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.

1 - 12 . (canceled)  
   
   
       13 . A method for associating at least one of a plurality of features with at least one of a plurality of categories, said method comprising at least one of manually or automatically associating at least one of said plurality of features to at least a first category, said plurality of features contributing to a decision to classify a document into said at least first category.  
   
   
       14 . The method of  claim 13 , further including classifying at least one document into said at least one category, provided the document includes a predetermined number of said plurality of features associated with said category.  
   
   
       15 . The method of  claim 13 , further comprising at least one of manually or automatically associating at least one of a plurality of attributes with at least one of said plurality of features, said plurality of attributes contributing to a decision to classify a document into said at least one category.  
   
   
       16 . The method of  claim 15 , further comprising: 
 determining whether said at least one feature was manually associated to said at least first category; and    associating an attribute with said at least one feature that indicates that the feature was Edited.    
   
   
       17 - 18 . (canceled)  
   
   
       19 . The method of  claim 15 , further comprising classifying a document into a category, provided the document does not contain a feature whose association with said category has a RejectConcept attribute  
   
   
       20 . The method of  claim 15 , further comprising classifying a document that contains a feature or a morphological variant of the feature into a category, provided the feature contains an attribute associated with the category that declares the feature to be morphologically variable.  
   
   
       21 . The method of  claim 15 , further comprising classifying a document that contains a feature into a category, provided the feature contains an attribute associated with the category that declares the feature to be morphologically invariant.  
   
   
       22 . The method of  claim 15 , further comprising the step of not classifying a document that contains a morphological variant of a feature into a category, provided the feature contains an attribute associated with the category that declares the feature to be morphologically invariant.  
   
   
       23 . The method of  claim 15 , further comprising classifying a document that contains a feature or a case variant of the feature into a category, provided the feature contains an attribute associated with the category that declares the feature to be case insensitive a feature associated with a category which association has an attribute declaring the feature to be case insensitive contributes to a decision to classify a document into said category, provided the document contains the feature or a case variant of the feature.  
   
   
       24 . The method of  claim 15 , further comprising classifying a document that contains a feature into a category, provided the feature contains an attribute associated with the category that declares the feature to be case insensitive.  
   
   
       25 . The method of  claim 15 , further comprising the step of not classifying a document that contains a case variant of a feature into a category, provided the feature contains an attribute associated with the category that declares the feature to be case invariant  
   
   
       26 . The method of  claim 15 , further comprising classifying at least one document into at least one of said categories, provided the document contains a feature whose association with said at least one category has an attribute entitled DirectHit.  
   
   
       27 . The method of  claim 15 , further comprising classifying at least one document into at least one of said categones, provided the document does not contain a feature whose association with said at least one category does not contain the attribute entitled Overlap.  
   
   
       28 . The method of  claim 15 , further comprising classifying a document containing an overlapping feature into a category, provided the feature contains an attribute associated with the category that declares the feature to be overlap insensitive.  
   
   
       29 . The method of  claim 15 , further comprising classifying a document containing a non-overlapping feature into a category, provided the feature contains an attribute associated with the category that declares the feature to be overlap sensitive.  
   
   
       30 . The method of  claim 15 , further comprising the step of not classifying a document that contains an overlapping feature into a category, provided the feature contains an attribute associated with the category that declares the feature to be overlap sensitive.  
   
   
       31 . The method of  claim 15 , further comprising at least one of manually or automatically assigning a weight to the feature, said weight indicative of a degree of association between said document and said category.  
   
   
       32 . The method of  claim 31 , further comprising: 
 determining whether said weight was manually assigned to said feature; and    associating an attribute with said feature that indicates that the weight was WeightEdited.    
   
   
       33 . The method of  claim 32 , further comprising at least one of manually or automatically replacing a value for said weight with another value, provided the feature does not contains an attribute associated with the category that declares the feature to be WeightEdited.  
   
   
       34 . The method of  claim 33 , further comprising manually replacing a value for said weight with another value, provided the feature contains an attribute associated with the category that declares the feature to be WeightEdited.  
   
   
       35 - 37 . (canceled)  
   
   
       38 . The method of  claim 15 , further comprising the steps of: 
 determining whether at least one of said plurality of features is a stop word; and    setting an attribute indicating that said feature is a stop word.    
   
   
       39 . The method of  claim 15 , further comprising the steps of: 
 at least one of manually or automatically determining a scope of at least one of said plurality of features; and    setting an attribute indicating that said at least one feature is for queries only, or for documents only, or for both.    
   
   
       40 . The method of  claim 15 , further comprising the step of setting an attribute indicating that said feature is FilteredOut, provided said feature has been manually or automatically filtered out of a classification.  
   
   
       41 . The method of  claim 31 , further comprising multiplying said weight by a scaling parameter, provided the decision to classify the document into said category was based on at least one feature automatically associated with the category.  
   
   
       42 . The method of  claim 41 , wherein said scaling parameter is between 0 and 1.  
   
   
       43 - 47 . (canceled)  
   
   
       48 . In a system including perspectives and categories, each perspective comprising at least one category representative of that perspective, a method for constructing a classifier to classify at least one item across multiple perspectives, the method including: 
 associating at least one feature with each category, in which each feature is configured for being detected in at least a portion of the at least one item for classification of that item;    determining an initial weight indicating a degree of association between each associated feature and category; and    in which weights for a category are initially related to weights for other categories of the same perspective but are initially substantially unrelated to weights for categories in different perspectives.    
   
   
       49 . The method of  claim 48 , in which the determining the weight indicating a degree of association between each associated feature and category includes using the corresponding feature's distribution in training data items tagged to categories from the same perspective as the category being associated with the corresponding feature.  
   
   
       50 . The method of  claim 48 , in which the determining the weight indicating a degree of association between each associated feature and category includes receiving a user input specifying the weight.  
   
   
       51 . The method of  claim 50 , further including deeming a feature to be unassociated with a category if no user input is received specifying the weight corresponding to the feature.  
   
   
       52 . The method of  claim 48 , further including deeming a feature to be unassociated with a category if a magnitude of a corresponding weight between the feature and the category does not exceed a predetermined threshold value.  
   
   
       53 . The method of  claim 52 , further including specifying the predetermined threshold value, for each perspective, independent of the predetermined threshold value for other perspectives.  
   
   
       54 . The method of  claim 48 , further including deeming a feature to be unassociated with a category if a number of features associated with the category exceeds a predetermined threshold value.  
   
   
       55 . The method of  claim 48 , further including limiting how many categories an item can be classified into within a particular perspective.  
   
   
       56 . In a system comprising perspectives and categories, each perspective comprising at least one category representative of that perspective, the system also comprising weights, each weight indicating a degree of an association between a feature and a category, a method for classifying at least one item across multiple perspectives, the method comprising: 
 identifying feature instances in the items;    representing, for each item, which features were identified in that item and the number of instances each such feature was identified in that item;    computing, for each item, a category score for each category associated with at least one feature identified in that item, the computing using the weight associating the category and the at least one feature identified in that item;    selecting one or more categories to represent each perspective according to the category scores; and    classifying the items across the selected categories representing the multiple perspectives.    
   
   
       57 . The method of  claim 56 , in which the selecting includes comparing, for each category, the category score to a predetermined threshold value of the perspective represented by the category, including filtering out those categories with a category score below the predetermined threshold value, and keeping those remaining categories with a category score that equals or exceeds the predetermined threshold value.  
   
   
       58 . The method of  claim 56 , further including, for each perspective, limiting the number of categories for that perspective to be less than or equal to a category count limit for that perspective, and in which the selecting one or more categories to represent each perspective includes eliminating categories in excess of the category count limit for that perspective based on their relatively lower category scores.  
   
   
       59 . The method of  claim 56 , in which the items include representations of documents or queries.  
   
   
       60 . The method of  claim 56 , in which the identifying features includes identifying vocabulary relevant to at least one category.  
   
   
       61 . The method of  claim 56 , in which the identifying feature instances in the items includes identifying features instances in documents, and further including using an attribute designating at least one region of the document to which the identifying feature instances is limited.  
   
   
       62 . The method of  claim 56 , further including receiving user input for determining at least one weight.  
   
   
       63 . The method of  claim 56 , further including statistically determining at least one weight using training data.  
   
   
       64 . The method of claim  claim 56 , further including providing, for each feature, at least one weight associating that feature with a corresponding category, and basing the at least one weight on at least one of: 
 automated processing of training data;    user-input data; and    a combination of automated processing of training data and user-input data.    
   
   
       65 . The method of  claim 64 , further including specifying, for each perspective, a degree for combining automated processing of training data and human-input data for weights associated with categories representative of that perspective.  
   
   
       66 . The method of  claim 56 , further including modifying weights initially associating features with one or more categories representing a first perspective based on other weights associating features with one or more other categories representing one or more perspectives different from the first perspective.  
   
   
       67 . The method of  claim 66 , in which the modifying weights includes, if a feature's initial weights indicates that the feature is strongly correlated with at least one category in a first perspective and weakly correlated to the categories of different perspectives, then doing at least one of: 
 reducing the feature's weights to the categories of the different perspectives; and    increasing the feature's weight to the category of the first perspective.    
   
   
       68 . The method of  claim 56 , further including incorporating a dependence between an item's category score for a category representing a first perspective and the item's category score for one or more other categories representing one or more perspectives different from the first perspective.  
   
   
       69 . The method of  claim 58 , in which the incorporating the dependence includes, if the item 's category score for a category representing a first perspective equals or exceeds a threshold value, then inhibiting classification of the item to one or more other categories representing one or more perspectives different from the first perspective.  
   
   
       70 . The method of  claim 68 , in which the incorporating the dependence includes, if the item's category score for a category representing a first perspective equals or exceeds a threshold value, then reducing the item's category score for one or more other categories representing one or more perspectives different from the first perspective.  
   
   
       71 . The method of  claim 56 , in which the items are documents, and in which the classifying the items includes limiting the regions of the document that are used for classifying the documents to at least one category in a second perspective based on a characteristic of a classification of the document to at least one category in the first perspective.  
   
   
       72 . In a system for classifying items to categories, a method including: 
 receiving user-input defining all associations between classification features and categories; and    statistically determining weights corresponding to the user-defined associations, each weight indicating a degree to which the association's feature identifies the association's category and discriminates against other categories.    
   
   
       73 . The method of  claim 72 , further including identifying candidate features in documents or queries, in which the candidate features include words or phrases in the documents or queries.  
   
   
       74 . The method of  claim 72 , further including providing attributes for the features.  
   
   
       75 . The method of  claim 74 , in which the providing attributes includes providing at least one of: 
 an Exact Match attribute to indicate whether a match to the feature requires both matching case and matching a stemming form;    a Case attribute to indicate whether a match to the feature requires matching case;    a Stemming attribute to indicate whether a match to the feature requires matching a stemming form; and    an EmbeddedTermsAllowed attribute to indicate whether a match to a first feature precludes a match to one or more other features embedded within the first feature.    
   
   
       76 . The method of  claim 72 , further including providing attributes for the associations of features to categories.  
   
   
       77 . The method of  claim 76 , in which providing attributes includes providing at least one of: 
 an Edited attribute indicating whether the association has been specified in a Recorded Evidence Edits table;    a WeightEdited attribute indicating whether a weight of the association was specified or edited by a human user;    a Stop attribute indicating whether the feature is a stop word;    a Scope of Feature attribute indicating whether the association of the feature to the category applies to topic spotting of queries only, topic spotting of documents only, or topic spotting of both queries and documents;    a FilteredOut attribute indicating whether the feature should be disregarded during topic spotting;    a DirectHit attribute that, if asserted, indicates that a document or query including the feature should be tagged to the category specified in the association between feature and category bearing the DirectHit attribute, regardless of what other features are included in the document or query;    a RejectConcept attribute that, if asserted, indicates that a document or query including the feature should not be tagged to the category specified in the association between the feature and category bearing the RejectConcept attribute, regardless of what other features are included in the document or query;    a Case attribute that indicates whether a to indicate whether a match in to the feature, in a document or query, requires matching case of the feature;    a Stemming attribute to indicate whether a match to the feature, in a document or query, requires matching a stemming form; and    an EmbeddedTermsAllowed attribute to indicate whether a match to a first feature, in a document or query, precludes a match to one or more other features embedded within the first feature, in the document or query.    
   
   
       78 . The method of  claim 72 , further comprising receiving user input for overriding at least one of the computed weights indicating a strength of the association of a feature to a category.  
   
   
       79 . The method of  claim 78 , in which the overriding at least one of the computed weights includes increasing the weight's strength of association of a feature and a category.  
   
   
       80 . The method of  claim 72 , in which the statistically determining weights is based at least in part on how often the associations' features are present in a set of training items.  
   
   
       81 . The method of  claim 72 , in which the determining the weight indicating a degree of association between each associated feature and category includes receiving a user input specifying the weight.  
   
   
       82 . The method of  claim 81 , further including deeming a feature to be unassociated with a category if no user input is received specifying the weight corresponding to the feature.  
   
   
       83 . The method of  claim 72 , further including receiving user input for determining at least one weight.  
   
   
       84 . The method of  claim 72 , further including statistically determining at least one weight using training data.  
   
   
       85 . The method of  claim 84 , in which the training data consists essentially of at least one document associated with a category.  
   
   
       86 . The method of  claim 84 , in which the training data consists essentially of at least one user-specified association between a feature and a category.  
   
   
       87 . The method of claim  claim 72 , further including providing, for each feature, at least one weight associating that feature with a corresponding category, and basing the at least one weight on at least one of: 
 automated processing of training data;    user-input data; and    a combination of automated processing of training data and user-input data.    
   
   
       88 . The method of  claim 87 , further including specifying, for each perspective, a degree for combining automated processing of training data and human-input data for weights associated with categories representative of that perspective.  
   
   
       89 . In a system for classifying items to categories, a method including: 
 receiving user-input creating user-defined associations between classification features and categories;    statistically determining machine-defined associations that are capable of being different from the user-defined associations; and    classifying items to the categories using weights corresponding to the user-defined associations and the machine-defined associations.    
   
   
       90 . The method of  claim 89 , further including controlling relative contributions of the weights corresponding to the user-defined associations and the weights corresponding to the machine-defined associations.  
   
   
       91 . The method of  claim 90 , in which the controlling includes obtaining a generally greater relative contribution of the user-defined associations with respect to the machine-defined associations.  
   
   
       92 . The method of  claim 89 , in which the classifying includes classifying the items to categories spanning multiple perspectives.