October, 2021

Fri Oct 01, 2021 in Notes

2021-10-01

Export all affiliations on CGSpace and run them against the latest RoR data dump:

localhost/dspace63= > \COPY (SELECT DISTINCT text_value as "cg.contributor.affiliation", count(*) FROM metadatavalue WHERE dspace_object_id IN (SELECT uuid FROM item) AND metadata_field_id = 211 GROUP BY text_value ORDER BY count DESC) to /tmp/2021-10-01-affiliations.csv WITH CSV HEADER;
$ csvcut -c 1 /tmp/2021-10-01-affiliations.csv | sed 1d > /tmp/2021-10-01-affiliations.txt
$ ./ilri/ror-lookup.py -i /tmp/2021-10-01-affiliations.txt -r 2021-09-23-ror-data.json -o /tmp/2021-10-01-affili
ations-matching.csv
$ csvgrep -c matched -m true /tmp/2021-10-01-affiliations-matching.csv | sed 1d | wc -l 
1879
$ wc -l /tmp/2021-10-01-affiliations.txt 
7100 /tmp/2021-10-01-affiliations.txt

So we have 1879/7100 (26.46%) matching already

2021-10-03

Dominique from IWMI asked me for information about how CGSpace partners are using CGSpace APIs to feed their websites
Start a fresh indexing on AReS
Udana sent me his file of 292 non-IWMI publications for the Virtual library on water management
- He added licenses
- I want to clean up the dcterms.extent field though because it has volume, issue, and pages there
- I cloned the column several times and extracted values based on their positions, for example:
  - Volume: value.partition(":")[0]
  - Issue: value.partition("(")[2].partition(")")[0]
  - Page: "p. " + value.replace(".", "")

2021-10-04

Start looking at the last month of Solr statistics on CGSpace
- I see a number of IPs with “normal” user agents who clearly behave like bots
  - 198.15.130.18: 21,000 requests to /discover with a normal-looking user agent, from ASN 11282 (SERVERYOU, US)
  - 93.158.90.107: 8,500 requests to handle and browse links with a Firefox 84.0 user agent, from ASN 12552 (IPO-EU, SE)
  - 193.235.141.162: 4,800 requests to handle, browse, and discovery links with a Firefox 84.0 user agent, from ASN 51747 (INTERNETBOLAGET, SE)
  - 3.225.28.105: 2,900 requests to REST API for the CIAT Story Maps collection with a normal user agent, from ASN 14618 (AMAZON-AES, US)
  - 34.228.236.6: 2,800 requests to discovery for the CGIAR System community with user agent Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1), from ASN 14618 (AMAZON-AES, US)
  - 18.212.137.2: 2,800 requests to discovery for the CGIAR System community with user agent Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1), from ASN 14618 (AMAZON-AES, US)
  - 3.81.123.72: 2,800 requests to discovery and handles for the CGIAR System community with user agent Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1), from ASN 14618 (AMAZON-AES, US)
  - 3.227.16.188: 2,800 requests to discovery and handles for the CGIAR System community with user agent Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1), from ASN 14618 (AMAZON-AES, US)
- Looking closer into the requests with this Mozilla/4.0 user agent, I see 500+ IPs using it:

# zcat --force /var/log/nginx/*.log* | grep 'Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1)' | awk '{print $1}' | sort | uniq > /tmp/mozilla-4.0-ips.txt
# wc -l /tmp/mozilla-4.0-ips.txt 
543 /tmp/mozilla-4.0-ips.txt

Then I resolved the IPs and extracted the ones belonging to Amazon:

$ ./ilri/resolve-addresses-geoip2.py -i /tmp/mozilla-4.0-ips.txt -k "$ABUSEIPDB_API_KEY" -o /tmp/mozilla-4.0-ips.csv
$ csvgrep -c asn -m 14618 /tmp/mozilla-4.0-ips.csv | csvcut -c ip | sed 1d | tee /tmp/amazon-ips.txt | wc -l

I am thinking I will purge them all, as I have several indicators that they are bots: mysterious user agent, IP owned by Amazon
Even more interesting, these requests are weighted VERY heavily on the CGIAR System community:

   1592 GET /handle/10947/2526
   1592 GET /handle/10947/2527
   1592 GET /handle/10947/34
   1593 GET /handle/10947/6
   1594 GET /handle/10947/1
   1598 GET /handle/10947/2515
   1598 GET /handle/10947/2516
   1599 GET /handle/10568/101335
   1599 GET /handle/10568/91688
   1599 GET /handle/10947/2517
   1599 GET /handle/10947/2518
   1599 GET /handle/10947/2519
   1599 GET /handle/10947/2708
   1599 GET /handle/10947/2871
   1600 GET /handle/10568/89342
   1600 GET /handle/10947/4467
   1607 GET /handle/10568/103816
 290382 GET /handle/10568/83389

Before I purge all those I will ask someone Samuel Stacey from the System Office to hopefully get an insight…
Meeting with Michael Victor, Peter, Jane, and Abenet about the future of repositories in the One CGIAR
Meeting with Michelle from Altmetric about their new CSV upload system
- I sent her some examples of Handles that have DOIs, but no linked score (yet) to see if an association will be created when she uploads them

doi,handle
10.1016/j.agsy.2021.103263,10568/115288
10.3389/fgene.2021.723360,10568/115287
10.3389/fpls.2021.720670,10568/115285

Extract the AGROVOC subjects from IWMI’s 292 publications to validate them against AGROVOC:

$ csvcut -c 'dcterms.subject[en_US]' ~/Downloads/2021-10-03-non-IWMI-publications.csv | sed -e 1d -e 's/||/\n/g' -e 's/"//g' | sort -u > /tmp/agrovoc.txt
$ ./ilri/agrovoc-lookup.py -i /tmp/agrovoc-sorted.txt -o /tmp/agrovoc-matches.csv
$ csvgrep -c 'number of matches' -m '0' /tmp/agrovoc-matches.csv | csvcut -c 1 > /tmp/invalid-agrovoc.csv

2021-10-05

Sam put me in touch with Dodi from the System Office web team and he confirmed that the Amazon requests are not theirs
- I added Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1) to the list of bad bots in nginx
- I purged all the Amazon IPs using this user agent, as well as the few other IPs I identified yesterday

$ ./ilri/check-spider-ip-hits.sh -f /tmp/robot-ips.txt -p
...

Total number of bot hits purged: 465119

2021-10-06

Thinking about how we could check for duplicates before importing
- I found out that PostgreSQL has a built-in similarity function:

localhost/dspace63= > CREATE EXTENSION pg_trgm;
localhost/dspace63= > SELECT metadata_value_id, text_value, dspace_object_id FROM metadatavalue WHERE dspace_object_id IN (SELECT uuid FROM item) AND metadata_field_id=64 AND SIMILARITY(text_value,'Molecular marker based genetic diversity assessment of Striga resistant maize inbred lines') > 0.5;
 metadata_value_id │                                         text_value                                         │           dspace_object_id
───────────────────┼────────────────────────────────────────────────────────────────────────────────────────────┼──────────────────────────────────────
           3652624 │ Molecular marker based genetic diversity assessment of Striga resistant maize inbred lines │ b7f0bf12-b183-4b2f-bbd2-7a5697b0c467
           3677663 │ Molecular marker based genetic diversity assessment of Striga resistant maize inbred lines │ fb62f551-f4a5-4407-8cdc-6bff6dac399e
(2 rows)

I was able to find an exact duplicate for an IITA item by searching for its title (I already knew that these existed)
I started working on a basic Python script to do this and managed to find an actual duplicate in the recent IWMI items
- I think I will check for similar titles, and if I find them I will print out the handles for verification
- I could also proceed to check other metadata like type because those shouldn’t vary too much
I ran my new check-duplicates.py script on the 292 non-IWMI publications from Udana and found twelve potential duplicates
- Upon checking them manually, I found that 7/12 were indeed already present on CGSpace!
- This is with the similarity threshold at 0.5. I wonder if tweaking that higher will make the script run faster and eliminate some false positives
- I re-ran it with higher thresholds this eliminated all false positives, but it still took 24 minutes to run for 292 items!
  - 0.6: ./ilri/check-duplicates.py -i ~/Downloads/2021-10-03-non-IWMI-publications.cs 0.09s user 0.03s system 0% cpu 24:40.42 total
  - 0.7: ./ilri/check-duplicates.py -i ~/Downloads/2021-10-03-non-IWMI-publications.cs 0.12s user 0.03s system 0% cpu 24:29.15 total
  - 0.8: ./ilri/check-duplicates.py -i ~/Downloads/2021-10-03-non-IWMI-publications.cs 0.09s user 0.03s system 0% cpu 25:44.13 total
Some minor updates to csv-metadata-quality
- Fix two issues with regular expressions in the duplicate items and experimental language checks
- Add a check for items that have a DOI listed in their citation, but are missing a standalone DOI field
Then I ran this new version of csv-metadata-quality on an export of IWMI’s community, minus some fields I don’t want to check:

$ csvcut -C 'dc.date.accessioned,dc.date.accessioned[],dc.date.accessioned[en_US],dc.date.available,dc.date.available[],dc.date.available[en_US],dcterms.issued[en_US],dcterms.issued[],dcterms.issued,dc.description.provenance[en],dc.description.provenance[en_US],dc.identifier.uri,dc.identifier.uri[],dc.identifier.uri[en_US],dcterms.abstract[en_US],dcterms.bibliographicCitation[en_US],collection' ~/Downloads/iwmi.csv > /tmp/iwmi-to-check.csv
$ csv-metadata-quality -i /tmp/iwmi-to-check.csv -o /tmp/iwmi.csv | tee /tmp/out.log
$ xsv split -s 2000 /tmp /tmp/iwmi.csv

I noticed each CSV only had 10 or 20 corrections, mostly that none of the duplicate metadata values were removed in the CSVs…
- I cut a subset of the fields from the main CSV and tried again, but DSpace said “no changes detected”
- The duplicates are definitely removed from the CSV, but DSpace doesn’t detect them
- I realized this is an issue I’ve had before, but forgot because I usually use csv-metadata-quality for new items, not ones already inside DSpace!
- I found a comment on thread on the dspace-tech mailing list from helix84 in 2015 (“No changes were detected” when importing metadata via XMLUI") where he says:

It’s very likely that multiple values in a single field are being compared as an unordered set rather than an ordered list. Try doing it in two imports. In first import, remove all authors. In second import, add them in the new order.

Shit, so that’s worth looking into…

2021-10-07

I decided to upload the cleaned IWMI community by moving the cleaned metadata field from dcterms.subject[en_US] to dcterms.subject[en_Fu] temporarily, uploading them, then moving them back, and uploading again
- I started by copying just a handful of fields from the iwmi.csv community export:

$ csvcut -c 'id,cg.contributor.affiliation[en_US],cg.coverage.country[en_US],cg.coverage.iso3166-alpha2[en_US],cg.coverage.subregion[en_US],cg.identifier.doi[en_US],cg.identifier.iwmilibrary[en_US],cg.identifier.url[en_US],cg.isijournal[en_US],cg.issn[en_US],cg.river.basin[en_US],dc.contributor.author[en_US],dcterms.subject[en_US]' ~/Downloads/iwmi.csv > /tmp/iwmi-duplicate-metadata.csv
# Copy and blank columns in OpenRefine
$ csv-metadata-quality -i ~/Downloads/2021-10-07-IWMI-duplicate-metadata-csv.csv -o /tmp/iwmi-duplicates-cleaned.csv | tee /tmp/out.log
$ xsv split -s 2000 /tmp /tmp/iwmi-duplicates-cleaned.csv

It takes a few hours per 2,000 items because DSpace processes them so slowly… sigh…

2021-10-08

I decided to update these records in PostgreSQL instead of via several CSV batches, as there were several others to normalize too:

cgspace=# SELECT DISTINCT text_lang, count(text_lang) FROM metadatavalue WHERE dspace_object_id IN (SELECT uuid FROM item) GROUP BY text_lang ORDER BY count DESC;
 text_lang |  count  
-----------+---------
 en_US     | 2603711
 en_Fu     |  115568
 en        |    8818
           |    5286
 fr        |       2
 vn        |       2
           |       0
(7 rows)
cgspace=# BEGIN;
cgspace=# UPDATE metadatavalue SET text_lang='en_US' WHERE dspace_object_id IN (SELECT uuid FROM item) AND text_lang IN ('en_Fu', 'en', '');
UPDATE 129673
cgspace=# COMMIT;

So all this effort to remove ~400 duplicate metadata values in the IWMI community hmmm:

$ grep -c 'Removing duplicate value' /tmp/out.log
391

I tried to export ILRI’s community, but ran into the export bug (DS-4211)
- After applying the patch on my local instance I was able to export, but found many duplicate items in the CSV (as I also noticed in 2021-02):

$ csvcut -c id /tmp/ilri-duplicate-metadata.csv | sed '1d' | wc -l 
32070
$ csvcut -c id /tmp/ilri-duplicate-metadata.csv | sort -u | sed '1d' | wc -l
19315

It seems there are only about 200 duplicate values in this subset of fields in ILRI’s community:

$ grep -c 'Removing duplicate value' /tmp/out.log
220

I found a cool way to select only the items with corrections
- First, extract a handful of fields from the CSV with csvcut
- Second, clean the CSV with csv-metadata-quality
- Third, rename the columns to something obvious in the cleaned CSV
- Fourth, use csvjoin to merge the cleaned file with the original

$ csvcut -c 'id,cg.contributor.affiliation[en_US],cg.coverage.country[en_US],cg.coverage.iso3166-alpha2[en_US],cg.coverage.subregion[en_US],cg.identifier.doi[en_US],cg.identifier.url[en_US],cg.isijournal[en_US],cg.issn[en_US],dc.contributor.author[en_US],dcterms.subject[en_US]' /tmp/ilri.csv | csvsort | uniq > /tmp/ilri-deduplicated-items.csv
$ csv-metadata-quality -i /tmp/ilri-deduplicated-items.csv -o /tmp/ilri-deduplicated-items-cleaned.csv | tee /tmp/out.log
$ sed -i -e '1s/en_US/en_Fu/g' /tmp/ilri-deduplicated-items-cleaned.csv
$ csvjoin -c id /tmp/ilri-deduplicated-items.csv /tmp/ilri-deduplicated-items-cleaned.csv > /tmp/ilri-deduplicated-items-cleaned-joined.csv

Then I imported the file into OpenRefine and used a custom text facet with a GREL like this to identify the rows with changes:

if(cells['dcterms.subject[en_US]'].value == cells['dcterms.subject[en_Fu]'].value,"same","different")

For these rows I starred them and then blanked out the original field so DSpace would see it as a removal, and add the new column
- After these are uploaded I will normalize the text_lang fields in PostgreSQL again
I did the same for CIAT but there were over 7,000 duplicate metadata values! Hard to believe:

$ grep -c 'Removing duplicate value' /tmp/out.log
7720

I applied these to the CIAT community, so in total that’s over 8,000 duplicate metadata values removed in a handful of fields…

2021-10-09

I did similar metadata cleanups for CCAFS and IITA too, but there were only a few hundred duplicates there
Also of note, there are some other fixes too, for example in IITA’s community:

$ grep -c -E '(Fixing|Removing) (duplicate|excessive|invalid)' /tmp/out.log
249

I ran a full Discovery re-indexing on CGSpace
Then I exported all of CGSpace and extracted the ISSNs and ISBNs:

$ csvcut -c 'id,cg.issn[en_US],dc.identifier.issn[en_US],cg.isbn[en_US],dc.identifier.isbn[en_US]' /tmp/cgspace.csv > /tmp/cgspace-issn-isbn.csv

I did cleanups on about seventy items with invalid and mixed ISSNs/ISBNs