ptravers / centi Goto Github PK

Add rule described as sentiment propagation whereby compositions in which the polarity of a neutral constituent is overridden by that of a non-neutral constituent ({(+)(N)} → (+); {(-)(N)} → (-)).

In layman's terms:
neutral + positive = positive
neutral + negative = negative

Situations whereby all sentiment is neutral will be propagated as neutral. Situations whereby sentiment is conflicting will be denoted as mixed (using the mixed constant).

It's a common thing in Python to name pseudo-constants in caps lock to indicate that they are constants. This helps differentiate them from local variables.

Source: PEP 8:

Constants are usually defined on a module level and written in all capital letters with underscores separating words. Examples include MAX_OVERFLOW and TOTAL .

Thoughts @ptravers, @alastairgillespie?

The current word to sentiment and word to modifier methods are case senitive. This means "like" is positive (+) but "Like" is neutral (-).

This should be a five second fix.

W.r.t. what is asked in the report.

Add function centi.get_sentiment in new file sentiment.py that maps a word to a triple containing the following two values:

• Sentiment [-1|0|1]: Whether the word is negative, neutral or positive.
• Modifier [P|N|¬|=]: Whether the word should override with positive/negative, invert the polarity or make no change

As discussed the function will start by assuming all values have no modifier [=]. After basic functionality is implemented I can look at obtaining a corpus of sentiment inverting words.

Creation of constants to allow for more readable code. Inclusive of standard functions.

Det:(Det|DP) >> Head:N

no crime = -1
too much wealth = -1

Labels should contain a head so prehead and posthead rules can be run against the current tree

Add class to hold a node's sentiment, modifier and label. Other properties and methods may be added later.

The paper describes words that can modify the meaning of entire phrases, such a "deficiency" [-]. These words are not just very positive or very negative words.

I have been searching through the citations of the paper as well as online. It seems we will be very limited as:

• Lack of time (many corpora require written permission)
• Lack of status (many corpora require proof of Phd and to be working at a uni)

None the less I think I at least have found a negative dataset I can process to make a used list of negative modifiers.

A SUPR with the negation tag should convert any SUB to the opposite of their current status