2018-04-10 Statusmeeting

As mentioned in previous updates last year, over the past months we have been working on our full-text indexing process and the search interface. Last week we opened access to the new version of our full-text search application "Archives de l'internet Labs", which is now integrated with the main interface. It has been updated with new graphics and new functions, in particular the grouping of identical URLs in search results. We have to thank Toke for his help with this, as we had real problems regarding performance when we applied this grouping to searches with large numbers of results - Toke advised us to deactivate in Solr the count of the number of groups generated (which we used to calculate the number of pages of results) and this made a huge improvement. We also tidied up the code of the application with a view to allowing its use by other institutions who use warc-indexer.

On the indexing side, the main objective of the project was to index our daily news crawl since its creation at the end of 2010; we originally aimed to index the period up to the end of 2016 but we were able to extend this to the end of 2017. This means we were also able to treat WARCs containing revisit records, which we have been producing since changing to Heritrix 3 last year. We worked with the community on the latest version of warc-indexer, in particular to define collection names based on W/ARC filenames, and therefore on the harvest definitions in NAS. The news crawl represented an increase in the amount of data indexed compared to the collections previously indexed (around 13 TB, compared to around 2.5 TB) and also in terms of the final index size, which roughly doubled to around 2.4 TB. To handle with this we also put in place a new infrastructure, and the performances are much better than out previous prototype, though we will aim to continue work on the configuration in future developments.

Work on the research project that is using the news crawl to study neologisms is ongoing. We are working with the research team to see if some of the analyses that they apply, such as Named Entity Recognition and Topic Modelling, can be included in our indexing and search systems.