|
Tag: Redirect target changed |
| (4 intermediate revisions by 2 users not shown) |
| Line 1: |
Line 1: |
| <blockquote style="font-size:14pt">
| | #REDIRECT[[Rules-Based (Classification Method)]] |
| The ''Rules-Based'' '''''[[Classification Method]]''''' is one of three methods of classifying documents available to Grooper. This approach uses [[Data Extractor]]s to find key words, phrases, or other text-based information in order to identify and classify a document (assigning a '''[[Document Type]]''' to the '''[[Batch Folder|Document Folder]]''').
| |
| </blockquote>
| |
| | |
| The ''Rules-Based'' method classifies documents according to "rules" using the '''''Positive Extractor''''' and '''''Negative Extractor''''' properties of '''Document Type''' objects in a '''[[Content Model]]'''.
| |
| | |
| If an extractor set as the '''''Positive Extractor''''' returns a result on a document, the document would be classified as that '''Document Type'''. The '''''Negative Extractor''''' works the opposite way. If this extractor finds a result on a document, it would be ''prevented'' from being classified as that '''''Document Type'''''. This type of classification can be useful if a document's structure is always predictable or has a fixed title heading or form number and [[OCR]] errors are not an issue.
| |
| | |
| Classification is then performed using the '''[[Classify]]''' activity, using the extraction rules established by the '''''Positive''''' and '''''Negative Extractor''''' properties of '''Document Types''' in a '''Content Model'''.
| |
| | |
| == About ==
| |
| | |
| ''Rules-Based'' classification can be enabled and configured on any '''Content Model''' object. To do so, select the '''''Classification Method''''' property and select ''Rules-Based''.
| |
| | |
| | |
| [[File:Rules-based-about-01.png|center|1000px]]
| |
| | |
| | |
| === What are you classifying? - Document Types ===
| |
| | |
| Classification is all about distinguishing one kind of document from another. The ''Rules-Based'' method uses extractors to do this. '''''Positive Extractors''''' positively identify documents of a certain kind. '''''Negative Extractors''''' prevent a document from being identified as a certain kind of document.
| |
| | |
| This may be obvious, but before you can assign these positive and negative extractor rules, you have to give a ''name'' to that type of document you're wanting to classify. In Grooper, we do this by adding '''Document Type''' objects to a '''Content Model'''.
| |
| | |
| For example, imagine you have a collection of human resources documents. For each employee, you'll have a variety of different kinds of documents in their HR file, such as a federal W-4 form, their employment application, various documents pertaining to their health insurance enrollment, and more. In order to distinguish those documents from one another (in other words, classify them), you will need to add a '''Document Type''' for each kind of document.
| |
| | |
| Take the four kinds of documents seen here: A federal W-4, an employee data sheet, an FSA enrollment form, and a pension enrollment form
| |
| | |
| {|cellpadding=10 cellspacing=5
| |
| |-style="text-align:center"
| |
| |Federal W-4||Employee Data Sheet||FSA Enrollment Form||Pension Enrollment Form
| |
| |-
| |
| |[[File:Lexical-classification-w4.png]]||[[File:Lexical-classification-datasheet.png]]||[[File:Lexical-classification-fsa.png]]||[[File:Lexical-classification-pension.png]]
| |
| |}
| |
| | |
| | |
| If we want to classify a '''[[Batch]]''' of these documents and assign the federal W-4 documents a "Federal W-4" classification and so on, we would need to create a '''Content Model''' and add one '''Document Type''' for each kind of document. So, the W-4s would get a "W-4" '''Document Type'''. The FSAs would get an "FSA" '''Document Type''', and so on.
| |
| | |
| {|cellpadding=10 cellspacing=5
| |
| |style="width:25%"|
| |
| [[File:Lexical-classification-content-model.png]]
| |
| |valign=top|
| |
| A '''Content Model''' is how we determine the ''taxonomy'' of our documents set. Taxonomy is just a fancy word for a classification scheme. Zoological taxonomy organizes organisms into a classification scheme, from domain all the way down to species. We do much the same thing with documents and a '''Content Model'''.
| |
| | |
| The whole set of HR documents belong to the top level in the hierarchy, the '''Content Model''' itself. Each individual kind of document are represented by '''Document Types''', which are next level down in that hierarchy. Each one is distinct from each other, but still part of the '''Content Model's''' scope. Just like insects, spiders, and lobsters are distinct from each other but are all part of the "arthropod" zoological class.
| |
| |}
| |
| | |
| === How are the documents classified? - Positive and Negative Extractors ===
| |
| | |
| {|cellpadding=10 cellspacing=5
| |
| |style="width:40%" valign=top|
| |
| The "rules" in the ''Rules-Based'' method are determined by extraction results set on the '''''Positive''''' and '''''Negative Extractor''''' properties of a '''Document Type'''.
| |
| | |
| # With a '''Document Type''' selected in the Node Tree...
| |
| # These properties will be found in the '''''Classification''''' properties in the property panel. | |
| |
| |
| [[File:Rules-based-about-02.png]] | |
| |}
| |
| | |
| ==== Positive Extractor Rules ====
| |
| | |
| {|cellpadding=10 cellspacing=5
| |
| |style="width:40%" valign=top|
| |
| If the '''''Positive Extractor''''' returns ''at least one'' result on a document, it will be assigned the '''Document Type'''. One common approach to ''Rules-Based'' classification is "title-matching". Often, a document's title will correspond to what '''Document Type''' you want to classify it as.
| |
| | |
| | |
| For example, the document here is titled "DATA INFORMATION SECTION", which is easily matched by the regular expression <code>DATA INFORMATION SECTION</code>.
| |
| |
| |
| [[File:Rules-based-about-03.png]]
| |
| |-
| |
| |valign=top|
| |
| If we create a '''Data Type''' returning this title, we can then assign it to the '''''Positive Extractor''''' property of the "Data Information Sheet" '''Document Type'''.
| |
| | |
| {|cellpadding="10" cellspacing="5"
| |
| |-style="background-color:#36b0a7; color:white"
| |
| |style="font-size:14pt"|'''FYI'''||The '''''Positive Extractor''''' can be set to a variety of extraction options, including ''Reference'', ''Text Pattern'', ''Read Zone'', ''Find Barcode'', ''Read Barcode'', ''Labeled OMR'', ''Ordered OMR'', and ''Zonal OMR''.
| |
| | |
| While using the ''Reference'' option, pointing to a '''Data Type''' in the Node Tree, is the most common configuration, any extractor returning a result will positively classify the document.
| |
| |}
| |
| |
| |
| [[File:Rules-based-about-04.png]]
| |
| |-
| |
| |valign=top|
| |
| When a '''[[Batch]]''' is classified, Grooper will execute each '''Document Type's''' '''''Positive Extractor''''' against the unclassified '''[[Batch Folder|Document Folder]]'''. When a '''''Positive Extractor''''' returns a result on the document, the '''Document Folder''' will be assigned the corresponding '''Document Type''' (Specifically, it will be assigned that '''Document Type''' as its '''''Content Type''''' property).
| |
| | |
| | |
| For example, all the documents labeled "Data Information Sheet" here returned a value matching the "DATA INFORMATION SECTION" title our extractor located. All the documents labeled "Folder" are unclassified documents. Since the "Data Information Sheet" '''Document Type's''' '''''Positive Extractor''''' did ''not'' return a result, they were not assigned the "Data Information Sheet" '''Document Type''' (We also did not assign any '''''Positive Extractors''''' for any of the other '''Document Types'''').
| |
| | |
| | |
| To put it simpler:
| |
| | |
| # If the '''''Positive Extractor''''' returns a result, the '''Document Folder''' is classified as the '''Document Type'''.
| |
| # If the '''''Positive Extractor''''' does not return a result, the '''Document Folder''' remains unclassified.
| |
| |
| |
| [[File:Rules-based-about-05.png]]
| |
| |}
| |
| | |
| ==== Negative Extractor Rules ====
| |
| | |
| {|cellpadding=10 cellspacing=5
| |
| |style="width:40%" valign=top|
| |
| The '''''Negative Extractor''''' property works the opposite way. If the extractor set here produces a result, the '''Document Folder''' will be ''prevented'' from being assigned the '''Document Type'''.
| |
| | |
| For example, let's say we use the following '''''Value Pattern''''' for our '''''Positive Extractor''''' to classify these Federal W-4 documents as a "Federal W-4" '''Document Type''': <code>W-4</code>
| |
| | |
| This certainly produces results for this Federal W-4 form here. It will accurately positively classify this document as a "Federal W-4" '''Document Type'''. However, this is a very general pattern. If the characters "W-4" are found on any other document that ''isn't'' a Federal W-4 form, it will classify it as a "Federal W-4" '''Document Type'''.
| |
| |
| |
| [[File:Rules-based-about-06.png]]
| |
| |-
| |
| |valign=top|
| |
| Upon classification, if the '''''Positive Extractor''''' returns a result on a document it shouldn't, it's going to produce a ''false positive'' result.
| |
| | |
| The document seen here is not a Federal W-4. It's a state W-4 form specific to the state of Iowa. However, since we were so loose with our regular expression pattern, all that had to match to produce a positive result are the characters "W-4".
| |
| | |
| # Those characters are sure enough on this document.
| |
| # So, it gets classified as a "Federal W-4" '''Document Type'''.
| |
| |
| |
| [[File:Rules-based-about-07.png]]
| |
| |-
| |
| |valign=top|
| |
| However, with a '''''Negative Extractor''''', if we can match something on this known document that shouldn't be a "Federal W-4", we can point to that as a rule to ''prevent'' it from being classified as a "Federal W-4" '''Document Type'''.
| |
| | |
| For example, we wouldn't expect to see the web address "www.iowa.gov/tax" on a federal form. But it definitely is on this state W-4 form.
| |
| |
| |
| [[File:Rules-based-about-08.png]]
| |
| |-
| |
| |valign=top|
| |
| If we can create an extractor to match and return that web address (or any other text unique to this Iowa W-4 that distinguishes it from the Federal W-4), we can assign it to the '''''Negative Extractor''''' of the "Federal W-4" '''Document Type'''.
| |
| | |
| Even if the '''''Positive Extractor''''' produces a result, if the '''''Negative Extractor''''' ''also'' produces a result, it will override classification. The document will be ''prevented'' from being classified as a "Federal W-4" '''Document Type'''.
| |
| |
| |
| [[File:Rules-based-about-09.png]]
| |
| |-
| |
| |valign=top|
| |
| If we classify this '''Batch''' with both the '''''Positive''''' and the '''''Negative Extractors''''' configured as described above, we get a different result.
| |
| | |
| # Even though the "Federal W-4" '''Document Type's''' '''''Positive Extractor''''' returns a result here, matching "W-4"
| |
| # The '''''Negative Extractor''''' returns a result here, matching "www.iowa.gov/tax"
| |
| # The '''''Negative Extractor''''' prevents the document from being assigned a "Federal W-4" '''Document Type'''.
| |
| |
| |
| [[File:Rules-based-about-10.png]]
| |
| |}
| |
| | |
| === Mixed Classification: Combining Training-Based and Rules-Based Approaches ===
| |
| | |
| Furthermore, a rules-based approach can be combined with a training-based approach when using the ''[[Lexical]]'' '''''Classification Method'''''. The ''Lexical'' method uses trained examples '''Document Types''' to classify documents. It uses a [[TF-IDF]] algorithm to weight the importance text features (such as words and phrases) based on this trained examples. However, even when choosing ''Lexical'' for the '''''Classification Method''''', the '''''Positive Extractor'''''' and '''''Negative Extractor''''' properties are still present on '''Document Types'''.
| |
| | |
| Generally, the '''''Positive Extractor's''''' result will "win out" over training based classification results, because the '''''Positive Extractor's''''' confidence result (as a percentage value) will be higher than the document's similarity to the trained examples (as a percentage value) for a '''Document Type'''. This way, if you have a value which can be extracted that you know is going to be on a '''Document Type''', you can take advantage of setting a '''''Positive Extractor''''' on the '''Document Type''' to classify them. For example, document titles are often used as "rules". If you can extract text to match a title to a corresponding '''Document Type''', this is often a quick and easy way to classify a document. But, if that extractor fails for whatever reason (because of bad OCR or a new title not matching the extractor's regex), you have training data which can act as a backup classification method.
| |
| | |
| Many of the best classification strategies involve combining the training-based ''Lexical'' method with a rules-based approach.
| |
