Olive XML importer

Olive Custom classes

This module contains the definition of the Olive importer classes.

The classes define newspaper Issues and Pages objects which convert OCR data in the Olive XML format to a unified canoncial format.

class text_preparation.importers.olive.classes.OliveNewspaperIssue(issue_dir: IssueDir, image_dirs: str, temp_dir: str)

Newspaper Issue in Olive format.

Upon object initialization the following things happen:

  • The Zip archive containing the issue is uncompressed.

  • The ToC file is parsed to determine the logical structure of the issue.

  • Page objects (instances of OliveNewspaperPage) are initialized.

Parameters:

  • issue_dir (IssueDir) – Identifying information about the issue.

  • image_dirs (str) – Path to the directory with the page images. Multiple paths should be separated by comma (“,”).

  • temp_dir (str) – Temporary directory to unpack ZipArchive objects.

id

Canonical Issue ID (e.g. GDL-1900-01-02-a).

Type:

str

edition

Lower case letter ordering issues of the same day.

Type:

str

journal

Newspaper unique identifier or name.

Type:

str

path

Path to directory containing the issue’s OCR data.

Type:

str

date

Publication date of issue.

Type:

datetime.date

issue_data

Issue data according to canonical format.

Type:

dict[str, Any]

pages

list of NewspaperPage instances from this issue.

Type:

list

rights

Access rights applicable to this issue.

Type:

str

image_dirs

Path to the directory with the page images. Multiple paths should be separated by comma (“,”).

Type:

str

archive

ZipArchive for this issue.

Type:

ZipArchive

toc_data

Table of contents (ToC) data for this issue.

Type:

dict

content_elements

All content elements detected.

Type:

list[dict[str, Any]]

content_items

Issue’s recomposed content items.

Type:

list[dict[str, Any]]

clusters

Inverted index of legacy ids; values are clusters of articles, each indexed by one member.

Type:

dict[str, list[str]]

class text_preparation.importers.olive.classes.OliveNewspaperPage(_id: str, number: int, toc_data: dict, image_info: dict, page_xml: str)

Newspaper page in Olive format.

Parameters:

  • _id (str) – Canonical page ID.

  • number (int) – Page number.

  • toc_data (dict) – Metadata about content items in the newspaper issue.

  • page_info (dict) – Metadata about the page image.

  • page_xml (str) – Path to the Olive XML file of the page.

id

Canonical Page ID (e.g. GDL-1900-01-02-a-p0004).

Type:

str

number

Page number.

Type:

int

page_data

Page data according to canonical format.

Type:

dict[str, Any]

issue

Issue this page is from.

Type:

NewspaperIssue | None

toc_data

Metadata about content items in the newspaper issue.

Type:

dict

image_info

Metadata about the page image.

Type:

dict

page_xml

Path to the Olive XML file of the page.

Type:

str

archive

Archive of the issue this page is from.

Type:

ZipArchive

add_issue(issue: NewspaperIssue) → None

Add to a page object its parent, i.e. the newspaper issue.

This allows each page to preserve contextual information coming from the newspaper issue.

Parameters:

issue (NewspaperIssue) – Newspaper issue containing this page.

parse() → None

Process the page XML file and transform into canonical Page format.

Note

This lazy behavior means that the page contents are not processed upon creation of the page object, but only once the parse() method is called.

Raises:

ValueError – No Newspaper issue has been added to this page.

Olive Detect functions

This module contains functions to detect Olive OCR data to be imported.

text_preparation.importers.olive.detect.OliveIssueDir

A light-weight data structure to represent a newspaper issue.

This named tuple contains basic metadata about a newspaper issue. They can then be used to locate the relevant data in the filesystem or to create canonical identifiers for the issue and its pages.

Note

In case of newspaper published multiple times per day, a lowercase letter is used to indicate the edition number: ‘a’ for the first, ‘b’ for the second, etc.

Parameters:

  • journal (str) – Newspaper ID.

  • date (datetime.date) – Publication date or issue.

  • edition (str) – Edition of the newspaper issue (‘a’, ‘b’, ‘c’, etc.).

  • path (str) – Path to the directory containing the issue’s OCR data.

  • rights (str) – Access rights on the data (open, closed, etc.).

>>> from datetime import date
>>> i = OliveIssueDir('GDL', date(1900,1,1), 'a', './GDL-1900-01-01/', 'open')
text_preparation.importers.olive.detect.dir2olivedir(issue_dir: IssueDir, access_rights: dict[str, dict[str, str]]) → OliveIssueDirectory

Helper function that injects access rights info into an IssueDir.

Note

This function is called internally by olive_detect_issues().

Parameters:

  • issue_dir (IssueDir) – Input IssueDir object.

  • access_rights (dict[str, dict[str, str]]) – Access rights information.

Returns:

New OliveIssueDir object.

Return type:

OliveIssueDir

text_preparation.importers.olive.detect.olive_detect_issues(base_dir: str, access_rights: str, journal_filter: set | None = None, exclude: bool = False) → list[OliveIssueDirectory]

Detect newspaper issues to import within the filesystem.

This function expects the directory structure that RERO used to organize the dump of Olive OCR data.

Parameters:

  • base_dir (str) – Path to the base directory of newspaper data.

  • access_rights (str) – Path to access_rights.json file.

  • journal_filter (set | None, optional) – IDs of newspapers to consider. Defaults to None.

  • exclude (bool, optional) – Whether journal_filter should determine exclusion. Defaults to False.

Returns:

List of OliveIssueDir instances, to be imported.

Return type:

list[OliveIssueDir]

text_preparation.importers.olive.detect.olive_select_issues(base_dir: str, config: dict[str, Any], access_rights: str) → list[OliveIssueDirectory]

Detect selectively newspaper issues to import.

The behavior is very similar to olive_detect_issues() with the only difference that config specifies some rules to filter the data to import. See this section for further details on how to configure filtering.

Parameters:

  • base_dir (str) – Path to the base directory of newspaper data.

  • config (dict[str, Any]) – Config dictionary for filtering.

  • access_rights (str) – Path to access_rights.json file.

Returns:

List of OliveIssueDir instances, to be imported.

Return type:

list[OliveIssueDir]

Olive parsers

Functions to parse Olive XML data.

text_preparation.importers.olive.parsers.olive_image_parser(text: bytes) → dict[str, str | list] | None

Parse the Olive XML file containing image metadata.

Parameters:

text (bytes) – Content of the XML file to parse.

Returns:

Dictionary of image metadata.

Return type:

dict[str, str | list] | None

text_preparation.importers.olive.parsers.olive_parser(text: str) → dict[str, dict | list]

Parse an Olive XML file (e.g. from Le Temps corpus).

The main logic implemented here was derived from <https://github.com/dhlab-epfl/LeTemps-preprocessing/>. Each XML file corresponds to one article, as detected by Olive. The final dictionary has keys meta, r, stats and legacy, each mapping to dictionaries or lists with the file’s parsed contents.

Parameters:

text (str) – Contents of the xml file to parse.

Returns:

Dictionary with parsed contents.

Return type:

dict[str, dict | list]

text_preparation.importers.olive.parsers.olive_toc_parser(toc_path: str, issue_dir: IssueDir, encoding: str = 'windows-1252') → dict[int, dict[str, dict]]

Parse the TOC.xml file (Olive format).

For each page, the a dict containing page data is created; mapping content item legacy IDs to their metadata.

Parameters:

  • toc_path (str) – Path to the ToC XML file.

  • issue_dir (IssueDir) – Corresponding IssueDir object.

  • encoding (str, optional) – File’s encoding. Defaults to “windows-1252”.

Returns:

Dictionary where keys are page numbers and

values the corresponding page data dictionary.

Return type:

dict[int, dict[str, dict]]

text_preparation.importers.olive.parsers.parse_styles(text: str) → list[dict[str, Any]]

Turn Olive styleGallery.txt file into a dictionary.

Style IDs may be referred to within the s property of token elements as defined in the impresso JSON schema for newspaper pages (see documentation). Each style has ID, font, font size, color (rgb).

Parameters:

text (str) – textual content of file styleGallery.txt.

Returns:

List of styles according to the impresso schema.

Return type:

list[dict[str, Any]]

Olive Helper methods

Helper functions used by the Olive Importer.

These functions are mainly used within (i.e. called by) the classes OliveNewspaperIssue and OliveNewspaperPage.

class text_preparation.importers.olive.helpers.BoxStrategy(value, names=None, *, module=None, qualname=None, type=None, start=1, boundary=None)
text_preparation.importers.olive.helpers.combine_article_parts(article_parts: list[dict[str, Any]]) → dict[str, Any]

Merge article parts into a single element.

Olive format splits an article into multiple components whenever it spans over multiple pages. Thus, it is necessary to recompose multiple parts.

Parameters:

article_parts (list[dict[str, Any]]) – One or more article parts.

Returns:

Dict with keys meta, fulltext, stats,`legacy`.

Return type:

dict[str, Any]

text_preparation.importers.olive.helpers.compute_box(scale_factor: float, input_box: str) → str | None

Compute IIIF box coordinates of input_box relative to scale_factor.

Parameters:

  • scale_factor (float) – Ratio between 2 images with different dimensions.

  • input_box (str) – String with 4 values separated by spaces.

Returns:

New box coordinates or None if the string had the wrong fromat.

Return type:

str | None

text_preparation.importers.olive.helpers.compute_scale_factor(img_source_path: str, img_dest_path: str) → float

Computes x scale factor bewteen 2 images.

Parameters:

  • img_source_path (str) – Full path ot the source image.

  • img_dest_path (str) – Full path to the destination image.

Returns:

X Scale factor between the two.

Return type:

float

text_preparation.importers.olive.helpers.convert_box(coords: list[int], scale_factor: float) → list[int]

Rescale iiif box coordinates relative to given scale factor.

Parameters:

  • coords (list[int]) – Original box coordinates.

  • scale_factor (float) – Scale factor based on image conversion necessary.

Returns:

Rescaled box coordinates.

Return type:

list[int]

text_preparation.importers.olive.helpers.convert_image_coordinates(image: dict[str, Any], page_xml: str, page_image_name: str, zip_archive: ZipArchive, box_strategy: str, issue: IssueDir) → dict[str, Any]

Convert coordinates of an Olive image element.

Note

This conversion is necessary since the coordinates recorded in the XML file were computed on a different image than the one used for display in the impresso interface.

Parameters:

  • image (dict[str, Any]) – Image metadata.

  • page_xml (str) – Content of Olive page XML.

  • page_image_name (str) – Name of page image file.

  • zip_archive (ZipArchive) – Olive Zip archive.

  • box_strategy (str) – Conversion strategy to apply.

  • issue (IssueDir) – IssueDie of the newspaper issue the page belongs to.

Returns:

Updated image metadata based on the conversion.

Return type:

dict[str, Any]

text_preparation.importers.olive.helpers.convert_page_coordinates(page: dict[str, Any], page_xml: str, page_image_name: str, zip_archive: ZipArchive, box_strategy: str, issue: NewspaperIssue) → bool

Convert coordinates of all elements in a page that have coordinates.

Note

This conversion is necessary since the coordinates recorded in the XML file were computed on a different image than the one used for display in the impresso interface.

Parameters:

  • page (dict[str, Any]) – Page data where coordinates should be converted.

  • page_xml (str) – Content of Olive page XML.

  • page_image_name (str) – Name of page image file.

  • zip_archive (ZipArchive) – Olive Zip archive.

  • box_strategy (str) – Conversion strategy to apply.

  • issue (NewspaperIssue) – Newspaper issue the page belongs to.

Returns:

Whether the coordinate conversion was successful or not.

Return type:

bool

text_preparation.importers.olive.helpers.get_clusters(articles: list[dict[str, Any]]) → dict[str, list[str]]

Created inverted index of legacy ids to article clusters.

Each cluster of articles is indexed by the legacy id of one its members. If a cluster contains only one element, the its id will be in the keys.

Parameters:

articles (list[dict[str, Any]]) – Articles to cluster by legacy ids.

Returns:

Article clusters dictionary.

Return type:

dict[str, list[str]]

text_preparation.importers.olive.helpers.get_scale_factor(issue_dir_path, archive, page_xml, box_strategy, img_source_name)

Returns the scale factor in Olive context, given a strategy to choose the source image.

Parameters:

  • issue_dir_path (str) – the path of the issue

  • archive (zipfile.ZipFile) – the zip archive

  • page_xml (bytes) – the xml handler of the page

  • box_strategy (str) – the box strategy such as found in the info.txt from jp2 folder

  • img_source_name – as found in the info.txt from jp2 folder

Returns:

the hopefully correct scale factor

Return type:

float

Background information

Impresso converts library images to JP2, taking the best image available: tif > highest png > jpg. Olive box coordinates were computed according to an image source which we have to identify among several. Image format coverage is different from issue to issue, and we have to devise strategies.

Case 1: tif

The tif is present and is the file from which the jp2 was converted. Dest: Tif dimensions can therefore be used as jp2 dimensions, no need to read the jp2 file. Source: Image source dimension is present in the page.xml (normally).

Case 2: several png

In this case the jp2 was acquired using the png with the highest dimension. Dest: It looks that in case of several png, Olive also took the highest for the OCR. It is therefore possible to rely on the resolution indicated in the page xml, which should be the same as our jp2. N.B.: the page width and heigth indicated in the xml do not correspond (usually) to the highest resolution png (there is therefore a discrepancy in Olive file between the tag ‘images_resolution’ on the one hand, and ‘page_width|height’on the other). It seems we can ignore this and rely on the resolution only in the current case. Source: the highest png Here source and dest dimension are equals, the function returns 1.

Case 3: one png only

To be checked if it happens. In this case, there is no choice and Olive OCR and JP2 acquisition should be from the same source => scale factor of 1. Here we do an additional check to see if the page_width|height are the same as the image ones. The only danger is if Olive used another image file and did not provide it.

Case 4: one jpg only

Same as Case 3, scale factor of 1. Here we do an additional check to see if the page_width|height are the same as the image ones. (there is only one image and things should fit, not like in case 2)

text_preparation.importers.olive.helpers.keep_title(title: str) → bool

Whether an element’s title should be kept.

The title should not be kept if it is one of “untitled article”, “untitled ad”, and “untitled picture”.

Parameters:

title (str) – Title to verify

Returns:

False if given title is in the black list, True otherwise.

Return type:

bool

text_preparation.importers.olive.helpers.merge_pseudo_tokens(line: dict[str, list[Any]]) → dict[str, list[Any]]

Remove pseudo tokens from a line by merging them.

Parameters:

line (dict[str, list[Any]]) – A line of OCR in JSON format.

Returns:

A new line object (with some merged tokens).

Return type:

dict[str, list[Any]]

text_preparation.importers.olive.helpers.merge_tokens(tokens: list[dict[str, Any]], line: str) → dict[str, Any]

Merge two or more tokens for the same line into one.

The resulting (merged) token will have new coordinates corresponding to the combination of coordinates of the input tokens.

Parameters:

  • tokens (list[dict[str, Any]]) – Tokens to merge.

  • line (str) – The line of text to which the input tokens belong.

Returns:

The new (merged) token.

Return type:

dict[str, Any]

text_preparation.importers.olive.helpers.normalize_hyphenation(line: dict[str, list[Any]]) → dict[str, list[Any]]

Normalize end-of-line hyphenated words.

Parameters:

line (dict[str, list[Any]]) – A line of OCR.

Returns:

A new line element.

Return type:

dict[str, list[Any]]

text_preparation.importers.olive.helpers.normalize_language(language: str) → str

Normalize the language’s string representation.

Parameters:

language (str) – Language to normalize.

Returns:

Normalized language, one of “fr”, “en” and “de”.

Return type:

str

text_preparation.importers.olive.helpers.normalize_line(line: dict[str, list[Any]], lang: str) → dict[str, list[Any]]

Apply normalization rules to a line of OCR.

The normalization rules that are applied depend on the language in which the text is written. This normalization is necessary because Olive, unlike e.g. Mets, does not encode explicitly the presence/absence of whitespaces.

Parameters:

  • line (dict[str, list[Any]]) – A line of OCR text.

  • lang (str) – Language of the text.

Returns:

The new normalized line of text.

Return type:

dict[str, list[Any]]

text_preparation.importers.olive.helpers.recompose_ToC(original_toc_data: dict[int, dict[str, dict]], articles: list[dict[str, Any]], images: list[dict[str, str]]) → list[dict[str, Any]]

Recompose the ToC of a newspaper issue.

Function used by OliveNewspaperIssue.

Parameters:

  • original_toc_data (dict[int, dict[str, dict]]) – ToC data.

  • articles (list[dict[str, Any]]) – List of articles in the issue.

  • images (list[dict[str, str]]) – List of images in the issue.

Returns:

List of final content items in the issue.

Return type:

list[dict[str, Any]]

text_preparation.importers.olive.helpers.recompose_page(page_id: str, info_from_toc: dict[str, dict], page_elements: dict[str, dict], clusters: dict[str, list[str]]) → dict[str, Any]

Merge a list of page elements into a single one.

Note

It is here that an n attribute is assigned to each region/paragraph/line/token.

Parameters:

  • page_id (str) – Page canonical id.

  • info_from_toc (dict[str, dict]) – Dictionary with page element IDs (articles, ads.) as keys, and dictionaries as values.

  • page_elements (dict[str, dict]) – Page’s articles or advertisements.

  • clusters (dict[str, list[str]]) – Inverted index of legacy ids; values are clusters of articles, each indexed by one member.

Returns:

Page data according to impresso canonical format.

Return type:

dict[str, Any]