LLM-based classification

openapi_classification.yml

@@ -27,21 +27,19 @@ paths:
       summary: Classify document by type
 
       description: |
-        Score a document's similarity to each document type you defined in your Sensible account and to each reference document in the highest-scoring type.
-        To retrieve the scores, poll the `download_link` in this endpoint's response until it returns a non-error response.
-        This endpoint is asynchronous. For more information about scores, expand the 200 response in the synchronous [classification](ref:classify-document-sync) endpoint.
+
+        Classify a document into one of the document types you defined in your Sensible account. For more information, see [Classifying documents by type](doc:classify).
 
         Use this endpoint:
 
          - In an extraction workflow. For example, determine which documents to extract prior to calling a Sensible extraction endpoint.
-         - Outside an extraction workflow. For example, to determine where to route each document or to label each document in a system of record.
+         - Outside an extraction workflow. For example, determine where to route each document or to label each document in a system of record.
 
         To post the document bytes, specify the non-encoded document bytes as the entire request body,and specify the `Content-Type` header, for example,"application/pdf" or "image/jpeg".
-
-        For supported file sizes, see [Supported file types](doc:file-types).
-
         
+        For supported file size and types, see [Supported file types](doc:file-types).
 
+      
       requestBody:
         $ref: '#/components/requestBodies/SupportedFileTypes'
 
@@ -77,7 +75,7 @@ paths:
 
         **Note:** Use this Classify endpoint for testing. Use the asynchronous Classify endpoint for production.
 
-        Score a document's similarity to each document type you defined in your Sensible account. Get scores for the document's similarity to document types and to their reference documents.
+        Classify a document into one of the document types you defined in your Sensible account. For more information, see [Classifying documents by type](doc:classify).
 
         Use this endpoint:
 
@@ -99,7 +97,7 @@ paths:
               schema:
                 $ref: '#/components/schemas/ClassifySingleResponse'
           description: |
-            The document type and reference documents in the Sensible account that are most similar to this document.
+            The document type in your Sensible account that's most similar to this document.
         '401':
           $ref: '#/components/responses/401'
         '400':
@@ -111,12 +109,6 @@ paths:
         '500':
           $ref: '#/components/responses/500'
 
-
-
-
-
-
-
 components:
 
   responses:
@@ -172,8 +164,6 @@ components:
             example: Sensible encountered an unknown error
   #parameters:
 
-
-
   securitySchemes:
     bearerAuth:       # arbitrary name for the security scheme
       type: http
@@ -243,7 +233,8 @@ components:
           type: string
           description: File format of the document for which you requested classification.
         download_link:
-          description: Poll until the download URL returns a non-error response. Links to a JSON download that contains the same response as from the synchronous Classify endpoint request.
+          description: |
+            Poll until the download URL returns a non-error response. Links to a JSON download that contains the same response as from the synchronous Classify endpoint request.
           type: string
           format: url
       example:
@@ -256,7 +247,7 @@ components:
       properties:
         document_type:
           type: object
-          description: Document type defined in the Sensible account that this doc is most similar to. To use a document type for classification, Sensible requires that the type contains at least one reference document.
+          description: The document type defined in your Sensible account that this document is most similar to.
           properties:
             id:
               type: string
@@ -266,64 +257,43 @@ components:
               description: User-friendly name for the document type.
             score:
               type: number
-              description: Similarity score comparing the document to the document type, between 0 and 1.
+              description: Deprecated. Similarity score comparing the document to the document type, where a score of 1 indicates a match.
         reference_documents:
           type: array
-          description: Reference documents uploaded to the Sensible account that this document is most similar to.
+          description: Deprecated. Scoring for embeddings-based classification, replaced by LLM-based classification. 
           items:
             type: object
             properties:
               id:
                 type: string
-                description: Unique ID for the reference document.
+                description: Deprecated. Unique ID for the reference document.
               name:
                 type: string
-                description: User-friendly name for the reference document.
+                description: Deprecated. User-friendly name for the reference document.
               score:
                 type: number
-                description: Similarity score comparing the document to the reference document, between 0 and 1.
+                description: Deprecated. Similarity score comparing the document to the reference document, between 0 and 1.
 
         classification_summary:
           type: array
-          description: Scores for this document's similarity to each document type in the Sensible account, excluding document types Sensible created in your account as tutorials, such as `senseml_basics`.
+          description: Deprecated. Scoring for embeddings-based classification, replaced by LLM-based classification.
           items:
             type: object
             properties:
               id:
                 type: string
-                description: Unique ID for the document type.
+                description: Deprecated. Unique ID for the document type.
               name:
                 type: string
-                description: User-friendly name for the document type.
+                description: Deprecated. User-friendly name for the document type.
               score:
                 type: number
-                description: Similarity score comparing the document to the document type, between 0 and 1.
+                description: Deprecated. Similarity score comparing the document to the document type, between 0 and 1.
 
       example:
         document_type:
           id: 77c2ab88-3389-4ea8-93c7-912c2bfd373a
           name: 1040s
-          score: 0.9637581544082083
-        reference_documents:
-        - id: b4fbc822-de99-4916-b43a-2902131f2619
-          name: 1040_2020_sample
-          score: 0.999649884175599
-        - id: 23680cc8-7855-4698-b51f-6a054704fd1e
-          name: 1040_2019_sample
-          score: 0.983879165384638
-        - id: 58aef918-7017-4576-ad5c-f987b98b4ae7
-          name: 1040_2021_sample
-          score: 0.9670293766923486
-        - id: 161b27ab-5218-4650-a919-65df03de3454
-          name: senior_1040_2021_sample
-          score: 0.939401195335292
-        - id: fb9ed1c3-0545-4f79-bd00-565838bd96a4
-          name: 1040_2018_sample
-          score: 0.9288311504531641
-        classification_summary:
-        - id: 28eee728-e51b-471c-ba92-827c995476f6
-          name: home_policy_declaration_pages
-          score: 0.7760597095611765
-        - id: 16b06941-9486-475a-a6bf-120cf433f6f3
-          name: bank_statements
-          score: 0.7639481987557378
+          score: 1
+        reference_documents: []
+        classification_summary: []

readme-sync/v0/document-type-classification/1000 - classify.md

@@ -11,18 +11,24 @@ Sensible supports two levels of document classification:
 
 This topic covers classifying a document by its type.
 
-For example, if you define a [bank statements](https://github.com/sensible-hq/sensible-configuration-library/tree/main/templates/Financial%20Services/Bank%20Statements) type and a [1040s](https://github.com/sensible-hq/sensible-configuration-library/tree/main/templates/Tax%20Forms/1040s) type in your account, you can classify 1040 forms, 1099 forms, Bank of America statements, Chase statements, and other documents, into those two types. In this scenario, for a  `2023-1-1_bankofamerica_statement_jon_doe.pdf` document, Sensible: 
+Sensible classifies a document by comparing it to the types you define in your account. For example, you can classify 1040 forms and bank statements if you define the following types in your account:
 
-- Classifies this document into the `bank_statements` document type.
-- Classifies the statement doc by its similarity to reference documents in the `bank_statements` document type. The highest score is for [a Bank of America sample statement](https://github.com/sensible-hq/sensible-configuration-library/blob/main/templates/Financial%20Services/Bank%20Statements/refdocs/bank_of_america_sample.pdf).
-- Provides metadata for the classification, including similarity scores for this document compared to each document type in your Sensible account and to each reference document in the `bank_statements` type.
+- a [bank statements](https://github.com/sensible-hq/sensible-configuration-library/tree/main/templates/Financial%20Services/Bank%20Statements) type
+
+- a [1040s](https://github.com/sensible-hq/sensible-configuration-library/tree/main/templates/Tax%20Forms/1040s) type
+
+Sensible uses a document type's name and its description for LLM-based classification:
+
+- If Sensible doesn't find an existing document type to which to match your document in your account, it returns an error.
+- Since Sensible doesn't use configs or reference documents for classification, Sensible can classify documents into your document types even if the document type lacks a config or example. For example, if you lack a  `citibank` config or reference document in your `bank_statements` type, Sensible can still classify a  `2023-1-1_citbank_statement_jon_doe.pdf` document as a bank statement. 
+
+ To improve classification results, describe each document type in your account in its **Settings** tab. For examples of descriptions, see [Document type descriptions](doc:descriptions). 
 
 Use document type classification:
 
 - Prior to an extraction workflow. For example, determine which documents to extract prior to calling a Sensible extraction endpoint. 
 
 - Independent from an extraction workflow. For example, determine where to route each document or to label each document in a system of record.
 
-To improve classification results, Sensible recommends that a document type includes a sample set of reference documents that represent the diversity you expect to see in the document type. To use a document type for classification, Sensible requires that the type contains at least one reference document.
-
 To classify documents, use the Sensible API or SDKs.
+

readme-sync/v0/senseml-reference/6800 - document-type-settings/000 - descriptions.md

@@ -1,11 +1,14 @@
 ---
-title: "LLM portfolio description"
+title: "Document type descriptions"
 hidden: false
 ---
 
-Describe the document type to enable segmenting documents' page ranges from a [portfolio](doc:portfolio) file using LLMs.  For example, describe a typical first page of a document type, a typical last page of a document type, and commonly found fields and their values.
+Describe a document type in its **Settings** tab to:
 
-Example of document type descriptions:
+- Enable segmenting documents' page ranges from a [portfolio](doc:portfolio) file using LLMs.  For example, describe a typical first page of a document type, a typical last page of a document type, and commonly found fields and their values.
+- Improve [classifying](doc:classify) a document into an existing document type in your account.
+
+Examples of document type descriptions:
 
 - `To accurately classify this type of document look at the bottom left of each page of the document and if you see ACORD 131 then it is an instance of an Acord 131 form.`
 - `This type of document is a scanned bank check. Usually only a single page.`