This project is a command line tool used by the National Library of Luxembourg (Bibliothèque nationale de Luxembourg - BnL) to parse METS/ALTO documents and export them into another format.

Note that this project is highly tailored to the BnL's METS/ALTO requirements and current workflows.

• Java 8
• Maven

This program is tailored for the data of the National Library of Luxembourg. You can download a small dataset here.

Download or clone this repository, then go inside and build it using Maven: mvn install

A default config_example folder is provided with this project. Make a copy and rename it to config.

For testing purposes, you can use the config as it is, but the different URLs should be modified for your needs in the export.yml.

To test the configuration you can run:

java -jar BnLMetsExporter.jar --testconfig

The tool supports 2 modes:

1. Primo: Parse METS/ALTO and exports each document unit to a XML document (OAI with Dublin Core). The result is directly saved into a tar.gz.
2. Solr: Parse METS/ALTO and load each document unit to Solr indices.

Note: A document unit is for example a newspaper article, section, illustration, advertisement, ... or book chapter, ... and is defined in the export.yml file.

Command for exporting to tar.gz from a local folder (where your METS/ATLO are):

java -jar BnLMetsExporter.jar -export primo -dir path/to/metsalto

For exporting to Solr, refer to the advanced section below.

The output will be saved in: ./import-for-primo.tar.gz.

For help and details on the available commands, just run:

java -jar BnLMetsExporter.jar -h

To build, simply call mvn install. The jar file will be BnLMetsExporter.jar in a folder named target.

See COPYING to see the full text.

Contributions to this project will be evaluated on a case-by-case basis. Some part of this software is general, but this project is highly tailored to the needs of the National Library of Luxembourg. If your needs are different, please fork your own version and modify it.

The program can fetch Mets files from an URL in the following format: https://127.0.0.1:8080/getMets?pid=. You have to provide a list of IDs in a line-by-line text file e.g. -pids 10pids.lst. The PID will be appended to the "get Mets Url". Note that Alto files should still be accessible via disk and so you have to provide the argument -dir path/to/metsalto that points to the root where alto files can be found relative to what is written in the METS file.

Custom

In addition to the already implemented methods, you can write your own implementation of the abstract class MetsGetter. It is located in the package lu.bnl.reader.

In case you want to do an export to Solr, the following chapters will present a guide to know how to get started with a Solr installation.

The easiest way to create and manage a Solr Cloud on a development environment is using the Solr cloud example as explained here: https://lucene.apache.org/solr/guide/6_6/getting-started-with-solrcloud.html

On the development environment, create a new configset in solr/server/solr/configsets/. For this, just copy the basic_configs/ to my_config_sets. Under my_config_set/conf/, upload your own configuration from the folder config_example/solr of this project.

Then the config must be reloaded into ZooKeeper (upconfig):

$ solr-6.6.0/server/scripts/cloud-scripts/zkcli.sh -zkhost localhost:9983 -cmd upconfig -confname my_config_set -confdir solr-6.6.0/server/solr/configsets/my_config_set/conf/

After this step. All collections must be reloaded.

IMPORTANT: The Solr config in this repository is for development only.

There are 2 indices:

1. Article Index: Stores the data of a single article, chapter, section, etc.
2. Page Index: Stores the data of a complete page

So, you must create two cores or collections using the configsets above. The default names are my_articles and my_pages as written in the default export.yml

In the config, note that articleURL and pageURL is for Solr running in single host mode (not cloud). articleCollection and pageCollection along with zkHost must be specified for a Solr running in Cloud mode.

Use the command -clean to clean the Solr article and page indices. Add -cloud if it is a Solr Cloud environment.

$ java -jar BnLMetsExporter.jar -clean -cloud

• Solr is in ClassicIndexSchemaFactory mode. The code <schemaFactory class="ClassicIndexSchemaFactory"/> has been added to the config/solr/solrconfig.xml file.
• There is only 1 schema. Articles and Pages use the same schema.
• Most dynamic fields are commented out.
• The custom field text_primo tries to follow the same analysis and query rules as the internal system of the BnL (Primo).

The full text is stored in 2 text fields, text_lines and text_words.

• text_lines: Stores the full text, which contains HTML and custom XML tags and is meant for web display to a user.
• text_word: Stores each word and its coordinate on the page in a custom format (using XML tags).

The current BnLViewer displays the full text and needs the line by line information. This allows to display the full text in the same format as it was on the physical paper.

The easiest solution is to store the line ending information inside the full text directly. HTML tags allows to have a standard way to mark this information without interfering with the analyser, because it uses an solr.HTMLStripCharFilterFactory (only during indexing, not during querying).

This has the advantage of being simple to parse, manipulate and to still use Solr's built-in highlight feature.

Line endings are stored using custom XML tags in the text_lines field. At the moment, each new line is marked with a <br>. Every line ending with hyphenated character is stored as <le s="X-" e="Y">XY</le>. s contains X- (with hyphen), which is the start of the word. e contains Y, which is the end of the word. X is located in line n and Y is located in line n+1. The word (without hyphen) XY is required to allow the analyser to search it fully. This word is also what is highlighted, meaning that some post-processing is required to recreate the correct line by line text.

Word coordinates are stored using custom XML tags in the text_words field. The full text is stored as as a sequence of <w> tags like this:

<w a="DTL1" b="BLOCK1" p="ALTO00001" x="300" y="1400" w="287" h="68">Hello</w>
<w a="DTL1" b="BLOCK2" p="ALTO00001" x="600" y="1400" w="290" h="67">World</w>
...

Again, using the solr.HTMLStripCharFilterFactory allows to ignore all tags, but still be able to search the words using the same analyser as if it were a full text. The Solr highlighting works perfectly with the correct options, such as hl.fragsize=0 and hl.maxAnalyzedChars=10000000. This solutions allows to easily retrieve the coordinates of the words for any page and article.