AN OVERBROAD PATENT ON automatic keyword extraction - 10 minutes of your time can help narrow US patents in the future. Follow @askpatents on twitter to help.

QUESTION - Have you seen anything that was published before 7/2/2009 that discusses using word co-occurrence scores to find key phrases?

I am looking for prior art on US 8,131,735 Rapid automatic keyword extraction for information retrieval and analysis

  • Publication Number: US8131735 B2
  • Assignee: Battelle Memorial Institute
  • Prior Art Date: Seeking prior Art predating 7/2/2009
  • Link to Google Prior Art Search - "Find Prior Art"

Claim 1 requires each and every step below:

A computer-implemented method of extracting keywords from an individual document, the method comprising:

  1. Parsing an individual document by delimiters and stop words to identify candidate keywords;

  2. Determining co-occurrences of words within the candidate keywords;

  3. Calculating word scores for each word within the candidate keywords based on a function of co-occurrence degree, co-occurrence frequency, or both;

  4. Calculating a keyword score for each candidate keyword based on a function of word scores for words within the candidate keyword; and

  5. Selecting a portion of the candidate keywords to extract as keywords based, at least in part, on the candidate keywords with highest keyword scores.

The notion of using word co-occurrence scores to find key phrases has been documented. (It seems like every information retrieval system ever built is prior art.)

For example, in this paper: http://citeseerx.ist.psu.edu/viewdoc/download?doi=

We find the following quote:

Various phrase-finding and indexing methods have been proposed in the past and generally, retrieval performance conclusions on the use of phrases as indexing units were inconsistent. Salton & McGill [23] suggested generating statistical phrases based on word co-occurrence and then incorporating them into document representation as additional index terms.

The Salton & McGill reference is from this book, published in 1983: http://books.google.ca/books/about/Introduction_to_modern_information_retri.html?id=7f5TAAAAMAAJ&redir_esc=y

  • From the background and summary section it seems that they acknowledge extracting keywords and key phases by co-occurance is old. What they believe they bring to the table is an efficient way to get multi-word candidates without a complex analysis of the whole document.
    – George White
    Commented Oct 24, 2013 at 2:55

5 Answers 5


This sounds literally like the textbook way of doing a search. As in, the textbook I used in my CS 553 class at Brigham Young University in Fall 2003 taught using this method.

  • 1
    My thoughts, exactly. Going through claim one, the steps seem to be: lex (although they call it 'parse'), check co-occurence, score, aggregate, sort, limit. These are all old tricks in CS. I also wonder about the line that "Claim one requires each and every step below". Does that mean that if I implement the algorithm, but I don't sort by score, or I sort by score but I use the whole list rather than just a portion of it, then I'm not infringing on the patent? Commented Oct 28, 2013 at 14:46
  • UYes, to infringe a claim one must do everything positively required by the claim wording. Determining what that is is usually a big part of any patent dispute. Also, the ease of implementing a invention after reading about it is not relevant in a patent ability analysis. It is hard to imagine a generic textbook provides a disclosure of this technique when, apparently, it was not specifically disclosed in the 10 references cited on the front of the patent. They include patents by IBM and Microsoft that are specific to this exact field.
    – George White
    Commented Oct 28, 2013 at 18:40

Here's a really cool web example of this very topic: http://fivefilters.org/term-extraction/

What is Term Extraction?

Term Extraction from FiveFilters.org is a free software project to help you extract terms (e.g. for use as tags) through a web service. Given some text it will return a list of terms with (hopefully) the most relevant first.


Here's a PHP Tip that does very much the same, assuming you consider a web page a document..

PHP tip: How to extract keywords from a web page

April 13, 2008 Topics: PHP, Text processing Technologies: PHP 5+, UTF-8 Web page keywords characterize the page's topic for a search engine. Extracting keywords requires that you recognize the page's character encoding, strip away HTML tags, scripts, and styles, decode HTML entities, and remove unwanted punctuation, symbols, numbers, and stop words. This article shows how.



The textbook Foundations of Statistical Natural Language Processing, published in 1999, written by Christopher Manning and Hinrich Shutze might contain prior art for co-occurence.

  • Thanks for your answer! If possible, could you include a brief relevant snippet from that book? That would help anyone investigating this understand whether it's worth looking deeper into. Commented Dec 11, 2015 at 18:55

Wouldn't Markov chains be related? Any Markov chain automatic configurator would naturally be doing step # 3 if not more.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .