Methods, systems, and apparatus, including computer programs encoded on computer storage media, for selecting a text span from an input electronic document that answers an input question. One of the methods includes obtaining a respective first numeric representations of text spans in the input document; for each of the text spans: determining, for a segment that contains the text span, a question-aware segment vector, determining, for the question, a segment-aware question vector, and processing the first numeric representation of the text span, the question-aware segment vector, and the segment-aware question vector using a second feedforward neural network to generate a second numeric representation of the text span; for each unique text span in the plurality of text spans: determining an aggregate representation for the unique text span, and determining, from the aggregate representation, a final score for the unique text span; and selecting a unique text span.

Embodiments of the present disclosure relate to a method and apparatus for determining a text color. The method may include: determining, in response to detecting a text frame on canvas, a subimage corresponding to the text frame from the canvas; acquiring color values of pixels in the subimage, to determine an average color value of the subimage; determining an average luminance value of the subimage based on the average color value of the subimage; and determining a color of text to be inputted into the text frame based on the average luminance value of the subimage.

The present disclosure provides for systems and methods for generating an image of a web resource to detect a modification of the web resource. An exemplary method includes selecting one or more objects of the web resource based on one or more object attributes; identifying a plurality of tokens for each selected object based on contents of the selected object; calculating a hash signature for each selected object of the web resource using the identified plurality of tokens; identifying potentially malicious calls within the identified plurality of tokens; generating an image of the web resource based on the plurality of hash signatures and based on the identified potentially malicious calls, wherein the image of the web resource comprises a vector representation of the contents of the web resource; and detecting whether the web resource is modified based on the image of the web resource.

The present disclosure provides for systems and methods for generating an image of a web resource to detect a modification of the web resource. An exemplary method includes selecting one or more objects of the web resource based on one or more object attributes; identifying a plurality of tokens for each selected object based on contents of the selected object; calculating a hash signature for each selected object of the web resource using the identified plurality of tokens; identifying potentially malicious calls within the identified plurality of tokens; generating an image of the web resource based on the plurality of hash signatures and based on the identified potentially malicious calls, wherein the image of the web resource comprises a vector representation of the contents of the web resource; and detecting whether the web resource is modified based on the image of the web resource.

An image processing device includes: an identifying unit that identifies a plurality of character strings that are candidates for a recording character string among a plurality of character strings acquired by recognizing characters included in a document image; an output unit that outputs a checking screen that represents positions of the plurality of character strings; and a feature quantity extracting unit that extracts a feature quantity of a character string corresponding to a position identified by a user on the checking screen as a feature quantity of the recording character string.

A non-transitory processor-readable medium stores instructions to be executed by a processor. The instructions cause the processor to receive a first trained machine learning model that generates a transcription based on a document. The instructions cause the processor to execute the first trained machine learning model and a second trained machine learning model to generate a refined transcription based on the transcription. The instructions cause the processor to execute a quality assurance program to generate a transcription score based on the document and the transcription. The instructions cause the processor to execute the quality assurance program to generate a refined transcription score based on the refined transcription and at least one of the document or the transcription. The at least one refined transcription score indicates an automation performance better than an automation performance for the at least one transcription score.

A non-transitory processor-readable medium stores instructions to be executed by a processor. The instructions cause the processor to receive a first trained machine learning model that generates a transcription based on a document. The instructions cause the processor to execute the first trained machine learning model and a second trained machine learning model to generate a refined transcription based on the transcription. The instructions cause the processor to execute a quality assurance program to generate a transcription score based on the document and the transcription. The instructions cause the processor to execute the quality assurance program to generate a refined transcription score based on the refined transcription and at least one of the document or the transcription. The at least one refined transcription score indicates an automation performance better than an automation performance for the at least one transcription score.

Provided is an information processing apparatus that extracts a character string to be a candidate of an item value corresponding to a predetermined item, from among multiple character strings included in a document image to be processed. The information processing apparatus first obtains information indicating a conceptual feature group of a preset extraction target, then identifies a conceptual feature group to which each of the character strings included in the document image to be processed belongs from among predetermined multiple conceptual feature groups based on a feature vector corresponding to each character string, and then extracts the character string whose identified conceptual feature group is the same as the conceptual feature group associated with the same attribute information as the conceptual feature group of the extraction target, from among the character strings included in the document image, as the candidate.

A form processing apparatus includes a processor. The processor is configured to, in a case of extracting text data located near a pre-registered item in a form as data associated with the item, display a form to be processed to register an item, accept from a user an area to be filled in with data to be extracted, the area being included in the form to be processed, and display a text located at a position near the accepted area as a candidate for the item.

A device may receive access information that permits access to messages associated with an electronic messaging account. The device may process the messages associated with the electronic messaging account to identify one or more of the messages that include at least one coupon after receiving the access information. The one or more of the messages may be identified using at least two of an image processing technique, a text processing technique, or a code processing technique. The device may store information identifying the one or more of the messages that include the at least one coupon after processing the messages. The device may perform one or more actions to facilitate use of the at least one coupon in association with completing a transaction.