Tag: extract

Extract PDF Content with Tika

The recently released Apache Hop 1.1 version comes with a very handy new transform step, that allows you to extract the metadata and content of different fileformats like PDF using Apache Tika.

The transform step is available in a pipeline:

Defining Input

It allows you to output the content as Plain text, XML, HTML and JSON:



Output Options

Result:

Scraping Elsevier abstract pages

Get a list of Elsevier’s IDs via Crossref

To get a first impression about the content and size I was using the Crossref Rest API facet search:

https://api.crossref.org/works?filter=member:78&facet=published:*

So for 2020 there are 542k entries. In order to download metadata data (or in my case just the DOI, date of DOI registration and the Elsevier internal ID) I’m using again the REST API of Crossref.

Actually a Crossref query is limited to provide max 1000 DOIs. Using cursors, it’s however to possible to loop further and get about 100k DOIs before the API times out. So in order to get all publications from 2020 I created a monthly batch based on the created date like:

https://api.crossref.org/works?filter=member:78,from-created-date:2020-01-01,until-created-date:2020-01-31&select=DOI,created,alternative-id,&rows=1000&cursor=*

In PDI I’ve created a job, that handles the cursor and repeats the transformation with the REST Query as long there is a new cursor coming back from Crossref.

Pentaho Job: Looping through cursor

So I get a list with all alternative IDs (like: S0960982219315106) from Elsevier articles. With that I can create an URL to all abstract pages of this article: https://www.sciencedirect.com/science/article/abs/pii/S0960982219315106

Getting HTML Abstract Page

Using the Pentaho HTTP Client, it’s now possible to get the HTML of the this abstract page.

https://www.sciencedirect.com/science/article/abs/pii/S0960982219315106

It’s important that there’s a HTTP-Header like: “Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_5)”, otherwise the Pentaho Client will be routed to a error page.

Extracting data from abstract HTML

Elsevier makes it very easy to extract information about the article without going deep into the HTML-Structure of ScienceDirect. Within the HTML there’s a whole section with structured data in JSON:

view-source:https://www.sciencedirect.com/science/article/abs/pii/S0960982219315106

Loading the whole HTML in one field, the following Regex can be used to extract the json block:

*<script type="application/json" data-iso-key="_0">(.*)</script>
<iframe style="display: none.*
Pentaho Step: Regex evaluation (Content Tab: Activate “Enable dotall mode” & “Enable multiline mode”)

Extracting data from JSON

Having the JSON data block separated from the rest of the HTML, we now can pass it to the JSON-input step. There is really plenty of information and much more than you would see via front end.

Exploring Json-Data and getting JSON-Path via https://jsonpathfinder.com/

For the moment I’m mostly interested in the author and affiliation data:

Extracting Affiliations

In the JSON we can extract:

  1. whole affiliation section
  2. individual affiliation
  3. id, label, name (textfn) of affiliation
JSON Path for affiliations

Extracting Authors

extracting the authors works similar like the affiliation.

  1. whole author section: $..[?(@.#name==’author’)]
  2. individual author: $.$.id ; $.$.orcid ; $..$$.[?(@.#name==’given-name’)]._
  3. label: $.$$.[?(@.#name==’cross-ref’)].$$.[?(@.#name==’sup’)]._
JSON-Path for authors

An author can have multiple labels/refids, emails, contributor-roles. In order to get those flat in a row separated by a coma, a group step can be used, while getting it back into the stream using the “stream lookup”-step:

Merge affiliations and authors

In a third step authors and affiliations are merged by the the label:

in order to get a list with authors and affiliations

Extracted authors and affiliations from Elsevier articles