Add types/components/resources related to sentiment analysis

[GoogleCodeExporter]

Currently, dkpro-core misses types, components and resources related to 
sentiment analysis. Since, there are people working in different projects in 
this area, I believe we should have a type system for that, along with some 
resources and even some components to be reused by everyone.

Original issue reported on code.google.com by pedrobss...@gmail.com on 11 Nov 2013 at 11:46

[GoogleCodeExporter]

This should not be a module in DKPro Core. It should better be an independent 
project and moved to DKPro Core if the following conditions are true:
- Others want to contribute and reuse
- There exist 3rd-party libraries for sentiment analysis

Original comment by nico.erbs@gmail.com on 11 Nov 2013 at 2:30

[GoogleCodeExporter]

I have been thinking about DKPro Sentiment for a while.
Is there a general model beyond dictionary-based methods that the project could 
model?

Original comment by torsten....@gmail.com on 11 Nov 2013 at 3:24

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

We couldn't think of any more generic model but we haven't invested too much 
time. I think it is the right time to start such a project and invite others 
with more experience about sentiment analysis to contribute.

Original comment by nico.erbs@gmail.com on 11 Nov 2013 at 4:56

[GoogleCodeExporter]

FTR: there is a demo module in DKPro-TC which does sentiment (polarity) 
detection.

Original comment by daxenber...@gmail.com on 12 Nov 2013 at 8:21

[GoogleCodeExporter]

It appears, there will be sentiment analysis in the next CoreNLP version:

http://nlp.stanford.edu/sentiment/code.html

Code for Deeply Moving: Deep Learning for Sentiment Analysis

The original code was written in Matlab. Due to the strong interest in this 
work we decided to re-write the entire algorithm in Java for easier and more 
scalable use without license restrictions. In this alpha release we focus on 
the testing portion of the code that gives the published results. The code 
already includes the training procedures. At the end of October we will update 
this package with further documentation and faster training code.

This model will be part of the next Stanford CoreNLP release and the below code 
includes the entire CoreNLP package. It includes the model source code, as well 
as the parser and sentence splitter among other functionality.

Download alpha version: stanford-corenlp-full-2013-10-21.zip (211MB)

You can run this code with our trained model on text files with the following 
command:

java -cp "*" -mx5g edu.stanford.nlp.sentiment.SentimentPipeline -file foo.txt

Original comment by richard.eckart on 12 Nov 2013 at 9:13

[GoogleCodeExporter]

Hi ..I am looking for a trusted and easier way to do text-based sentiment 
analysis with Matlab .. and what I expect is that in the result of each row of 
my matrix (each row includes a text file)  show me positive or negative or even 
if there is any professional tools or code which let me to have Author 
personality could be very very useful.
thanks in advance

Original comment by asan...@gmail.com on 18 May 2014 at 12:47

[GoogleCodeExporter]

@asana86: Please do check out page from Stanford mentioned above. We plan do 
integrate these components into DKPro Core at some point, but currently we did 
not do so. Mind that the everything below the link in the comment above has 
been copied from the linked page for our convenience. We DKPro Core people did 
*not* create or contribute to this sentiment analysis code.

Original comment by richard.eckart on 18 May 2014 at 1:21

[GoogleCodeExporter]

Hi, I've implemented a DKPro annotator for sentiment analysis on sentence level 
with 5-scores granularity of sentiment (wrapper of Stanford CoreNLP 3.4 
sentiment analyzer, http://www-nlp.stanford.edu/sentiment/code.html). Pros: Out 
of box functionality, pre-trained models, good results. Cons: Currently it runs 
the whole Stanford pipeline (tokenization, POS, parsing), so it's slow; better 
seamless integration would require mapping existing DKPro annotations (POS, 
parse tree) to Stanford annotations, not sure how demanding that is. What will 
be the best way to integrate it to the next release of DKPro core?

Original comment by ivan.hab...@gmail.com on 17 Jul 2014 at 11:57

[GoogleCodeExporter]

I believe we have code somewhere to transform DKPro Core annotations into the 
Stanford structures. I need to dig around a bit to find those.

I would suggest to add the component as is to DKPro Core GPL trunk into the 
CoreNLP module and then we'll look what needs to be done to improve it. I'd 
also suggest to open a separate issue for the CoreNLP sentiment integration - 
this issue is just about adding annotation types for sentiment analysis which 
should be independent of a particular sentiment analysis implementation. We 
should probably add a new api.sentiment module for these types.

Original comment by richard.eckart on 17 Jul 2014 at 12:02

[GoogleCodeExporter]

Ok, I opened a separate bug for it #424 and submitted two patches (one for 
dkpro-asl with typesystem, one for dkpro-gpl with wrapper for Stanford 
sentiment analyzer).

Original comment by ivan.hab...@gmail.com on 18 Jul 2014 at 11:59