[Netarchivesuite-curator] BnF NAS update for February

[geraldine.camile at bnf.fr]

Hello,

Our broad crawl finished on December the 23rd. It represents 2.1 billion 
URLs and 106.46 TB. Due to technical difficulties it took a long time: 11 
weeks (compared to 6 weeks in 2017). The technical difficulties came from 
the new computer architecture and the hardware, the broker and the version 
of NAS, resulting in multiples jobs being created that failed and thus an 
overall slowdown of the crawl. We will discuss this subject during the NAS 
workshop.The percentage of domains that are fully crawled has also 
decreased. We haven't finished anlaysing this collection but we've chosen 
to focus on the websites published for the young.

We have finished analysing all the 2018 crawl reports. Over the year we 
crawled 2.6 billion URLs and 136.15 TB. This is 9 TB less than 2018 due to 
deduplication: we've crawled more in 2019 but with deduplication, 
especially for the broad crawl.  The proportion of the broad crawl 
compared to the selective crawls is still growing: the broad crawl 
represents 78% of the 2018 collections and 70% in 2017. Our collections 
now represent more than 1 Petabyte (1 074.73 TB).

From mid-December to mid-January, we organised an internal workshop to 
improve the harvesting of social media (Facebook, Instagram, Twitter). We 
are able to crawl Facebook with the same Heritrix template we used for 
Twitter. But the quality of the crawl isn't guaranteed: the quality is 
significantly downgraded when there are more than 500 accounts in the job, 
and from one crawl to another the quality is very variable (sometimes we 
crawl nothing). We crawl basically the homepage, the posts and a lot of 
images: it's difficult to know exactly which images we crawl because a lot 
of them are not visible in the Wayback. During the workshop, we tried to 
crawl social media with Umbra. Umbra is very complex to install and 
there's no information exchange between Umbra and NAS: sometimes Umbra 
failed and Heritrix continued to collect. However Umbra allow us to crawl 
the images on Instagram that we couldn't crawl  with Heritrix. We compared 
also the restitution of the web archives in Python Wayback with 
OpenWayback. The restitution is better with Python especially for 
Instagram: the images are displayed while in the OpenWayback we have just 
a white page. For Twitter, the scroll down seems to work in the access 
tool (but we must do more tests). But for Facebook, we hardly noticed any 
change.


Best regards,
The BnF digital legal deposit team
Exposition  Les Nadar, une légende photographique  – jusqu'au 3 février 2019 | François-Mitterrand Avant d'imprimer, pensez à l'environnement. 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://ml.sbforge.org/pipermail/netarchivesuite-curator/attachments/20190204/a9c46453/attachment.html>