Lexica

Extract lexicon features from text. Available lexica are:

• LIWC 2007
• LIWC 2015
• NRC EmoLex v0.92
• Agency and Authority Connotation frame

Running the code

Import it into python using from util import *

Select the dictionary you want to use:

• lex = liwc.parse_liwc("2007") for LIWC 2007
• lex = liwc.parse_liwc("2015") for the 2015 version
• lex = nrc.parse_emolex() for the NRC EmoLex
• lex = nrc.parse_optpess() for the NRC Optimism/Pessimism lexicon (weighted)
• lex = conno.parse_connotation("agency") for the Agency connotation frame
• lex = conno.parse_connotation("authority") for the Authority connotation frame

Optionally only select certain categories: lex = liwc.parse_liwc("2007",whitelist=["posemo","negemo"])

Extract features using the extract function:

• extract(lex,"this is a text"): will return a dictionary of {category: percentage}
• extract(lex,"this is a text",percentage=False): will return a dictionary of {category: raw word count}

If lex is a weighted lexicon, each matched word is multiplied by it's category weight

Example:

In [1]: from util import *

In [2]: lex = nrc.parse_emolex()

In [3]: extract(lex,"This is a story about a girl named lucky",False)
Out[3]: {'anticipation': 1, 'joy': 2, 'positive': 2, 'surprise': 2}

In [4]: lex = liwc.parse_liwc("2015")

In [5]: extract(lex,"This is a story")
Out[5]:
{'article': 0.25,
 'auxverb': 0.25,
 'focuspresent': 0.25,
 'function': 0.75,
 'ipron': 0.25,
 'pronoun': 0.25,
 'social': 0.25,
 'verb': 0.25}
       

Connotation frames

You can also work at a verb-level, instead of the word-level. Specifically, the connotation frames of agency and authority only work with verbs.

Use extractVerbs to only count verbs towards connotation frames. Verbs are detected using the SpaCy POS tagger and lemmatizer.

Note that the results will be different since some verbs are nouns/adjectives sometimes.

Example:

In [1]: from util import *

In [2]: lex = conno.parse_connotation()

In [3]: extractVerbs(lex,"They grabbed and pulled the cool machine")
Out[3]: {'agency_pos': 1.0}

In [4]: extract(lex,"They grabbed and pulled the cool machine")
Out[4]: {'agency_neg': 0.14285714285714285}