Marco Fossati comments

Results 34 comments of


                                            Marco Fossati

Entity Linker

Currently supporting the [Dandelion Entity Extraction API](https://dandelion.eu/docs/api/datatxt/nex/v1/) via 1db502eae9f6a1a0b599b4ef3ebca9faff18c855 .

Ensure proper encoding from scraped sources

Should be a minor issue for English, but not for other languages.

Problem on windows

Thanks @moissinac for the report. We will investigate the issue.

Improve documentation

Points 3,4 and 5 above are to be implemented yet.

More comments. These are the minimal requirements to make the documentation more usable: 1. do not generate non-browsable lists, e.g., https://www.mediawiki.org/wiki/StrepHit#Subpackages 2. avoid empty sections, e.g., https://www.mediawiki.org/wiki/StrepHit#Module_contents

Investigate undocumented CrowdFlower API request to enable internal and external channels

This seems to be bound to the **launch** of a job: ``` curl -X POST -d "channels[0]=on_demand&channels[1]=cf_internal&debit[units_count]={100}" https://api.crowdflower.com/v1/jobs/{job_id}/orders.json?key={api_key} ```

Provide reference URLs that point to human-readable resources

There seems to be no easy way to deal with point 2 above

Inconsistent CrowdFlower API behavior when converting gold

The support team replied that there is no fix for this. Hence, we will use the platform UI for test questions creation, instead of uploading them directly.

Parallel processing crashes on Mac

Also with `python -m strephit commons entity_linking`. Full stack trace attached. [stacktrace02.txt](https://github.com/Wikidata/StrepHit/files/314193/stacktrace02.txt)

Parallel processing crashes on Mac

And with `python -m strephit commons serialize` Full stack trace attached. [stacktrace03.txt](https://github.com/Wikidata/StrepHit/files/314241/stacktrace03.txt)