.-;;tLj
`
`I
`Attorney Docket No: S96-213/PROV
`
`IN THE UNITED STATES PATENT AND TRADEMARK OFFICE
`
`!$;~~\
`.
`!# ,-.. ,,
`.. .
`~.~. 1BS7
`: ·; .. ~ ~:
`t
`..zr:.J
`..... ._ .... ;)/
`~:::.-</
`li
`t~·~~~~/
`~Q!;,:Aj>plication No.:
`Filed:
`Title:
`-Applieant(s):
`Examiner:
`Art Unit:
`
`60/035,205
`10Jan 97
`hnproved Text Searching in Hypertext Systems
`- Lawrence Page- -
`mot yet assigned
`not yet assigned
`
`TRANSMITTAL OF MISSING PARTS
`
`~ TilE COMMISSIONER OF PATENTS AND TRADEMARKS
`Washington, DC 20231
`
`Sir:
`
`I HEREBY CERTIFY THAT TinS CORRESPONDENCE IS BEING DEPOSITED WITH THE UNmiD STA1ES POSTAL
`SERVICE AS FIRST CLASS MAll.. ON THE DATE OF DEPOSIT INDICATED BELOW AND IS ADDREsSED TO:
`ASSISTANT COMMISSIOW ~R PATENTS, WASHINGfON, DC 20231.
`DATEOFDEPOSIT $ 2.~ 'rt
`SIGNED: ~lj)ut~
`THOMAS J, McFARLANE REG. No. 39,299
`
`DATE ~{-z,~ l qi-
`
`'-f=i!:
`I ' F'
`
`~-
`
`Transmitted herewith in the Provisional patent application identified above are:
`
`[X]
`Small Entity Declaration; and
`[X] A check in the amount of $25 for the surcharge
`
`· Respectfully submitted,
`
`~(~
`Thomas J. McFarlane
`Reg. No. 39,299
`426 Lowell Avenue
`Palo Alto, CA 94301
`(415) 321-6630
`
`3(>0 SH 04·/23197 60(•35205
`25~00 CK
`1 227
`
`EXHIBIT 2077
`Facebook, Inc. et al.
`v.
`Software Rights Archive, LLC
`CASE IPR2013-00480
`
`

`

`20230 Appendix
`71632 U.S. PTO
`
`bU/UJ;JuU~
`
`PTO/SB/16(6-95)
`
`\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\
`01/10/91
`
`PROVJ:SJ:ONAL APPLJ:CATJ:ON
`
`COVER SHEET
`
`This is a request for filing a PROVISIONAL APPLICATION under 37 CFR 1 53(b) (2)
`
`inside this box -->
`
`!Docket Number jS96-213 ~Type a plus sign (+) j+
`
`LAST NAME
`
`INVENTOR(s)/APPLICANTS(s)
`FIRST NAME M.I.
`RESIDENCE
`(CITY AND EITHER STATE OR FOREIGN COUNTRY)
`t'"'l7
`La_!G'en_ce
`Sta~d, California
`
`TITLE OF INVENTION (280 characters ma:x:. l
`
`~r~~·--------------------------------------------------------------------------~
`~OKRESPONDENCE ADDRESS
`
`!rHOMAS _iL
`426 LOWELL
`PALO ALTO
`STATE rcA)
`
`McFARLANE_
`AVENUE
`
`I ZIP CODE
`
`94301
`
`I COUNTRY
`
`!UNITED STATES
`
`[X]
`[
`]
`[
`]
`[
`]
`
`Number of Pages:
`SPECIFICATION
`Number of Sheets:
`Drawing(s)
`Small Entity Declaration
`Other (specify)
`
`METHOD OF PAYMENT
`
`[X] A check or money order is enclosed to cover the Provisional filing fee
`[
`]
`The Commiss.ioner is hereby authorized to charge filing fees and credit
`Deposit Account Number
`
`PROVISIONAL FILING FEE AMOUNT ($): 75~00
`
`The invention was made by an agency of the United States Government or under a
`contract with an agency of the United States Government.
`
`] No
`[
`[>'] Yes, the name of the U.S. Government agency and the Government contract
`number are: National Science Foundation IRI-9411306-4
`
`Respectfully submitted,
`
`SIGNATURE:
`NAME:
`
`THOMAS J. McFARLANE
`39,299
`REG. NO.:
`
`DATE:
`
`[
`
`] Additional inventors are being named on separately numbered sheets
`attached hereto
`
`

`

`OU/UJ~~U~
`
`Attorney Docket No: 896-213
`
`THE COMMISSIONER OF PATENTS AND TRADEMARKS
`Washington, DC 20231
`
`11.632 \}. $ ~\ ?'! ~XPRESS MAIL" MAILING LABEL NO.: ',E F 0 55 9 6 9 2 9 8 us . DATE OF DEPOSIT
`lO .J'A.t-.:'1_ 'i1_
`~
`~tEREBY CERTIFY mAT TillS CORRESPONDENCE IS BEING DEPOSITED WITH THE UNITED STATES POSTAL SERVICE "EXPRESS
`\ ~lr~.~ 'I BOX PROVISIONAL PATENT APPLICATIONS,ASSISTANTCOMMISSIONERFORPATENTS, WASHINGTON, DC 20231
`' ' I t'l., MAIL POST OFFICE TO ADDRESSEE" SERVICE UNDER 37 CFR 1.10 ON THE DATE INDICATED ABOVE AND IS ADDRESSED TO:
`SIGNED _ ~_j~~--------DATE _l~-.s!::~IIJ -~-!, _____ _
`THOMAS J. MCFARLANE, REG. No. 39,299
`
`Sir:
`Transmitted herewith for filing is the provisional patent application of
`Inventor(s): Lawrence Page
`Improved Text Searching in Hypertext Systems
`For:
`
`Enclosed are:
`[ X]
`Provisional Application Cover Sheet
`J...$;' pages of Specification
`[>']
`[ ] __ sheet( s) of drawings
`[ ] A verified statement to establish small entity status under 37 CPR 1.9 and CFR 1.27
`[')(..]
`A check in the amount of $75 to cover the filing fee.
`
`Dated:
`
`\ /' 0 lCJ:r
`
`Respectfully submitted,
`
`~\~-
`Thomas J. McFarlane
`Reg. No. 39,299
`426 Lowell A venue
`Palo Alto, CA 94301
`(415) 321-6630
`
`

`

`0
`
`5
`
`10
`
`15
`
`20
`
`S96-213
`
`/'S-
`?-IY-
`60/035205
`
`A-jPR.Dv,
`
`Provisional Patent Application of
`Lawrence Page
`for
`~proved Text Searching in Hypertext Systems
`
`FIELD OF THE INVENTION
`This invention relates generally to techniques for database
`searching. More particularly, it relates to improved techniques
`for hypertext database searching.
`
`DETAILED DESCRIPTION
`
`The following appendix, attached hereto, provides a detailed
`description of the invention.
`lA·
`Appendix A:
`pages;
`TOTAL appended pages:
`(description and drawings).
`~+
`
`1
`
`

`

`1
`
`~'\-">"\-£~\)\~ p...
`Improved Text Searching
`in Hypertext Systems
`There is a demonstration system, called BackRub, usually available at
`http://zam.stanford.edu:1972/ from within Stanford. Or, contact me,
`page@cs.stanford.edu to arrange for a demo.
`
`Introduction and Summary
`
`Existing search engines on the web produce very poor results when the query
`matches large numbers of documents. Yet, these simple queries are very frequently
`issued by users. Described here is a system which yields radically improved results
`for these queries using the additional information available from a large database of
`web links. This database of web citations is used to determine a citation importance
`ranking for every web page, which is then used to sort the query results. This system
`has been implemented, and yields excellent results, even on a relatively small
`database of four million web pages. Not only does the system yield better results, but
`it does so at significantly reduced computational cost, which can be a very large
`expense for web search engines!. Demonstrating the improvement is as easy as
`picking a general query, for example "weather", and comparing the results to the
`results from a traditional web search engine, like AltaVista (the results section
`shows some sample queries).
`
`Motivation and Discussion of System
`
`Only recently with the advent of the web, have large numbers of people started to
`frequently search huge databases. The indexable part of the web is currently at least
`60 million documents totaling around 480 gigabytes2• AltaVista currently services
`23 million user queries per weekday3. Many of these queries do not give people
`reasonable results, and improving this situation is a very important problem,
`especially as the web is growing very quickly4. The web search engines are the
`busiest sites on the entire web.
`
`Many of the queries users perform in existing web search engines are simple one
`word queries which typicially return tens of thousands of documents that match the
`query. This problem is only going to get worse as the web grows. Since people are
`only able to examine a handful of documents, only the highest ranked documents
`
`1 Alta Vista is run on several large computers that each cost around a million dollars. And, every time
`they add another machine, the demand increases instantly to fully utilize the new equipment.
`21 recently downloaded 30 million web pages, and estimated this figure.
`3From http:/ I altavista.digital.com/.
`4Some researches have claimed that the web has been doubling in size every 6 months or so.
`
`

`

`2
`
`are ever seen. Most current search engines rank documents according to how often
`the search terms appear relative to the size of the document, and how close the
`matched terms are to the beginning of the document. So, a typical search for
`"university" will yield all the documents that have the term "university" several
`times very close to the beginning of the document. This ranking method yields
`documents that are quite random and of low quality (see the attached results
`comparison section). Many experts would claim that the results for a query
`"university" are not good because the user has not specified enough information to
`narrow their search, and they should be searching for something like "stanford
`university professor winograd". However, if all the user knows is that they are
`interested in universities, it makes much more sense to return general documents
`of high "quality" or "importance" that seem to be about universities, rather than the
`top ten out of 50,000 documents which happen to "best" include the query terms. By
`analyzing all the links on every web page we can compute a useful ranking that
`approximates a "quality" or "importance" criteria.
`
`My demonstration system, called BackRub, returns almost entirely university
`homepages for the query "university" (see the attached section for comparisons of
`the query results). These are much more reasonable results for this "university"
`query then the standard approach. For this query, and most other simple queries,
`this system gives greatly improved results over the standard searching methods in
`use today, and at significantly lower computational cost.
`
`While the system gives great improvement for simple, under constrained, queries,
`it also works slightly better for very specific queries, like "Jeffrey Ullman". These
`specific queries work quite well in the existing search engines, because there are
`usually few documents that match, and any that match a query like "Jeffrey
`Ullman" are usually close to what you are looking for. So, since these queries work
`well in existing search engines, it is impossible to radically improve the results.
`However, my search engine will tend to return the most significant page that
`matches.
`In this case, BackRub finds Jeff's main home page and returns it first. It is
`interesting to note that the home page Jeff does not maintain, in the College of
`Engineering, with very little information, is returned last in my system as it should
`be (again see the attached results section). However, Alta Vista returns Jeff's college
`of engineering page as the first matching Stanford page, which is clearly not correct.
`So, my system still improves the search a fair amount, but not as much as an under
`constrained query where we have to pick between tens of thousands of possible
`documents to return.
`
`PageRank -- An Approximation to "Importance"
`I
`The reason why my $ystem works so well, is that it decides which documents to
`return, and in what <f>rder, by using an approximation to how well cited, or
`"important" the mat&ing documents are. I will call this aproximation to
`importance PageRank from now on. Web pages get a higher PageRank from being
`mentioned on other pages. But, the PageRank a page gains from a citation is based
`
`

`

`3
`
`on the PageRank of the page that cites it. This definition may sound circular because
`it is in fact circular. But, it turns out with a few small modifications, we can still
`compute a PageRank of this form. So a page can have high PageRank even if it only ·
`has one citation, such as from an ad on the Yahoo home page, which of course has
`high PageRank. Or, a web page can get a high PageRank by being pointed to from
`thousands of other pages, like the Netscape download page which has 31,284 pages
`that point to it in my system. The intuition is that if your query matches tens of
`thousands of documents, you would be happier looking at documents that many
`people thought to mention in their web pages, or that people who had important
`pages mentioned at least a few times.
`
`Detailed Description
`
`Although there are many components required to make the demonstration system,
`BackRub work, only the two main components, the PageRank system, and the
`search engine will be discussed here. The PageRank model and especially its use in
`search is what we believe to be new knowledge.
`
`PageRank
`
`This section will discuss the model behind PageRank, in practice what kind of
`documents have high PageRank, and possible future improvements in the model.
`It is interesting to note, that while we have been discussing use of PageRanks in
`queries, it is useful in its own right to have an objective measure for web documents
`in general.
`
`PageRank Model
`
`The simple motivation for the PageRank model is the following: Assume you have
`a set of weights for each page such that a page's weight is distributed evenly among
`its children. And, each page's weight is the sum of all the weights distributed from
`its backlinks. The analogy is that we have a hypothetical user who surfs the web
`randomly. Whichever page the user is on, they keep following links randomly
`from page to page, with an equal probability of picking any link on the current page.
`The probability that they will visit a given page is its PageRank. Obviously, any user
`who surfs randomly is much more likely to end up at the Netscape home page, then
`my modest home page, for example. In any case, the model turns out to be
`equivalent to computing the first eigenvector of the web which can be easily done
`using an iterative method.
`
`The catch to this method is that the web has lot$ of cycles in its graph, and is not
`fully connected. So, we have what I like to calli the cyclotron effect, where all the
`ranking "energy" ends up getting sucked into s~allloops, and goes around forever.
`In our random surfer analogy, this corresponds to the user getting stuck clicking
`back and forth between two pages which only point to each other. If such a "loop"
`exists, the random surfer will always end up in' one if we assume surfing goes on
`
`

`

`4
`
`forever. In reality, users do not surf forever only following links on the page they
`are currently on. Usually, people do another unrelated task after clicking through
`just a few links. To roughly model this, we assume that there is a small probability
`(15% in the current system) that the random surfer will jump to any random page
`in the system from the current page. This means a page's influence decays to be
`very small after traversing several links. · In practice, the modified model works
`very well, but determining the 15% figure above is a matter of experimentation. If
`we set the damping amount to be 100% we are simply computing the number of
`backlinks (citations) that point to each page. This would probably work fairly well,
`but not nearly as well as the iterative method described here which takes into
`account things like advertisements on major sites. If we set the damping factor to
`zero, then we have the cyclotron effect, and no reasonable ranking. So clearly, some
`intermediate value for the damping is reasonable. An alternative method to
`damping is to simply only iterate a limited number of times, for example two, then
`we have computed something like backlinks of backlinks.
`
`Possible Improvements to Model
`
`One problem with this model is that it basically assumes that every page has some
`importance, because we are distributing 15% of the total PageRank to every page
`uniformly. But, importance has to start somewhere-- it needs to have some basis.
`As a malicious web page author, I could automatically create a large number of pages
`that all point to my home page, which will cause my home page to get a high
`PageRank. This is a problem, because there is a great commercial interest to be at the
`top of the results of any search engine, and many people try to 'beat" the system by
`this sort of "spamming". I have considered distributing the damping 15% just to the
`major sites, like Netscape, which would likely completely alleviate this problem,
`and change the ranking little. But I need to do experiments to verify this. Another
`likely option for improvement that also helps solve the "spamming" problem is to
`compute how much of a page's PageRank is due to only the pages on the top several
`contributing servers. This would give a measure of how "general" a page's appeal
`is, which also might be useful for general search. Pages which had a very high
`PageRank that was based on only a very small number of servers would likely be
`not of general interest. This is because if the page were of general interest, many
`servers would likely have links to the page. Or, the page could be a page of general
`interest that was only recently created, and no one but major sites have known
`about it to create the link. This is a very difficult problem to solve in general.
`Copyright pages, which are often linked to from every page on a commercial site,
`often get a high PageRank only from links on one server. Almost no one links to
`these pages except for the site itself. The modification to consider on and pff site
`links differently would help alleviate this copyright page problem. In pra,ctice, even
`though the copyright pages have a high PageRank, they are not too much jof a
`problem when querying because they do not match that many queries.
`!
`
`The model could be much more sophisticated and take into account actua,l user
`behavior based on many parameters of the links on a page. For example, the model
`
`

`

`5
`
`could take the font size and position on the page of a link as factors which
`determine how likely a link is to be clicked on. Another possible improvement is to
`weight more recently modified documents more highly. These types of
`improvements could yield significant gains in the quality of search results.
`
`Implementation of PageRank
`
`The iterative method used to compute the model can be very efficiently
`implemented. My implementation uses only half the RAM required to hold all the
`weights for the pages. The rest of the link information is kept on disk, and read
`sequentially. I have computed this model to· reasonable convergence for 30 million
`pages and about 400 million links in several hours on a large workstation. This is a
`very small computational cost compared with building a full text index of the same
`amount of information. So, this computation required to compute PageRank is
`insignificant compared to the rest of the computation required to run a web search
`engine.
`
`PageRank in Practice
`
`In practice, a high PageRank seem to indicate several things. One is popular
`content, that is pages that people like a lot, and add to their public bookmarks, or
`otherwise mention on their pages. Good examples of this category include things
`such as search engines like Alta Vista or popular web soap operas such as The Spot,
`both of which have high PageRanks, and large numbers of backlinks. A root of a
`hierarchy also tends to get a very high PageRank because all the nodes in the
`hierarchy point to it, as do many other people because it is the root, and you can rely
`on it being there and being a good place to start navigation. Examples of such pages
`might be a university homepage, company homepage, personal homepage, or
`Yahoo. Things people tend to put on their pages also end up with very high
`PageRanks. This includes things like the EFF's Blue Ribbon Campaign, and several
`companies that provide web counters automatically when you create a link to them,
`and the Excite search box which searches a local site.
`
`While PageRank is not intended to approximate usage, it probably has a fairly high
`correlation with actual usage of web pages for most types of pages. Notably, things
`like sex, which people do not feel comfortable mentioning in their web pages, have
`a low PageRank, but high usage. PageRank may be better than usage data for
`measuring "eye traffic", that is how many people see various things. For example,
`many people see the Blue Ribbon logo all the time, but likely rarely follow it because
`they already know what it is. Brand name ads would fall into a similar category,
`where people see them a lot, but rarely follow them.
`
`Search Engine
`
`The search engine is quite simple and is primarily based on PageRank. Currently, it
`searches only the titles of the web documents due to space constraints, and returns
`
`

`

`6
`
`the documents that match sorted by PageRank. It also does duplicate detection, and
`grouping by site. If the search engine was extended to search full text, some function
`of the full-text match value, and the PageRank value would have to be constructed
`to yield good results. Developing a reasonable rank-merge function would likely be
`a matter of some simple experimentation. If I simply did the same thing I am doing
`now on full text, any query that matched any part of Netscape's home page would
`always return Netscape's home page first, which is not a reasonable response.
`
`When searching with an existing search engine, often a user finds a
`semi-relevant page, and then surfs from that page to what they are actually looking
`for. This means, that the user tends to end up at pages that have a lot of links, which
`tend to be of higher quality. My search engine helps automate this process, simply
`returning the higher quality pages first rather than relying on intermediate
`searching. Also, since the pages my system returns tend to be roots of hierarchies,
`they are often the most efficient place to start navigating from, assuming the user
`can not express exactly what they want using a boolean query specific enough to
`return only a few results.
`
`Possible Enhancements to Search Engine
`
`Besides full-text searching which is a necessary enhancement, another possible
`enhancement to the search engine is to search the titles of the backlinks to a
`document as wen; and include them in the text-matching rating. Or, the text that is
`right around the neighborhood of a backlink could also be included in the text(cid:173)
`matching rating. The intuition is that by looking at the text that occurs in the
`neighborhood around all the links to a page, we could get a good summary of a
`pages content. I am working on implementing these enhancements in the near
`future.
`
`For certain queries, like "weather", the search engine already yields excellent Yahoo
`like categorizations automatically. Generating a set of queries for common
`categories and simply recording the results might be enough to build a reasonable
`automatically generated Yahoo like categorization system.
`
`Possible Close or Related Patents
`
`There is a company, which I have recently become aware of, located at
`http:/ /www.linkrank.com/. They seem to be getting similar results, but they do not
`explain their methods in any detail. They claim they have a patent pending.
`
`Results Explanation
`
`I have chosen to display Alta Vista vs. BackRub (my system). Most web search
`engines seem to give similar results to Alta Vista.
`
`

`

`7
`
`The PageRank citation importance number, which determines the order of results,
`is depicted by the bar graph and number below it on the BackRub results. High
`PageRank indicates a well cited page. The bar graph gives you some idea of the
`absolute ranking of a page. A full bar graph is only displayed for Netscape's
`download page (the most highly ranked page at 6033.37). Obscure pages with only
`one link to them have a completely empty bar graph and a PageRank of much less
`than ones, and so are usually not returned unless there are very few documents that
`match a query.
`
`BackRub's database is smaller (4 million pages) versus Alta Vista's 30 million. Since
`BackRub can only search titles because of space and processing constraints, we have
`restricted Alta Vista to search titles as well. Title searching generally seems to
`improve AltaVista's results for the kinds of queries we have displayed here, so we
`have not handicapped Alta Vista.
`
`Please feel free to try it out yourself and compare the results:
`http:/ /zam.stanford.edu:1972/ If it does not work, let me know and I'll start it up
`for you. Please keep in mind the smaller database, and lack of full text search. Also,
`the database only contains web pages which are inside of the US. Queries for city
`names, and state names, and companies with a reasonable web presence generally
`yield excellent res~lts, and are good things to try first.
`
`Example Queries (Printouts Follow)
`
`1. "university"
`BackRub returns major universities homepages, with only a few inappropriate
`pages. Alta Vista returns pages which have university twice in their title first, which
`seem to be on relatively random topics. Note that there is a slight Stanford bias in
`the BackRub data, since I started downloading web documents at Stanford, so the
`Stanford ratings are slightly higher than you would normally expect.
`
`2. "weather"
`Note that the first four documents returned by BackRub are the major weather sites,
`The Weather Underground, The National Weather Service, Intellicast, and The
`Weather Channel. The indented results are the results that are on the same server,
`and should be generally considered as a unit with the documents above. A member
`of one of these weather companies considers the results BackRub returned to be
`very representative of the good weather services that are available. Indeed the
`results even represent a reasonable Yahoo like categorization of the topic. None of
`these four main sites that BackRub returned first are even il1. the first three pages of
`AltaVista's results! Note that if you were looking for weather information using
`Alta Vista, you would scan through the list returned by Alta Vista, and you would
`
`5The bar graph is a log scale, because the PageRanks seem to have a Zipfian distribution, as would be
`expected for any citation ranking system.
`
`

`

`find a page that appeared to list a bunch of weather sites. 'fl!ten, you would go to that
`page and then try to follow a link to a major site. This is exactly what BackRub and
`PageRank are automating for you, by returning the pointers to the major sites first.
`
`3. "Jeffrey Ullman"
`Refer to the previous Motivation and Discussion section for an explanation of the
`results for this query.
`
`8
`
`

`

`Bil.ckR.ub Search: university
`
`BackRub Query Results
`
`Bac:kRub's Highest Ranked Sites
`
`University of Illinois at Urbana-Champaign
`http://www .uiuc.edu/
`694.687 8460 backlinks 12k - 10125196 - 11/1196
`
`Stanford University Homepage
`http://www.stanford.edu/
`609.303 8857 backlinks 4k - none - 1111196
`
`Stanford University: Portfolio Collection
`--f~JU~ http://www. stanford.edu/home/administration/portfo lio. html
`167.919 34 backlinks
`
`University of California, Irvine
`http://www.uci.edu/
`273.621 2390 backlinks 2k - 10120/96 - 1111/96
`
`Baylor University
`http://www .baylor.edu/
`218.372 2761 backlinks 6k - none - 1111196
`
`Glimpse Working Group - University of Arizona, CS Dept.
`http://glimpse.cs.arizona.edu/
`214.736 663 backlinks 9k - none - 1111/96
`--~!l!l®l Duplicate: http:// glimpse. cs.arizona.edu.·19941
`
`Northwestern University: NUinfo
`http://nuinfo.nwu.edu/
`204.169 1227 backlinks 3k - 9128196 - 1111196
`--~!l!l®lDuplicate: http://www.nwu.edu/
`
`University of Colorado at Boulder
`http://www .colorado.edu/
`182.67'1 3015 backlinks 4k - none - 11/1196
`
`Iowa State University Homepage
`http://www .iastate.edu/
`178.431 2341 backlinks 3k - 9113196, - 1111196
`I
`The George Washington University ~orne Page
`http://www.gwu.edu/
`i
`174.856 2222 backlinks 2k - none - [ 1111/96
`
`.
`University of Virginia
`http://www. virginia. edu/
`168.576 2724 backlinks 3k - none - 11/1/96
`
`Monday, December 16, 1996
`
`http://zam. stanford. edu: 1972/cgi/title/
`?searchterm=university
`
`

`

`BackRub Search: university
`
`University of Wisconsin-Madison: WisciNFO
`http://www. wisc.edu/
`164.481 2331 backlinks 1 k - 9121/96 - 1111196
`
`University of Pennsylvania
`http://www .upenn.edu/
`163.961 2755 backlinks 1k - 11/1196 - 11/1/96
`
`The University of Michigan
`http://www. umich.edu/
`151.114 2874 backlinks 1k - none - 1111196
`
`Indiana University
`http://www .indiana.edu/
`146.445 1409 backlinks 1k - 9/28196 - 1111/96
`
`Indiana University copyright statement
`--~!l\!l~~ http://www .indiana.edu/copyright.html
`88.9323 703 backlinks
`
`Carnegie Mellon University
`http://www .cmu.edu/
`144.987 1762 backlinks 2k - 3/24196 - 1111/96
`
`JHUNIVERSE: Johns Hopkins University on the Web
`http://www .jhu.edu/
`140.429 2060 backlinks 4k - 10/27196 - 1111196
`
`University of Minnesota
`http://www.umn.edu/
`136.827 2489 back1inks Ok - 10111196 - ll/1196
`
`University of Minnesota, Twin Cities Campus
`--all~ http://www.umn.edu/tc/
`102.56 424 backlinks
`
`The University of Delaware
`http://www. udel.edu/
`132.334 1598 backlinks 4k - 10122196 - 1111196
`
`Welcome to the University of Chicago
`http://www .uchicago.edu/
`130.815 2770 backlinks 4k - 10118/96 - 1111/96
`
`The University of Iowa
`http://www.uiowa.edu/
`118.989 1315 backlinks 2k - 9/3196 - 1111/96
`
`University of Florida Home Page
`http://www. ufl.edu/
`117.691 1997 backlinks 5k - 10114/96 - 11/1/96
`
`USCWeb, the University of Southern California
`http://www.usc.edu/
`115.256 1573 backlinks 4k - none - 1111196
`Monday, December 16, 1996
`
`http://zam.stanfo rd.edu: 1972/cgl/title/
`?searchterm=universlty
`
`

`

`BackRub Search: university
`
`Welcome to New York University
`http://www.nyu.edu/
`113.217 1884 backlinks 5k - 10/22/96 - 1111196
`
`Michigan State University
`http://www.msu.edu/
`109.508 1623 backlinks 2k - 10127/96 - 1111196
`
`Cornell University Home Page
`http://www .cornell.edu/
`108.117 2319 backlinks 4k - none - 1111196
`
`Purdue University
`http://www.purdue.edu/
`106.65 1495 backlinks 3k - 9/3/96 - 1111196
`
`Harvard University WWW Home Page
`http://www.harvard.edu/
`101.376 2014 backlinks 4k - 10112196 - 1111196
`
`Mississippi State University
`http://www.msstate.edu/
`99.1736 1465 backlinks 3k - none - 1111196
`
`Alfred University
`http://www .alfred.edu/
`96.4112 1281 backlinks 'Jk - 9111/96 - 11/1196
`
`UNICORN: Kansas State University's Information System
`http://www .ksu.edu/
`94.9499 1388 backlinks 3k - 8122/96 - 11/1196
`
`University of Washington Home Page
`http://www.washington.edu/
`93.514 1376 backlinks 4k - 1111196 - 1111196
`
`University of California. Santa Barbara
`http:/lwww. ucsb .edu/
`91.8778 1685 backlinks 3k - none - 1111196
`
`Boston University Home Page
`http://web.bu.edu/
`91.7343 1435 backlinks 14k - 10122196 - 1111/96
`
`North Dakota State University
`http://www.ndsu.nodak.edu/
`90.9534 770 backlinks 4k - none - 11/1196
`
`University of Maryland at College Park
`http://www. umcp. umd.edu/
`85.7528 1278 backlinks 4k - 10125196 - 1111/96
`
`University of Missouri - Rolla
`http://www.umr.edu/
`85.5691 1233 backlinks 2k - none - 1111196
`Monday, December 16, 1996
`
`http://zam.stanford.edu: 1972/cg i/title/
`?searchterm=university
`
`

`

`BackRub Search: university
`
`Princeton University - Home Page
`http://www .princeton.edu/
`85.179 1489 backlinks 4k - none - 1111/96
`
`Bradley University
`http://www. bradley.edu/
`84.435 1171 backlinks 5k - none - 1111196
`
`Monday, December 16, 1996
`
`http://zam.stanford.edu: 1972/cgi/title/
`?searchterm=university
`
`

`

`I and Display the Results I in Standard Form
`Search
`the Web
`Lt:!~_l:~~:""!:~!Y~~_!ty_ ~~"'"'-~-----,--"~-~~---"~-'"'"_, __ . ____ , ___ ", ____ .J ( Submit)
`Tip: To find a page from a given site, try: host:thls.slte.eom
`
`Word count: title:university: about 60000
`
`Documents 1-10 of about 60000 matching the query, best matches first.
`
`DePauw University Introdnctjonllntent of Uniyersity Policy
`Introduction/Intent of University Policy. Both use and abuse of alcohol have become widespread in our society. While
`most people are able to use alcohol...
`httn·!Jwww.depauw.edubtulife5/gctionl.htm -size 2K- 20 Feb 96
`
`University Medical Center <UMC) • University Physicians • Medical Offices -·Ar
`Internal Medicine Office 6th floor at University Medical Center 1501 N. Campbell Avenue Tucson, Arizona 85724
`(520)694-8888. Description of Facility. The ..
`httrr!Jwww ahsc arizonq edu!-umclinmedoff htm -size 7K- 12 Apr 96
`
`-- Uniyersity Physicians .. Physicians Resourc
`Unjyersity Medical Center <UMC)
`Physicians' Resource Service (for physicians only) The Physicians' Resource Service provides 24-hour priority physician
`access to faculty physicians, ...
`htm·/lwww.ahsc.arizona edul-umclresourc.htm ·size 3K- 12 Apr 96
`
`UNIVERSITY ARCHIVES. UNivERSITY OF MISSOURI-ST. LOUIS
`UNIVERSITY ARCHIVES, UNIVERSITY OF MISSOURI-ST. LOUIS. ADDRESS: 8001 Natural Bridge Rd. Thomas Jefferson
`Library St. Louis, MO 6312~. TELEPHONE: (314) ...
`httrrlllibrary wustl edul-speclarchjyey/aslaa/umsl-qrchjves html -size 2K- 16 Jun 96
`
`The University of Queensland Enrolments 1995
`The University of Queensland. STUDENT LOAD. 1994 1995. By Course Level: Higher Degree 2891 3032 Bachelor Degree
`16165 16269 Other 1545 1630. By Group: ...
`httv:flwww uq.edu.aul-adbamesblstats/[oad.html -size 1K- 14 Dec 95
`
`DePauw University Violations of University Regulations
`Violations of University Regulations. The consupmtion and serving of alcoholic beversges involves responsibility for
`both individuals and group