate a percentage of out going links going to non affiliated web pages. The example used in the paper is '5' or 5 percent?

The paper goes on to state that if a broad classification is known for each page, then the expert pages can additionally be distinguished by those that fall into the search term classification. That helps to seperate resource directories from random collections of links. Regular maintenance of the page is also mentioned as a means of helping to determine relevance.

2. Expert pages point to "non-affiliated" hosts It is actually easier to define what makes up affiliated hosts. Affiliated hosts:

a. Share the same first 3 octets of the IP address.
b. Share the rightmost non-generic token in the hostname. (ex: ibm.com & ibm.co.mx are considered affiliated)

What's done with expert pages?

Expert pages are subgrouped into an inverted index apart from the general population of web pages. These are indexed to map 'key words' that match user queries found on expert pages' 'key phrases'. Key phrases are text within the title, heading, and anchor text of the page.

a. The Title qualifies all the urls found in the document.
b. The Heading qualifies all the urls found until the next heading is found.
c. The Anchor Text qualifies only the url it is associated with.

The index of expert pages is then organized as a list of 'match positions' within expert pages. In other words, where does the key word appear in a key phrase of an expert page.

There are two sets of scores given for placement of key words in key phrases.

A level score is given for the key word depending on which key phrase it is found in. For example, in the paper, if the key word was found in the Title, a score of 16 was given. If the key word was found in the heading, a score of 6 is given. And if the key word is found in the anchor text, a score of 1 is given.

A fullness score is subtracted based on how many additional terms (not included in the search query) that appear in the key phrase. In other words, how diluted is the key words within the key phrase.


How are expert pages used?

The "top" expert pages are reduced to a predetermined number. The example used for the paper is 200. All the "target" pages they point to are placed into a query subset, if and only if two or more experts point to it providing that the experts are not affiliated to each other or to the target page.

A target score (authority score) is computed based on the number and relevance of the expert pages pointing to it and the relevance of the phrases qualifying the links.



Conclusion: From this more detailed analysis of the Hilltop paper, I would say that my initial hypothisis that an expert page being the next best relevant page to authority pages doesn't have any support within the paper.

I will continue my experiment to see if there is a connection, but I'm beginning to have my doubts.