Method for generating descriptors for the classification of texts

• US 6,038,527 A
• Filed: 03/14/1997
• Issued: 03/14/2000
• Est. Priority Date: 07/19/1995
• Status: Expired due to Fees

• Current Assignee: Daimler-Benz AG (Mercedes-Benz Group AG)
• Original Assignee: Daimler-Benz AG (Mercedes-Benz Group AG)
• Inventors: Renz, Ingrid
• Primary Examiner(s): Isen, Forester W.
• Assistant Examiner(s): EDOUARD, PATRICK NESTOR
• Application Number:
  US08/809,080
• Time in Patent Office: 1,096 Days
• Field of Search: 704/1, 704/9, 707/500, 707/530, 707/531
• US Class Current: 704/9
• CPC Class Codes: G06F 16/353 into predefined classes
• CPC Class Codes: G06F 40/268 Morphological analysis
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

The proposed method for generating descriptors for the classification of texts provides a breakdown of more complex word forms by way of matching with the entirety of word forms occurring within a compilation of training texts. No morphological or linguistic knowledge base is necessary for the preferably cyclically continued breakdown, nor for the accompanying drawing up of stop word prefix and suffix lists. Simple morphological knowledge is provided by prescribing minimum requirements with respect to the form of descriptors and text sections. The method is particularly flexible and can be easily adapted to new applications. The method is also very error-tolerant and thus particularly suited for the classification of digitized texts which are produced from written texts by means of character recognition methods or from spoken texts by means of language recognition methods.

Claims

1. A method of generating descriptors for natural language texts, using a plurality of training texts having a plurality of words, comprising the steps of:

extracting words from a text during a training phase on the basis of the training texts;

predetermining a minimum structure of said descriptors;

breaking down words in the text into shorter word segments, wherein each shorter word segment within a longer word segment must meet said minimum structure for said breaking down to be permitted; and

matching said word segments that remain in the text against each other to generate a list of descriptors.

2. A method according to claim 1 wherein the breakdown step is carried out repeatedly, with the word segments formed in preceding breakdown cycles being treated like word forms that occur in the text, and the descriptors are formed from the word forms and word segments which are left after the last breakdown cycle.

3. A method according to claim 1 wherein, during the formation of the descriptors from the remaining word forms and word segments, suffixes and prefixes are separated from them.

4. A method according to claim 3, wherein the suffixes and prefixes are statistically determined from the word forms occurring in the training texts.

5. A method according to claim 1, wherein a stop word list is predetermined and word forms contained therein are not considered in the text.

6. A method according to claim 5, wherein the word forms of the stop word list are statistically determined from the training texts as the word forms which most frequently occur therein.

7. A method according to claim 1, wherein a frequency threshold is predetermined for the descriptors and only those descriptors are used whose frequency in the training texts exceeds the threshold.

8. The method according to claim 4, comprising the following steps:

removing word segments not meeting the minimum structure from consideration as descriptors, wherein the minimum structure includes word segments having at least three characters, one of which must be a vowel; and

separating said prefixes and suffixes from said word segments at a conclusion of said breaking down step.