| System and method for automatically classifying text -> Monitor Keywords |
|
|
  System and method for automatically classifying textUSPTO Application #: 20060143175Title: System and method for automatically classifying text Abstract: A method is provided for automatically classifying text into categories. In operation, a plurality of tokens or features are manually or automatically associated with each category. A weight is then coupled to each feature, wherein the weight indicates a degree of association between the feature and the category. Next, a document is parsed into a plurality of unique tokens with associated counts, wherein the counts are indicative of the number of times the feature appears in the document. A category score representative of a sum of products of each feature count in the document times the corresponding feature weight in the category for each document is then computed. Next, the category scores are sorted by perspective, and a document is classified into a particular category, provided the category score exceeds a predetermined threshold. (end of abstract)
Agent: Schwegman, Lundberg, Woessner & Kluth, P.A. - Minneapolis, MN, US Inventors: Igor Ukrainczyk, Max Copperman, Scott B. Huffman USPTO Applicaton #: 20060143175 - Class: 707006000 (USPTO) Related Patent Categories: Data Processing: Database And File Management Or Data Structures, Database Or File Accessing, Query Processing (i.e., Searching), Pattern Matching Access The Patent Description & Claims data below is from USPTO Patent Application 20060143175. Brief Patent Description - Full Patent Description - Patent Application Claims
RELATED APPLICATIONS [0001] The following application is relied upon and hereby incorporated by reference in this application: [0002] U.S. Provisional Application No. 60/206,975, entitled "System and Method for Providing a Text Classifier," bearing attorney docket no. 07569-6004-00000. TECHNICAL FIELD [0003] The present invention is directed to text classification and, more particularly, to a computer-based system for text classification that provides a resource that can be utilized by external applications for text classification. BACKGROUND [0004] Throughout the entire period of recorded history, people have memorialized their thoughts, actions, hopes and dreams on a daily basis. Prior to the latter part of the 20.sup.th century, this recorded history was typically written for exchange between human beings without any expectation that the information would be stored in a machine or otherwise converted into a machine-readable format. At that time, archives of this information resided in countless document depositories, complicating access to, and retrieval of the information contained in the documents. In the past 30 years, efforts have been underway to archive these "natural language" documents on various other media More specifically, the development of the personal computer has led to the creation of an unprecedented amount of machine-readable information. Improvements in scanner technology have additionally led to the conversion of documents from hardcopy documents into machine-readable documents. This technology, together with similar advances in mass storage has led to the conversion of natural language documents into machine-readable documents at an unprecedented level. Today, documents generated by a computer (e.g., word processor, spreadsheet, or database software), can now be stored directly on magnetic or optical media, further increasing the opportunities for subsequent access and retrieval of them. [0005] The growing volume of publicly available, machine-readable textual information makes it increasingly necessary for businesses to automate the handling of such information to stay competitive. Other establishments like educational institutions, medical facilities and government entities can similarly benefit from this automated handling of information. By automating the handling of text, these organizations can decrease costs and increase the quality of the services performed that require access to textual information. [0006] One approach for automating the conversion of natural language documents into machine-readable text is to use a text classification system which, given a portion of data, can automatically generate several categories describing major subject matter contained in the data. Automated text classification systems identify the subject matter of a piece of text as belonging to one or more categories of a potentially large, predefined set of categories. Text classification also includes a class of applications that can solve a variety of problems in the indexing and routing of text. Efficient routing of text is particularly useful in large organizations where there is a large volume of individual pieces of text that needs to be sent to specific persons (e.g., technical support specialists inside a large customer support center). Text routing also plays a pivotal role in the area of text retrieval in response to user queries on the Internet. [0007] A number of different approaches have been developed for automatic text processing of user queries. One approach is based upon information retrieval techniques utilizing Boolean keyword searches. While this approach is efficient, it suffers from problems relating to the inaccuracy of the retrieved information. A second approach borrows natural language processing using deep linguistic knowledge from artificial intelligence technology to achieve higher accuracy. While deep linguistic processing improves accuracy based upon an analysis of the meaning of input text, speed of execution is slow and range of coverage is limited. This is especially problematic when such techniques are applied to large volumes of text [0008] Another approach is rule-based text classification systems which classify documents according to rules written by people about the relationship between words in the documents and the classification categories. Text classification systems which rely upon rule-base techniques also suffer from a number of drawbacks. The most significant drawback is that such systems require a significant amount of knowledge engineering to develop a working system appropriate for a desired text classification application. It becomes more difficult to develop an application using rule-based systems because individual rules are time-consuming to prepare, and require complex interactions. A knowledge engineer must spend a large amount of time tuning and experimenting with the rules to arrive at the correct set of rules to ensure that the rules work together properly for the desired application. [0009] Another approach to text classification is to use statistical techniques to enable the system to "learn" from the input text. In essence, these systems develop a statistical model of the vocabulary used in the different classification categories. Such systems take training data in the form of documents classified by people into appropriate categories, and in a training phase, develop the statistical model from these documents. These statistical models quantify the relationships between vocabulary features (words and phrases) and classification categories. Once developed, these statistical models may be used to classify new documents. In systems that do utilize a learning component (a training phase), the narrower and more closely related the categories are, the more training data is needed. Exacerbating the problem is the fact that in most applications, training data is hard to locate, often does not provide adequate coverage of the categories, and is difficult and time-consuming for people to categorize, requiring manual effort by experts in the subject area (who are usually scarce and expensive resources). Further, badly categorized training data or correctly categorized training data with extraneous or unusual vocabulary degrades the statistical model, causing the resulting classifier to perform poorly. [0010] Of the prior art systems that utilize training data, most do not have the capability to interactively take advantage of human knowledge. History has shown that a person will often know what results from sound training data, what results from poor training data, and what may not be adequately expressed in the training data. Those prior art systems that do utilize user input, do not allow users to directly affect the quantified relationship between vocabulary features and classification categories, but simply allow the user to change the training data. Yet another shortcoming of prior art text classification systems lies in the fact that they only deal with categories which are from a single perspective. Consider three perspectives on news stories: geography (where the story took place), business entities (what companies the story is about) and topic. To categorize stories according to geographic location would require a different classifier than one that classified the stories according to business entities, which in turn would require a different classifier than one that classified stories according to their topic. These classifiers cannot interact, consequently one cannot benefit from the other. For example, a false correlation between text about Germany and the category "pharmaceutical companies" may arise in the business entity classifier because many pharmaceutical companies are located in Germany. The fact that text about Germany is known to be an important feature of geographical classification (by the geography classifier) cannot be used to ameliorate the false correlation in the business entity perspective. [0011] Thus, there is a need to overcome these and other problems of the prior art and to provide an effective method for classifying text in which user knowledge may be utilized very early in the construction of the statistical model. The present invention, as illustrated in the following description, is directed to solving one or more of the problems set forth above. SUMMARY OF THE INVENTION [0012] In accordance with the present invention, a method for simultaneously classifying at least one document into a plurality of categories is disclosed. The method comprises the steps of associating a plurality of category features with each category, wherein each category feature represents one of a plurality of tokens; producing a category vector for each category, wherein each category vector includes the plurality of category features with a weight corresponding to each category feature, wherein the weight is indicative of a degree of association between the category feature and the category; associating a plurality of document features with each document, wherein each document feature represents one of a plurality of tokens found in the document; producing a feature vector for each document, wherein each feature vector includes the plurality of document features with a count corresponding to each document feature, wherein the count is indicative of the number of times the document feature appears in the document; multiplying the category vector by the document vector, in accordance with the mathematical convention of multiplication of a vector by a vector, to produce a plurality of category scores for each document; and for each perspective, classifying a document into a category provided the category score exceeds a predetermined threshold. [0013] In accordance with another embodiment of the present invention, a method for associating one of a plurality of features with at least one of a plurality of categories is disclosed. The method comprises the steps of manually or automatically associating a plurality of features with at least one category, wherein the plurality of features contribute to a decision to classify a document into said at least one category. BRIEF DESCRIPTION OF THE DRAWINGS [0014] FIG. 1, is shown a block diagram of the general layout of the present invention; [0015] FIG. 2 is a flow diagram depicting the training phase of the present invention; [0016] FIG. 3 is an example of a term list in accordance with one embodiment of the present invention; [0017] FIG. 4 is an example of a symbol table in accordance with one embodiment of the present invention; [0018] FIG. 5 is an example of a symbol table and a feature table in accordance with the present invention; [0019] FIG. 6 is an example of a Recorded Evidence Edits Table in accordance with one embodiment of the present invention; Continue reading... Full patent description for System and method for automatically classifying text Brief Patent Description - Full Patent Description - Patent Application Claims Click on the above for other options relating to this System and method for automatically classifying text patent application. ###  How KEYWORD MONITOR works... a FREE service from FreshPatents1. Sign up (takes 30 seconds). 2. Fill in the keywords to be monitored. 3. Each week you receive an email with patent applications related to your keywords. Start now! - Receive info on patent apps like System and method for automatically classifying text or other areas of interest. ### Previous Patent Application: Method, apparatus, and system for implementing vehicle identification Next Patent Application: System and method for measuring image similarity based on semantic meaning Industry Class: Data processing: database and file management or data structures ### FreshPatents.com Support Thank you for viewing the System and method for automatically classifying text patent info. IP-related news and info Results in 1.03992 seconds Other interesting Feshpatents.com categories: Novartis , Pfizer , Philips , Polaroid , Procter & Gamble , |
||
