You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In lesson 7, "Diktionäre und Sentiment-Analysen" (https://polmine.github.io/UCSSR/sentiments.html#8), you propose a workflow in which context() and partition_bundle() are deployed to identify hits and their surrounding words. The proposed workflow doesn't work if one of the query term's context extends into the next date (or whatever struc is used for grouping the sentiment).
If I see this correctly, the partition_bundle method here splits the context object into partitions, omitting the node (i.e. the query) itself. This doesn't regard boundaries which is why later, when set_names() is called, the s_attributes() methods returns more individual dates than there are partitions in the partition bundle.
After writing this, I remembered that there is a boundary argument for context(). If the general behavior of this workflow is okay, then adding a reasonable boundary (probably speech, which isn't annotated in all corpora or speaker, but there certainly are a lot of situations in which the same speaker might end one debate and start the next) would improve the UCSSR example both functionally as well as substantially as we don't want to measure other speaker's words as context for the sentiment analysis.
The text was updated successfully, but these errors were encountered:
In lesson 7, "Diktionäre und Sentiment-Analysen" (https://polmine.github.io/UCSSR/sentiments.html#8), you propose a workflow in which
context()
andpartition_bundle()
are deployed to identify hits and their surrounding words. The proposed workflow doesn't work if one of the query term's context extends into the next date (or whatever struc is used for grouping the sentiment).If I see this correctly, the partition_bundle method here splits the context object into partitions, omitting the node (i.e. the query) itself. This doesn't regard boundaries which is why later, when
set_names()
is called, thes_attributes()
methods returns more individual dates than there are partitions in the partition bundle.Here is a reproducible example to see the error:
After writing this, I remembered that there is a
boundary
argument forcontext()
. If the general behavior of this workflow is okay, then adding a reasonable boundary (probably speech, which isn't annotated in all corpora or speaker, but there certainly are a lot of situations in which the same speaker might end one debate and start the next) would improve the UCSSR example both functionally as well as substantially as we don't want to measure other speaker's words as context for the sentiment analysis.The text was updated successfully, but these errors were encountered: