Method, apparatus, and computer program product for classification of documents

• US 10,339,375 B2
• Filed: 02/27/2017
• Issued: 07/02/2019
• Est. Priority Date: 08/16/2012
• Status: Active Grant

• Current Assignee: Groupon Incorporated
• Original Assignee: Groupon Incorporated
• Inventors: Castillo, Roger Henry, Humphrey, Brian Andrew
• Primary Examiner(s): Entezari, Michelle M
• Application Number:
  US15/443,550
• Publication Number:
  US 20170228588A1
• Time in Patent Office: 855 Days
• Field of Search: None
• CPC Class Codes: G06F 16/3347 using vector based model
• CPC Class Codes: G06F 16/93 Document management systems
• CPC Class Codes: G06F 16/957 Browsing optimisation, e.g. caching or content distillation
• CPC Class Codes: G06F 40/295 Named entity recognition
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06V 30/413 Classification of content, e.g. text, photographs or tables

Abstract

Provided herein are systems, methods and computer readable media for classification of documents using a location hierarchy. An example method may include receiving a feature vector r that represents occurrence counts of references in a document'"'"'s text to each of a group of named entities, and determining whether the document is associated with the particular location by querying, to determine a query result, using feature vector r, at least one location-specific classifier from a group of location-specific classifiers, wherein the location-specific classifier is associated with the particular location, and wherein the location-specific classifier is configured to generate a positive output value in response to receiving an input feature vector representing occurrence count of at least one reference to the particular named entity and determining that the document is associated with the particular location in an instance in which the query result includes data indicating that the positive output value was generated by the location-specific classifier that is associated with the particular location.

Claims

1. A computer-implemented method for identifying content to represent web pages and creating thumbnails from the content, the computer-implemented method comprising:

retrieving a web document using a uniform resource locator (URL) contained in a dequeued work item, the dequeued work item parsed using a markup language parser;

determining, from the web document, candidate images for thumbnail creation,

wherein the determination of the candidate images for thumbnail creation comprises at least;

identifying a desired thumbnail size and aspect ratio;

extracting data content from the parsed markup to determine one or more candidate images for thumbnail creation; and

utilizing one or more heuristics to discard candidate images having predefined undesirable characteristics, including at least discarding, from among the extracted one or more images, any images failing to meet the desired thumbnail size and aspect ratio; and

creating a thumbnail image, wherein generation of the thumbnail image comprises at least;

cropping a chosen image, the chosen image selected from among the candidate images, to each of one or more predefined sizes and encoding the chosen image with predefined compression settings, each in accordance with an environment in which the thumbnails will be used.

2. The computer-implemented method of claim 1, further comprising:

extracting one or more image references from the parsed markup, wherein the image references are extracted by identifying image tags with source attributes, values of which are URLs locating images;

filtering the URLs using a blacklist of expressions designed to match the URLs of images comprising one or more predefined undesirable characteristics; and

retrieving the images which do not match any of the expressions using an HTTP client, the retrieved images being candidate images.

3. The computer-implemented method of claim 1, further comprising:

extracting one or more video references from the parsed markup, wherein the video references are extracted by finding video tags containing source tags with source attributes, embed tags with source attributes, or object tags with data attributes, values of which are URLs locating videos;

retrieving the videos using an HTTP client,

wherein the video retriever stops retrieval of a video after receiving an initial segment of a predefined minimum size;

decoding the retrieved videos; and

extracting image frames from the videos from predefined points in the video, the extracted image frames being candidate images.

4. The computer-implemented method of claim 1, further comprising:

scoring and aggregating discovered features of the candidate images to determine an overall score for each image; and

choosing the image with the highest score to represent the document.

5. The computer-implemented method of claim 1, further comprising:

detecting duplicate images, detection of duplicate images performed by (i) storing, for each image processed, a cryptographic hash of the image'"'"'s binary data, (ii) subsequently when an image is processed, computing its hash and comparing to previously computed hashes, and (iii) upon finding a match, excluding the image from further processing thus making each thumbnail unique;

decoding remaining images; and

for each image, identifying a size and aspect ratio, and discarding, in accordance with a desired thumbnail size and aspect ratio, images of comprising sizes and aspect ratios failing to meet predefined thresholds.

6. The computer-implemented method of claim 1, further comprising:

computing a color histogram of each decoded image by determining a number of pixels of each color in the image; and

in an instance in which more than a desired fraction of the pixels are of a single color, discarding the image.

7. The computer-implemented method of claim 1, further comprising:

performing an optical character recognition (OCR) process on each candidate image;

returning text found in the candidate image;

performing a text classification process to determine if the image is an advertisement, wherein, in advance of the performance of the text classification process, a collection of images labelled “

advertisement”

/“

not advertisement”

are used to train the classifier; and

upon processing an image, the classifier is provided the text found by OCR and returns either “

advertisement”

or “

not advertisement”

; and

discarding the candidate image in an instance in which the classifier returns “

advertisement”

for the image.

8. The computer-implemented method of claim 1, further comprising:

performing a face detection process on the candidate image.

9. The computer-implemented method of claim 1, further comprising:

performing a measurement of a sharpness of the candidate image by first performing an edge detection by convolving the candidate image with a convolution filter, thresholding the resulting pixel values at a predefined value resulting in a partition of original image pixels into “

edge” and

“

non-edge”

pixels, computing the ratio of the number of “

edge”

pixels to the number of “

non-edge”

pixels; and

discarding the candidate image if this ratio lies outside predefined bounds thus eliminating candidate images with excessive edges or with too few images.

10. The computer-implemented method of claim 9, further comprising:

blurring the candidate image using a standard Gaussian blur, re-performing the edge detection process on the blurred image to determine a blurred “

edge”

/“

non-edge”

ratio; and

measuring a ratio of the original “

edge”

/“

non-edge”

ratio to the blurred ratio to determine a sharpness of the image.

11. An apparatus for identifying content to represent web pages and creating thumbnails from the content comprising at least one processor and at least one memory including computer program code, the at least one memory and the computer program code configured to, with the processor, cause the apparatus to at least:

retrieve a web document using a uniform resource locator (URL) contained in a dequeued work item, the dequeued work item parsed using a markup language parser;

determine, from the web document, candidate images for thumbnail creation,

wherein the computer program code configured to, with the processor, cause the apparatus to determine the candidate images for thumbnail creation, comprises computer program code configured to, with the processor, cause the apparatus to at least;

identify a desired thumbnail size and aspect ratio;

extract data content from the parsed markup to determine one or more candidate images for thumbnail creation;

utilize one or more heuristics to discard candidate images having predefined undesirable characteristics, including computer program code configured to, with the processor, cause the apparatus to at least discard, from among the extracted one or more images, any images failing to meet the desired thumbnail size and aspect ratio; and

create a thumbnail image, wherein the computer program code configured to, with the processor, cause the apparatus to generate the thumbnail image comprises computer program code configured to, with the processor, cause the apparatus to at least;

crop a chosen image, the chosen image selected from among the candidate images, to each of one or more predefined sizes and encoding the chosen image with predefined compression settings, each in accordance with an environment in which the thumbnails will be used.

12. The apparatus of claim 11, wherein the at least one memory and the computer program code are further configured to, with the processor, cause the apparatus to:

extract one or more image references from the parsed markup, wherein the image references are extracted by identifying image tags with source attributes, values of which are URLs locating images;

filter the URLs using a blacklist of expressions designed to match the URLs of images comprising one or more predefined undesirable characteristics; and

retrieve the images which do not match any of the expressions using an HTTP client, the retrieved images being candidate images.

13. The apparatus of claim 11, wherein the at least one memory and the computer program code are further configured to, with the processor, cause the apparatus to:

extract one or more video references from the parsed markup, wherein the video references are extracted by finding video tags containing source tags with source attributes, embed tags with source attributes, or object tags with data attributes, values of which are URLs locating videos;

retrieve the videos using an HTTP client,

wherein the video retriever stops retrieval of a video after receiving an initial segment of a predefined minimum size;

decode the retrieved videos; and

extract image frames from the videos from predefined points in the video, the extracted image frames being candidate images.

14. The apparatus of claim 11, wherein the at least one memory and the computer program code are further configured to, with the processor, cause the apparatus to:

score and aggregating discovered features of the candidate images to determine an overall score for each image; and

choose the image with the highest score to represent the document.

15. The apparatus of claim 11, wherein the at least one memory and the computer program code are further configured to, with the processor, cause the apparatus to:

detect duplicate images, detection of duplicate images performed by (i) storing, for each image processed, a cryptographic hash of the image'"'"'s binary data, (ii) subsequently when an image is processed, computing its hash and comparing to previously computed hashes, and (iii) upon finding a match, excluding the image from further processing thus making each thumbnail unique;

decode remaining images; and

for each image, identifying a size and aspect ratio, and discard, in accordance with a desired thumbnail size and aspect ratio, images of comprising sizes and aspect ratios failing to meet predefined thresholds.

16. The apparatus of claim 11, wherein the at least one memory and the computer program code are further configured to, with the processor, cause the apparatus to:

compute a color histogram of each decoded image by determining a number of pixels of each color in the image; and

in an instance in which more than a desired fraction of the pixels are of a single color, discard the image.

17. The apparatus of claim 11, wherein the at least one memory and the computer program code are further configured to, with the processor, cause the apparatus to:

perform an optical character recognition (OCR) process on each candidate image;

return text found in the candidate image;

perform a text classification process to determine if the image is an advertisement, wherein, in advance of the performance of the text classification process, a collection of images labelled “

advertisement”

/“

not advertisement”

are used to train the classifier; and

upon processing an image, the classifier is provided the text found by OCR and returns either “

advertisement”

or “

not advertisement”

; and

discard the candidate image in an instance in which the classifier returns “

advertisement”

for the image.

18. The apparatus of claim 11, wherein the at least one memory and the computer program code are further configured to, with the processor, cause the apparatus to:

perform a face detection process on the candidate image.

19. The apparatus of claim 11, wherein the at least one memory and the computer program code are further configured to, with the processor, cause the apparatus to:

perform a measurement of a sharpness of the candidate image by first performing an edge detection by convolving the candidate image with a convolution filter, thresholding the resulting pixel values at a predefined value resulting in a partition of original image pixels into “

edge” and

“

non-edge”

pixels, computing the ratio of the number of “

edge”

pixels to the number of “

non-edge”

pixels; and

discard the candidate image if this ratio lies outside predefined bounds thus eliminating candidate images with excessive edges or with too few images.

20. The apparatus of claim 19, wherein the at least one memory and the computer program code are further configured to, with the processor, cause the apparatus to:

blur the candidate image using a standard Gaussian blur, re-performing the edge detection process on the blurred image to determine a blurred “

edge”

/“

non-edge”

ratio; and

measure a ratio of the original “

edge”

/“

non-edge”

ratio to the blurred ratio to determine a sharpness of the image.

21. A computer program product for identifying content to represent web pages and creating thumbnails from the content comprising at least one non-transitory computer-readable storage medium having computer-executable program code instructions stored therein, the computer-executable program code instructions comprising program code instructions for:

retrieving a web document using a uniform resource locator (URL) contained in a dequeued work item, the dequeued work item parsed using a markup language parser;

determining, from the web document, candidate images for thumbnail creation,

wherein the determination of the candidate images for thumbnail creation comprises at least;

identifying a desired thumbnail size and aspect ratio;

extracting data content from the parsed markup to determine one or more candidate images for thumbnail creation;

utilizing one or more heuristics to discard candidate images having predefined undesirable characteristics, including at least discarding, from among the extracted one or more images, any images failing to meet the desired thumbnail size and aspect ratio; and

creating a thumbnail image, wherein generation of the thumbnail image comprises at least;

cropping a chosen image, the chosen image selected from among the candidate images, to each of one or more predefined sizes and encoding the chosen image with predefined compression settings, each in accordance with an environment in which the thumbnails will be used.