cgspace-notes/content/posts/2024-05.md

---
title: "May, 2024"
date: 2024-05-01T10:39:00+03:00
author: "Alan Orth"
categories: ["Notes"]
---

## 2024-05-01

- I dumped all the CGSpace DOIs and resolved them with my `crossref_doi_lookup.py` script
  - Then I did some work to add missing abstracts (about 900!), volumes, issues, licenses, publishers, and types, etc

<!--more-->

## 2024-05-05

- Spend some time looking at duplicate DOIs again...

## 2024-05-06

- Spend some time looking at duplicate DOIs again...

## 2024-05-07

- Discuss RSS feeds and OpenSearch with IWMI
  - It seems our OpenSearch feed settings are using the defaults, so I need to copy some of those over from our old DSpace 6 branch
- I saw a patch for an interesting issue on DSpace GitHub: [Error submitting or deleting items - URI too long when user is in a large number of groups](https://github.com/DSpace/DSpace/issues/9544)
  - I hadn't realized it, but we have lots of those errors:

```console
$ zstdgrep -a 'URI Too Long' log/dspace.log-2024-04-* | wc -l
1423
```

- Spend some time looking at duplicate DOIs again...

## 2024-05-08

- Spend some time looking at duplicate DOIs again...
  - I finally finished looking at the duplicate DOIs for journal articles
  - I updated the list of handle redirects and there are 386 of them!

## 2024-05-09

- Spend some time working on the IFPRI 2020–2021 batch
  - I started by checking for exact duplicates (1.0 similarity) using DOI, type, and issue date

## 2024-05-12

- I couldn't figure out how to do a complex join on withdrawn items along with their metadata, so I pull out a few like titles, handles, and provenance separately:

```psql
dspace=# \COPY (SELECT i.uuid, m.text_value AS uri FROM item i JOIN metadatavalue m ON i.uuid = m.dspace_object_id WHERE withdrawn AND m.metadata_field_id=25) TO /tmp/withdrawn-handles.csv CSV HEADER;
dspace=# \COPY (SELECT i.uuid, m.text_value AS title FROM item i JOIN metadatavalue m ON i.uuid = m.dspace_object_id WHERE withdrawn AND m.metadata_field_id=64) TO /tmp/withdrawn-titles.csv CSV HEADER;
dspace=# \COPY (SELECT i.uuid, m.text_value AS submitted_by FROM item i JOIN metadatavalue m ON i.uuid = m.dspace_object_id WHERE withdrawn AND m.metadata_field_id=28 AND m.text_value LIKE 'Submitted by%') TO /tmp/withdrawn-submitted-by.csv CSV HEADER;
```

- Then joined them:

```console
$ csvjoin -c uuid /tmp/withdrawn-title.csv /tmp/withdrawn-handles.csv /tmp/withdrawn-submitted-by.csv > /tmp/withdrawn.csv
```

- This gives me an insight into who submitted at 334 of the duplicates over the past few years...
- I fixed a few hundred titles with leading/trailing whitespace, newlines, and ligatures like ff, fi, fl, ffi, and ffl

## 2024-05-13

- Export a list of IFPRI information products with handle links and CONTENTdm links:

```
$ csvgrep -c 'dc.description.provenance[en_US]' -m 'CONTENTdm' cgspace.csv \
  | csvcut -c 'id,dc.description.provenance[en_US],dc.identifier.uri[en_US]' \
  | tee /tmp/ifpri-redirects.csv \
  | csvstat --count
2645
```

- I discovered the `/server/api/pid/find` endpoint today, which is much more direct and manageable than the `/server/api/discover/search/objects?query=` endpoint when trying to get metadata for a Handle (item, collection, or community)
  - The "pid" stands for permanent identifiers apparently, and we can use it like this:

```
https://dspace7test.ilri.org/server/api/pid/find?id=10568/118424
```

## 2024-05-15

- I got journal titles for 2,900 journal articles that were missing them from Crossref

## 2024-05-16

Helping IFPRI with some DSpace 7 API support, these are two queries for items issued in 2024:
  - https://dspace7test.ilri.org/server/api/discover/search/objects?query=dcterms.issued:2024
  - https://dspace7test.ilri.org/server/api/discover/search/objects?query=dcterms.issued_dt%3A%5B2024-01-01T00%3A00%3A00Z%20TO%20%2A%5D — note the Lucene search syntax is URL encoded version of `:[2024-01-01T00:00:00Z TO *]`

Both of them return the same number of results and seem identitical as far as I can see, but the second one uses Solr date indexes and requires the full Lucene datetime and range syntax

I wrote a new version of the `check_duplicates.py` script to help identify duplicates with different types
  - Initially I called it `check_duplicates_fast.py` but it's actually not faster
  - I need to find a way to deal with duplicates from IFPRI's repository because there are some mismatched types...

## 2024-05-20

Continue working through alternative duplicate matching for IFPRI
  - Their item types are sometimes different than ours...
  - One thing I think I can say for sure is that the default similarity factor in my script is 0.6, and I rarely see legitimate duplicates with such similarity so I might increase this to 0.7 to reduce the number of items I have to check
  - Also, the difference in issue dates is currently 365, but I should reduce that a bit, perhaps to 270 days (9 months)

## 2024-05-22

- Finalize and upload the IFPRI 2020–2021 batch set
  - I used a new technique to get missing licenses via Crossref (it's Python 2 because of OpenRefine's Jython):

```python
import urllib2

doi = cells['cg.identifier.doi[en_US]'].value
url = "https://api.crossref.org/works/" + doi
useragent = "Python (mailto:a.o@cgiar.org)"

request = urllib2.Request(url.encode("utf-8"), headers={"User-Agent" : useragent})
get = urllib2.urlopen(request)

return get.read().decode('utf-8')
```

## 2024-05-23

- Finalize last of the duplicates I found for the IFPRI 2020–2021 batch set (those that we missed initially due to mismatched types)
- Export a new list of IFPRI redirects from CONTENTdm:

```console
$ csvgrep -c 'dc.description.provenance[en_US]' -r 'Original URLs? from IFPRI CONTENTdm' cgspace.csv \
  | csvcut -c 'id,dc.description.provenance[en_US],dc.identifier.uri[en_US]' \
  | tee /tmp/ifpri-redirects.csv \
  | csvstat --count
4004
```

I found a way to get abstracts from PLOS
  - They offer an API that returns XML including the JATS-formatted abstracts
  - I created a new column in OpenRefine by fetching specially crafted URLs based on the DOIs using this GREL:

```console
"https://journals.plos.org/plosone/article/file?id=" + cells['doi'].value + '&type=manuscript'
```

Then used `value.parseXml()` on the resulting text to extract the abstract's text:

```console
value.parseXml().select("abstract")[0].xmlText()
```

This doesn't preserve `<p>` tags though...
  - Oh, nice, this does!

```console
forEach(value.parseHtml().select("abstract p"), i, i.htmlText()).join("\r\n\r\n")
```

For each paragraph inside an abstract, get the inner text and join them as one string separated by two newlines...
  - Ah, some articles have multiple abstracts, for example: https://journals.plos.org/plosone/article/file?id=https://doi.org/10.1371/journal.pntd.0001859&type=manuscript
  - I need to select the abstract that does **not** have any attributes (using [Jsoup selector syntax](https://jsoup.org/apidocs/org/jsoup/select/Selector.html))

```console
forEach(value.parseXml().select("abstract:not([*]) p"), i, i.xmlText()).join("\r\n\r\n")
```

Testing `xsv` (Rust) versus `csvkit` (Python) to filter all items with DOIs from a DSpace dump with 118,000 items:

```console
$ time xsv search -s doi 'doi\.org' /tmp/cgspace-minimal.csv | xsv select doi | xsv count
27339
xsv search -s doi 'doi\.org' /tmp/cgspace-minimal.csv  0.06s user 0.03s system 98% cpu 0.091 total
xsv select doi  0.02s user 0.02s system 40% cpu 0.091 total
xsv count  0.01s user 0.00s system 9% cpu 0.090 total
$ time csvgrep -c doi -m 'doi.org' /tmp/cgspace-minimal.csv | csvcut -c doi | csvstat --count
27339
csvgrep -c doi -m 'doi.org' /tmp/cgspace-minimal.csv  1.15s user 0.06s system 95% cpu 1.273 total
csvcut -c doi  0.42s user 0.05s system 36% cpu 1.283 total
csvstat --count  0.20s user 0.03s system 18% cpu 1.298 total
```

## 2024-05-27

- Working on IFPRI datasets batch migration
  - 732 items total
  - 6 duplicates on CGSpace
  - 6 duplicates within set that need investigation

## 2024-05-28

- I'm thinking of increasing the frequency of thumbnail generation on CGSpace
  - Currently the `dspace filter-media` script runs once at 3AM for all media types and seems to take ~10 minutes to run for all 118,000 items...
  - I think I will make the thumbnailer run explicitly more often using `-p "ImageMagick PDF Thumbnail"`

<!-- vim: set sw=2 ts=2: -->