resolved token vs. character indexing issues

[davidberenstein1957]

Added character indexing, instead of the faulty, token based indexing. Note that doc.[index_x, index_y] refers to token, however, entity-fishing works with characters.
Additionally, now you also assign the entity_fishing variables to completely different spans, which will also likely overlap with one another.

[davidberenstein1957]

@Lucaterre I also wrote some code, to be able to include entities, and Wikipedia matches from the entity-fishing API too, however, since this is a more elaborate feature, I want to get your input on this before creating a PR.

[davidberenstein1957]

Also, currently the pipeline isn´t optimized for using the .pipe function in spaCy. I can help to add this too.

[Lucaterre]

Hi @davidberenstein1957, thank you very much for this bug fix and for your contributions. 👍

I note that you are right: entity fishing use offsets at character level and not at token level, this is a mistake on my part. 😞

However, your implementation does not fully work (if you test it on the examples in the doc, attribute kb_qid returns None).

I need to added some complements so that Spacy uses the offsets well to recover the spans and update correctly entities in final doc output (replace index access eg. doc[start:end] with doc.char_span() method) and I update the nil_clustering part too. I'll take care of that, after tests checking and merging this PR.

I will create a new release after this, because it is an important fix. thanks again 😄

[davidberenstein1957]

@Lucaterre https://hacktoberfest.com/participation/#pr-mr-details could you add your repo to Hacktoberfest please🤓
I want to get a tree planted in my honor.