Intelligent document system
An intelligent document system provides one or more users, who are recipients of documents, with a way to “interact” with the documents for example for the purpose of understanding the content of the documents, and taking appropriate action in response to receiving the documents. Interacting with one or more documents can include navigating the documents guided by semantic content of the documents, asking questions that are answered based on the content of the documents. In some examples, the documents are “dynamic” in that users can manipulate data in the document for example for multiple different views or analyses. In some examples, the documents are augmented semantics and ontology that will allow the user to accurately navigate the document and achieve the natural interfacing they desire.
1 . A method comprising:
processing a plurality of received documents to generate a corresponding plurality of augmented documents, each augmented document of the plurality of augmented documents including a content of a received document of the plurality of received documents and metadata including one or more questions and corresponding answers in the content of the received document,
wherein at least some of one or more questions are automatically generated natural language questions determined from the received document with respective answer locations in the content of the received document; and
providing an interface configured to receive a natural language question about the plurality of received documents from a user as input and to provide an answer to the natural language question to the user, wherein the answer is based at least in part on the metadata associated with one or more augmented documents of the plurality of augmented documents.
2 . The method of claim 1 , wherein providing the answer to the natural language question includes identifying a question from the one or more questions that is semantically similar to the natural language question.
3 . The method of claim 2 , wherein providing the answer to the natural language question includes providing a part of the content at the location associated with the identified question to the user.
4 . The method of claim 1 , wherein at least some received documents of the plurality of received documents are structured documents.
5 . The method of claim 4 , wherein the structured documents include section identifiers associated with different sections of the documents.
6 . The method of claim 1 , wherein at least some received documents of the plurality of received documents are unstructured documents.
7 . The method of claim 1 , wherein the interface is configured, for each section of a plurality of sections of the received documents, to generate an indication of whether the section includes an answer to the natural language question.
8 . The method of claim 7 , wherein the indication includes a location in the section of the answer to the natural language question.
9 . The method of claim 1 , wherein:
processing the plurality of received documents comprises processing said documents using a natural language processing service to perform semantic procedures for said augmented document including the at least some questions and corresponding answers in the content of said documents, said semantic procedures being a part of the metadata for said received document; and
wherein the answers comprise respective pointers to parts of the content of received documents.
10 . The method of claim 1 , wherein the determining of the one or more questions from the one or more documents is performed during ingestion of the one or more documents and prior to receiving of a user's natural language question.
11 . The method of claim 10 , wherein the determining of the one or more questions comprises determining multiple questions from the documents, each question being associated with an answer in the documents.
12 . The method of claim 11 , wherein the answer to the received natural language question is based at least in part on a similarity of a question of the one or more questions to the received natural language question.
13 . The method of claim 1 , wherein the respective answer location comprises one or more pointers into the content of the received document.
14 . The method of claim 1 , wherein the respective answer location comprises a document part of the received document.
15 . The method of claim 1 , wherein processing the plurality of received documents comprises including, in the metadata of at least one augmented document, one or more natural language questions provided in the received document.
16 . The method of claim 1 , wherein providing the answer to the natural language question comprises automatically comparing text of the natural language question received from the user with the one or more natural language questions provided in the received document, determining that the natural language question received from the user is similar to one of the natural language questions provided in the received document, and, in response to determining similarity, providing the user with information in the received document that corresponds to the similar natural language question.
17 . The method of claim 1 , wherein processing the plurality of received documents comprises processing structured data of a table into a text representation and generating at least some of the natural language questions based on the text representation.
18 . The method of claim 1 , wherein processing the plurality of received documents comprises performing named entity recognition to detect named entities within the content of the received documents and generating at least some of the natural language questions such that a corresponding answer location comprises a location in the content of a detected named entity.
19 . The method of claim 1 , wherein generating at least some of the natural language questions comprises processing text of the received documents using a transformer natural language processing model.
20 . A system comprising:
one or more processors for processing a plurality of received documents to generate a corresponding plurality of augmented documents, each augmented document of the plurality of augmented documents including a content of a received document of the plurality of received documents and metadata including one or more questions and corresponding answers in the content of the received document,
wherein at least some of one or more questions are automatically generated natural language questions determined from the received document with respective answer locations in the content of the received document; and
an interface configured to receive a natural language question about the plurality of received documents from a user as input and to provide an answer to the natural language question to the user, wherein the answer is based at least in part on the metadata associated with one or more augmented documents of the plurality of augmented documents.
21 . A non-transitory, computer-readable medium comprising software tangibly embodied thereon, the software including instructions for causing one or more processors to:
process a plurality of received documents to generate a corresponding plurality of augmented documents, each augmented document of the plurality of augmented documents including a content of a received document of the plurality of received documents and metadata including one or more questions and corresponding answers in the content of the received document,
wherein at least some of one or more questions are automatically generated natural language questions determined from the received document with respective answer locations in the content of the received document; and
provide an interface configured to receive a natural language question about the plurality of received documents from a user as input and to provide an answer to the natural language question to the user, wherein the answer is based at least in part on the metadata associated with one or more augmented documents of the plurality of augmented documents.