Florida #2 ?

[Johan007]

Whats TSPR?

[Rob]

Originally Posted by Johan007

Whats TSPR?

Topic-Sensitive PageRank

http://www.seoresearchlabs.com/seo-r...gle-report.pdf

[compar]

LocalRank - aka LocalScore - is a completely different type of algo

Maybe we are talking about different Local Ranks. But I think Local Rank and Hilltop are very similar. Read the Patent application for
Local Rank

One of the authors of this patent is Krishna Bharat. He is also one of the two authors of the Hilltop paper. In my view Local Rank and Hilltop both do almost exactly the same thing. And Ive even had confirmation of this from other people.

BTW here is the link to Hilltop if anybody wants to read about both of these technologies or methodologies.

[I, Brian]

My reading of LocalRank has been that it's principle purpose is that of a filter, rather than a core means of selecting relevant documents in the first place - which is what Hilltop is.

[compar]

Have you read the patent paper?

Here is the abstract from the Local Rank Patent:

Ranking search results by reranking the results based on local inter-connectivity

Abstract

A search engine for searching a corpus improves the relevancy of the results by refining a standard relevancy score based on the interconnectivity of the initially returned set of documents. The search engine obtains an initial set of relevant documents by matching a user's search terms to an index of a corpus. A re-ranking component in the search engine then refines the initially returned document rankings so that documents that are frequently cited in the initial set of relevant documents are preferred over documents that are less frequently cited within the initial set.

Here is part of the abstract from Hilltop:

Abstract:
In this paper, we propose a novel ranking scheme for broad queries that places the most authoritative pages on the query topic at the top of the ranking. Our algorithm operates on a special index of "expert documents." These are a subset of the pages on the WWW identified as directories of links to non-affiliated sources on specific topics. Results are ranked based on the match between the query and relevant descriptive text for hyperlinks on expert pages pointing to a given result page.

Show me the major difference in the results of these two statements. They are virtually identical. What Local Rank refers to as a "index of a corpus", Hilltop calls "index of "expert documents."". They are both saying the same thing and attempting to modify the SERP based on this second level ranking.

While Local Rank mentions inter-connectivity as a criteria in the abstract if you read Hilltop it establishes exactly the same set of inter-connectivity criteria for compiling it's "index of "expert documents.""

Local Rank is Google's patented version of Hilltop. There is no doubt in my mind.

[I, Brian]

The LocalRank algo is an algo that seems to "refine" a set according to standard practice - it searches, then closes the set according to additional criteria. In my reading, inter-connectivity with regards to LocalRank means nothing particular other than link pop.

The point about Hilltop is that it requires a pre-defined set of terms in the first place. The update this last weekend makes perfect sense if Google has extended the initial set of "Florida" terms into a wider set - probably based on search frequency values.

I'm not quite sure how both algos can be perceived to have the same net effect.

How in your mind would LocalRank in particular have created the current results, and especially rate .edu and certain directories?

[compar]

I don't know why I get in these long debates with you. The English language doesn't change that much as it crosses the pond. But let me try a little longer.

LR patent says:

What is claimed:

1. A method of identifying documents relevant to a search query, comprising:
obtaining an initial set of relevant documents from a corpus;
ranking the initial set of documents to obtain a relevance score for each document in the initial set of documents;

Hilltop Says:

In response to a query, we first compute a list of the most relevant experts on the query topic. Then, we identify relevant links within the selected set of experts, and follow them to identify target web pages. The targets are then ranked according to the number and relevance of non-affiliated experts that point to them.

Again in my humble Canadian version of the English language these two pieces of text are saying the same thing.

[I, Brian]

Number 1 for LocalRank reads to me exactly like any basic ranking algo - it's looking for pages associated with the terms, and then filters that list.

Hilltop, on the other hand, already has a predefined set to work from, from which it then refines the search from.

By analogy, if LocalRank is faced with a search for fruit, it will obtain a set of documents simply on the basic that they refer to "fruit", and refines that list. Hilltop, on the other hand, has already decided that "apples" and "oranges" are much more relevant "fruit", and therefore these types of "fruit" get distinct prominence in returns for "fruit".

Of course, I could be wrong - these are merely my current interpretations. My opinions are really quite flexible.

[Emancipator]

Where did you get the data that helped you determine this is how local rank and pagerank work Brian? I agree with Bob and he has shown his sources, can you share yours, or examples?

Interesting discussion, Cheers.

[compar]

Local Rank Patent

In one aspect, the present invention is directed to a method of identifying documents relevant to a search query. The method includes generating an initial set of relevant documents from a corpus based on a matching of terms in a search query to the corpus.

Hilltop Paper:

In a pre-processing step, a subset of the pages crawled by a search engine are identified as experts..........
Given an input query, a lookup is done on the expert-index to find and rank matching expert pages. This phase computes the best expert pages on the query topic as well as associated match information.

I have added the bold.

Both systems are aimed at exactly the same end result. To apply a second rank to a query based on a set of highly relevant or "expert pages".

[I, Brian]

Originally Posted by Emancipator

Where did you get the data that helped you determine this is how local rank and pagerank work Brian?

By reading the papers online, of course.

There are actually many more:

http://www.business-talk.co.uk/showthread.php?t=68

What is at odds here is not the information in the papers, as much as the interpretation of the information in the papers.

[Emancipator]

lol. I have no idea what I was thinking when I put "local rank and page rank' I meant "hilltop" .

I am not trying to put you on the spot, I just like to see reference materials. so I can interpret it my own way as well

You are 100% right, everything is about interpretation until somebody proves us all wrong..

[I, Brian]

Originally Posted by compar

Local Rank Patent

Hilltop Paper:

I have added the bold.

Both systems are aimed at exactly the same end result. To apply a second rank to a query based on a set of highly relevant or "expert pages".

compar, I do not at all interpret the LocalRank paper's initial set of relevant documents as meaning experts - it's just a set of returns for the search term.

So if you do a search within LocalRank for "widgets", then it will merely return an initial set of returns for "widgets" - like any algo would.

The difference with LocalRank being that it then ranks the pages in that list in a manner particular to the linking relationships between different pages.

Hilltop, on the other hand, already has a pre-selected set of "experts" to decide which results it will return in the first place - for "widgets" Hilltop will only rank pages if they already have mulitple backlinks from the pre-selected "expert" pages.

And that's a particular distinction. LocalRank doesn't do that - it isn't picky about what it starts with. Hilltop is.

If you'd like a little more comment, Phil makes a note of LocalRank, and why he doesn't think it can apply to the Florida results, here:

http://www.webworkshop.net/florida-update.html#theories

[I, Brian]

Originally Posted by Emancipator

I am not trying to put you on the spot, I just like to see reference materials. so I can interpret it my own way as well

You are 100% right, everything is about interpretation until somebody proves us all wrong..

LOL! As I get to wear a "moderator" badge for the SEO forums, I have to expect to be put on the spot - and answer - over SEO issues.

The problem is that, without access to inside knowledge from Google, SEO in itself has to be the practice of "considered speculation".

And within that "considered speculation" there is room for a variety of opinions that may not all be in perfect agreement, but can still remain valid none-the-less.

SEO is a very dynamic field, and speculation coupled with observation can make for some very important - and challenging - ideas. In the wake of shocks such as Florida, that is extremely necessary.

Our appraisal of the Florida situation, IMO, increases with every serious idea. And compar made a very interesting connection elsewhere between Google's stemming and use of Applied Semantics technology that I hadn't realised.

I may personally disagree with compar's interpretation of LocalRank - but as I don't own the patent of "truth" I can merely try and justify why I personally disagree.

But every serious mooted possibility deserves every serious consideration. There is still a lot to learn here.

[Emancipator]

As a moderator on another SEO forum I never take anything at face value. If I dont see examples, and not just theories I let it go in one ear, and out the other SEO is my hobby, webdesign my business... but I take my hobbies seriously.

Play hard or go home! Thats my motto.

[compar]

compar, I do not at all interpret the LocalRank paper's initial set of relevant documents as meaning experts - it's just a set of returns for the search term.

But look at the Patent and the Hilltop paper and they both use essentially the same methods to get their comparison set of pages. Hilltop doesn't do anything wonderful to determine Expert pages. They are only "Expert" to the degree that they are relevant and other people link to them.

Local rank calls it a "corpus" but they then go on to describe how this 'corpus' is determined and it is determine in the same way that Hilltop is suggesting. Links from non affiliate pages and from different IP blocks.

You seem to think that there is something magic about "expert". There isn't. A page just gets to be expert by virtue of its content and the people who are linking to it. That's the only possible way that the search engine could judge and determine the subset that they want to use for a second ranking pass.

From LR Patent

2. The method of claim 1, wherein calculating the local score value for a particular one of the relevant documents further includes: forming a sub-set of documents from the initial set of documents as the sub-set of documents that contain a hyperlink to the particular one of the relevant document, and removing documents from the sub-set that are from the same host or from an affiliated host as the particular one of the relevant documents.

How is that any different from Hilltop method of determining 'expert' pages?

[I, Brian]

Originally Posted by compar

You seem to think that there is something magic about "expert".

That's very much the point of it - there is supposed to be a very specific set of criteria for determining expert pages:

From the Hilltop paper:

Originally Posted by Hilltop

Experts in our definition are directories of links pointing to many non-affiliated sites. This is an indication that these pages were created for the purpose of directing users to resources, and hence we regard their opinion as valuable. Additionally, in computing the level of relevance, we require a match between the query and the text on the expert page which qualifies the hyperlink being considered. This ensures that hyperlinks being considered are on the query topic. For further accuracy, we require that at least 2 non-affiliated experts point to the returned page with relevant qualifying text describing their linkage.

The high rankings of directories, .edu and .gov pages were orginally taken as evidence of a Hilltop-like expert system. I note with interest, though, that there's no specific mention of .edu sites in themselves on Hilltop. I really should look to see where that idea came from.

There are features of both LocalRank and Hilltop that overlap - but everything depends upon the core selection of expert documents, and how that differs.

LocalRank seems to only groups pages together according to topic linkage, and then modifies that list according to affiliations - but Hilltop goes further and groups pages according to not just topic linkage, but according to how those pages are linked to from a subset of "expert pages".

To this I might add that although Hilltop - and TSPR - have been raised as particular contenders for the Florida system, I think it's going to be very important for myself to read the full list of Google papers, that I linked to earlier. I may be some time.

[glengara]

Don't know if it has to be an either/or situation, seems possible Localrank could be used as secondary step to Hilltop, using the broadmatch "corpus" of HT to further narrow down the relevance of pages.

[I, Brian]

Originally Posted by glengara

Don't know if it has to be an either/or situation, seems possible Localrank could be used as secondary step to Hilltop, using the broadmatch "corpus" of HT to further narrow down the relevance of pages.

I quite argee - I've posted about both having been applied in Florida before.

[Emancipator]

Brian I am not following you. Why dont you post in point form what you are saying happens starting with me searching on goole for "emancipator" for example.

- search for emancipator
- google then...
- then.....
- then....

I realize you are only assuming, but I am trying to see what it is you are saying and I am afraid I am missing the boat entirely on what you are saying.