matheuscas/fuzzy_opinion_mining

Propor e avaliar o desempenho de um sistema de mineração de opiniões baseado na lógica fuzzy

fuzzy_opinion_mining

Propor e avaliar o desempenho de um sistema de mineração de opiniões baseado na lógica fuzzy

fuzzy_opinion_mining's Issues

Add Amazon Corpus

From this source: http://patty.isti.cnr.it/~baccianella/reviewdata/index.php?download

Pass map and reduce methods to new classes

They weren't re-factored to new class model 2 .

Create index files to the polarities files

This is useful to relate each line of polarities to its origin file. Hence, it is possible to see from which file those polarities came from.

Create a method in util to parse a text from a given document from collection

It should receive a Id (or name) and access direct to model or receive model as well as parameter and with that the method will parse the text. Again the user should pick which parser will be used.

Create a method in util to parse a given text given for debug purpose

Let the user pick which parser he will use (perceptron, standford or pattern)

Words tags are different from parsed text tags

Ngrams has been extracted using a method different from that are parsing the text, causing discrepancies between what is really is in the text and what has been used to predict polarities.

Using textblob, ngrams has been gathered using "tags" property and the text tagging has been done using the "parse" method from the chosen parser.

Add more TripAdvisor corpora

Only onde part of TripAdvisor corpora was added. Add the other part.

Parameterize POS Taggers

In the new classes only the Averaged Perceptron POS Tagger (APPT) is been used. It was perceived that in some simulations there was some (little, less than 1%) decrease of performance using this tagger, despite of some researches says that this tagger has better performance than Stanford Tagger and Pattern Clips lib.

So, BaseModel methods of pre-processing should allow the user choose what tagger he will use: APPT, Stanford or Pattern