System for classifying words

• US 20070136048A1
• Filed: 12/13/2005
• Published: 06/14/2007
• Est. Priority Date: 12/13/2005
• Status: Active Grant

• Current Assignee: R2 Solutions LLC (Acacia Research Corporation)
• Original Assignee: Yahoo! Inc. (Apollo Global Management, Inc.)
• Inventors: Riise, Soren, Richardson-Bunbury, David, Stipp, Eugene, Patel, Devish, Grealish, Paul
• Granted Patent: US 7,729,901 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 16/367 Ontology
• CPC Class Codes: G06F 40/44 Statistical methods, e.g. probability models
• CPC Class Codes: G06Q 30/0277 Online advertisement

Abstract

A system is disclosed for determining probable meanings of a word. A prior probability is established of the probable meanings of the word. A context frequency probability is established of the probable meanings of the word. The probability that each meaning is a correct meaning may be provided in accordance with both the prior probability and the context frequency probability.

Claims

1. A method for determining probable meanings of a word, the method comprising:

establishing a prior probability of the probable meanings of the word;

establishing a context frequency probability of the probable meaning of the word; and

providing the probable meanings of the word including the probability that each meaning is a correct meaning in accordance with both the prior probability and the context frequency probability.

2. The method of claim 1, wherein the context frequency probability comprises the meaning of the word in a context of terms immediately preceding or immediately following the word.

3. The method of claim 2, wherein the context frequency probability comprises the meaning of the word in a context of terms both immediately preceding and immediately following the word.

4. The method of claim 3 further comprising applying an inverse to combine function to combine the probabilities of the context frequency for the terms immediately preceding and immediately following the word.

5. The method of claim 1, wherein the prior probability comprises a probability that the word refers to a predetermined meaning irregardless of the context in which the word is used.

6. The method of claim 1, wherein the prior probability and the context frequency probability are combined using combine mathematical function.

7. The method of claim 1 wherein establishing a context frequency probability further comprises calculating a conditional probability that a given context implies a predetermined meaning of the word.

8. The method of claim 7 further comprising performing a training of the meanings of the words to determine an actual probability that a given context implies a predetermined meaning of the word.

9. The method of claim 8 further comprising combining two probabilities using an inverse combine function to derive a conditional probability

10. The method of claim 1 further comprising applying an adjustment value to normalize the probabilities of the meaning of the words.

11. The method of claim 10 wherein after the adjustment value is applied, a sum all of the probabilities of the possible meanings of the words equals about one-hundred percent.

12. The method of claim 8 wherein the adjustment value is determined by an iterative mathematical process, not directly.

13. The method of claim 1 wherein the prior probability comprises a location related probability which includes the probability that the word refers to a location.

14. The method of claim 1 further comprising moderating the prior probability for small samples by adding a value of one to the numerator and a value of two to the denominator.

15. The method of claim 1 further comprising establishing a disambiguators in accordance with a text in which the word is used.

16. The method of claim 15 wherein the disambiguators comprise specific disambiguators which include the meaning of the word as determined in a context of a text that includes the word, other than the text immediately preceding or following the word.

17. The method of claim 15 wherein the disambiguators comprise general disambiguators which includes a term in a text that uses the word, wherein the term shares an attribute with the word being disambiguated.

18. The method of claim 1 further comprising determining an immediate small relative context to moderate a long context.

19. The method of claim 18 further comprising storing a probability value for the long context only if sufficiently different from the small context.

20. A method for training an actual meaning of names, the method comprising:

identifying a name to be trained;

collecting texts that include the name;

searching the texts for the name;

identifying an ambiguity level of the name from the search;

obtaining an input regarding the name as used in the text; and

analyzing results based on the inputs to generate statistics about probable meanings of the name.

21. The method of claim 20 further comprising utilizing the generated statistics to re-analyze the texts.

22. A method for analyzing double-sided context frequency values, the method comprising:

applying a combine function;

applying an inverse combine function to determine a resulting value;

determining whether the resulting value differs from a neutral context frequency; and

utilizing the double-sided context frequency value if the resulting value differs from the neutral context frequency.

23. The method of claim 22 further comprising storing the inverse combine function value only if the resulting value differs from the neutral context frequency.

24. A method of determining probable meaning of a word, the method comprising:

creating a document classifier;

applying the document classifier across a number of documents; and

calculating probabilities for different meanings of the word.

25. The method of claim 24 further comprising disambiguating the word in the context of a text.

26. The method of claim 25 further comprising re-calculating the probabilities for the different meanings of the word in accordance with the disambiguating.

27. A system to determine probable meanings of a word, comprising:

means for establishing a prior probability of the probable meanings of the word;

means for establishing a context frequency probability of the probable meaning of the word; and

means for providing the probable meanings of the word including the probability that each meaning is a correct meaning in accordance with both the prior probability and the context frequency probability.

28. The system of claim 27, wherein the context frequency probability comprises the meaning of the word in a context of terms immediately preceding or immediately following the word.

29. The system of claim 27, wherein the prior probability comprises a probability that the word refers to a predetermined meaning irregardless of the context in which the word is used.

30. The system of claim 27 further comprising means for applying an adjustment value to normalize the probabilities of the meaning of the words, wherein after the adjustment value is applied, all of the probabilities of the possible meanings of the words are summed to equal about one-hundred percent, and the adjustment value is determined by an iterative mathematical process, not directly.

31. The system of claim 27 further comprising means for establishing a disambiguators in accordance with a text in which the word is used.

32. The system of claim 31 wherein the disambiguators comprise specific disambiguators which include the meaning of the word as determined in a context of a text that includes the word, other than the text immediately preceding or following the word.

33. The system of claim 31 wherein the disambiguators comprise general disambiguators which includes a term in a text that uses the word, wherein the term shares an attribute with the word being disambiguated.