search

dbpedia / neural-qa

📚 A Neural QA Model for DBpedia using Neural SPARQL Machines.
MIT License
85 stars 20 forks source link

sep_dot encoding problem #28

Closed BaiBlanc closed 4 years ago

BaiBlanc commented 4 years ago

encode() method in generator_utils.py will potentially encode the abbreviation dot into a token, which would affect the dataset and the final performance, for example:

select ?x where { dbr_Bandon_A.F.C. dbo_ceo ?x }

will be encoded into 'select var_x where brack_open dbr_Bandon_A.F.C sep_dot dbo_ceo var_x brack_close' but not 'select var_x where brack_open dbr_Bandon_A.F.C. dbo_ceo var_x brack_close'

divyanshugit commented 4 years ago

Hey, I would love to work on this issue. Can you suggest me ways to get started ?

BaiBlanc commented 4 years ago

Hi @divyanshugit , first of all, you could get start with the NSpM model, try to follow the instruction and run the model if you haven't got any idea. To solve this particular issue, you may need some knowledge of query language Sparql, at least you should know the use of a sep_dot in the SPARQL query, and why it shouldn't be there in our case. Once you make this clear, I suggest you to see my commit, in particular the modifications in generator_utils.py, This may give you some idea on this issue.