doctalk

Python-based summary and keyword extractor and question answering system with optional BERT-based post-processing filter and spoken output

##INSTALL See the requirements.txt file if installing directly. To make an editable package locally, use

pip3 install -e .

To embed in a system as is, from pypi.org use

pip3 install -U doctalk

##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 need 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.

Web App

To use the included ``docbootWeb app, after installingFlask``` and ```waitress```, run the ```app.py``` file in directory ```docbot``` with ```python3 app.py```.

The docbot uses a JSON-based API, documented in doctalk/api.py. In fact, this is the simplest way to integrate the summarizer and the dialog agent into a production system.