AI::Categorizer::Learner::NaiveBayes(3pm)

1AI::Categorizer::LearneUrs:e:rNaCiovnetBraiybeust(e3d)PAeIr:l:CDaotceugmoernitzaetri:o:nLearner::NaiveBayes(3)
2
3
4

NAME

6       AI::Categorizer::Learner::NaiveBayes - Naive Bayes Algorithm For
7       AI::Categorizer
8

SYNOPSIS

10         use AI::Categorizer::Learner::NaiveBayes;
11
12         # Here $k is an AI::Categorizer::KnowledgeSet object
13
14         my $nb = new AI::Categorizer::Learner::NaiveBayes(...parameters...);
15         $nb->train(knowledge_set => $k);
16         $nb->save_state('filename');
17
18         ... time passes ...
19
20         $nb = AI::Categorizer::Learner::NaiveBayes->restore_state('filename');
21         my $c = new AI::Categorizer::Collection::Files( path => ... );
22         while (my $document = $c->next) {
23           my $hypothesis = $nb->categorize($document);
24           print "Best assigned category: ", $hypothesis->best_category, "\n";
25           print "All assigned categories: ", join(', ', $hypothesis->categories), "\n";
26         }
27

DESCRIPTION

29       This is an implementation of the Naive Bayes decision-making algorithm,
30       applied to the task of document categorization (as defined by the
31       AI::Categorizer module).  See AI::Categorizer for a complete
32       description of the interface.
33
34       This module is now a wrapper around the stand-alone
35       "Algorithm::NaiveBayes" module.  I moved the discussion of Bayes'
36       Theorem into that module's documentation.
37

METHODS

39       This class inherits from the "AI::Categorizer::Learner" class, so all
40       of its methods are available unless explicitly mentioned here.
41
42   new()
43       Creates a new Naive Bayes Learner and returns it.  In addition to the
44       parameters accepted by the "AI::Categorizer::Learner" class, the Naive
45       Bayes subclass accepts the following parameters:
46
47       •   threshold
48
49           Sets the score threshold for category membership.  The default is
50           currently 0.3.  Set the threshold lower to assign more categories
51           per document, set it higher to assign fewer.  This can be an
52           effective way to trade of between precision and recall.
53
54   threshold()
55       Returns the current threshold value.  With an optional numeric
56       argument, you may set the threshold.
57
58   train(knowledge_set => $k)
59       Trains the categorizer.  This prepares it for later use in categorizing
60       documents.  The "knowledge_set" parameter must provide an object of the
61       class "AI::Categorizer::KnowledgeSet" (or a subclass thereof),
62       populated with lots of documents and categories.  See
63       AI::Categorizer::KnowledgeSet for the details of how to create such an
64       object.
65
66   categorize($document)
67       Returns an "AI::Categorizer::Hypothesis" object representing the
68       categorizer's "best guess" about which categories the given document
69       should be assigned to.  See AI::Categorizer::Hypothesis for more
70       details on how to use this object.
71
72   save_state($path)
73       Saves the categorizer for later use.  This method is inherited from
74       "AI::Categorizer::Storable".
75

CALCULATIONS

77       The various probabilities used in the above calculations are found
78       directly from the training documents.  For instance, if there are 5000
79       total tokens (words) in the "sports" training documents and 200 of them
80       are the word "curling", then P(curling|sports) = 200/5000 = 0.04 .  If
81       there are 10,000 total tokens in the training corpus and 5,000 of them
82       are in documents belonging to the category "sports", then P(sports) =
83       5,000/10,000 = 0.5> .
84
85       Because the probabilities involved are often very small and we multiply
86       many of them together, the result is often a tiny tiny number.  This
87       could pose problems of floating-point underflow, so instead of working
88       with the actual probabilities we work with the logarithms of the
89       probabilities.  This also speeds up various calculations in the
90       categorize() method.
91

TO DO

93       More work on the confidence scores - right now the winning category
94       tends to dominate the scores overwhelmingly, when the scores should
95       probably be more evenly distributed.
96

AUTHOR

98       Ken Williams, ken@forum.swarthmore.edu
99

COPYRIGHT

101       Copyright 2000-2003 Ken Williams.  All rights reserved.
102
103       This library is free software; you can redistribute it and/or modify it
104       under the same terms as Perl itself.
105