doctalk: Dialogue agent handling queries about a given text document
Project description
doctalk
Python-based summary and keyword extractor and question answering system with optional BERT-based post-processing filter and spoken output
USAGE:
python3 -i
>>> from doctalk.talk import run_with
>>> run_with(fname)
>>> from doctalk.think import reason_with
>>> reason_with(fname)
This activates dialog about document in <fname>.txt
with questions in <fname>_quests.txt
See some examples at :
https://github.com/ptarau/pytalk , where, after installing the system, you can run
python3 -i tests.py
>>> go()
>>> tgo()
To run the system one will neet to start the Stanford Corenlp Server, listening on port 9000 with all annotators in params.py started, i.e., with something like:
java -mx16g -cp "stanford-corenlp-full-2018-10-05/*" edu.stanford.nlp.pipeline.StanfordCoreNLPServer
-preload tokenize,ssplit,pos,lemma,depparse,natlog,openie
To play with various parameter settings, edit the doctalk/params.py
file.
Please see the install hints for stanfordnlp, that might involve torch binaries, and require anaconda on some systems.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.