prosecco

Simple, extendable nlp engine that can extract data based on provided conditions.


Keywords
language-processing, nlp, python, text-mining
License
MIT
Install
pip install prosecco==0.0.6

Documentation

prosecco

GitHub pypi GitHub commits since tagged version GitHub last commit

Description

NLP engine with text extraction capabilities that can be easily extended to desired needs.

Can be used to build chat bots, question answer machines (see example/qa.py), text converters.

Extract words or even whole sentences in ordered manner.
Get position of found text.
Use Condition class and mark data using regex or string comparasion.
Extend each part of it in easy manner. ( see example/custom_condition_class.py).

Install

pip install prosecco

Usage

Basic

example/basic.py

from prosecco import Prosecco, Condition, EnglishWordNormalizer

# Read wikipedia https://en.wikipedia.org/wiki/Superhero
with open("superhero.txt") as f:
    text = f.read()

# 1. Create conditions with hero names
conditions = [
    Condition(lemma_type="hero|dc", compare=["batman", "superman", "wonder woman"], lower=True),
    Condition(lemma_type="hero|marvel", normalizer=EnglishWordNormalizer(),
              compare=["spiderman", "iron man", "black panther"], lower=True)
]
# 2. Create prosecco
p = Prosecco(conditions=conditions)
# 3. Let's drink and print output
p.drink(text, progress=True)
lemmas  = set(p.get_lemmas(type="hero"))
print(" ".join(map(str, lemmas)))

Output

Batman[hero|dc][start:1089] Wonder Woman[hero|dc][start:2100] Iron Man[hero|marvel][start:2184] Superman[hero|dc][start:2070] Spider-Man[hero|marvel][start:2080] Black Panther[hero|marvel][start:17690]

Advanced

example/advanced.py

from prosecco import *

text = """Chrząszcz brzmi w trzcinie w Szczebrzeszynie.
Ząb zupa zębowa, dąb zupa dębowa.
Gdzie Rzym, gdzie Krym. W Pacanowie kozy kują.
Tak, jeśli mam szczęśliwy być, to w Gdańsku muszę żyć! 
"""

# 1. Create condition with city names
cities = ["szczebrzeszyn", "pacanow", "gdansk", "rzym", "krym"]
animals = ["koz", "chrzaszcz"]
# 2. Normalizer to remove polish specific charset
n = CharsetNormalizer(Charset.PL_EN)
# 3. Stemmer to remove suffix
s = SuffixStemmer(language="pl")
# 4. Conditions for city and animal
city_condition = Condition(lemma_type="city", compare=cities, normalizer=n, stemmer=s, lower=True)
animal_condition = Condition(lemma_type="animal", compare=animals, normalizer=n, stemmer=s, lower=True)
conditions = [city_condition, animal_condition]
# 5. Create tokenizer for polish charset
tokenizer = LanguageTokenizer(Charset.PL)
# 6. Get list of tokens
tokens = tokenizer.tokenize(text)
# 7. Create visitor with conditions provided in step 1
visitor = Visitor(conditions=conditions)
# 8. Parse tokens based on visitor conditions
lexer = Lexer(tokens=tokens, visitor=visitor)
# 9. Get list of lemmas
lemmas = lexer.lex()
# 10. filter found cities and print output
found = filter(lambda l: l.type == "city", lemmas)
print(" ".join(map(str, found)))
# 11. filter found anumals and print output
found = list(filter(lambda l: l.type == "animal", lemmas))
print(" ".join(map(str, found)))
# 12. print exact words from text
for l in list(found):
    print(text[l.start:l.start+len(l.sentence)])

Output

Szczebrzeszynie[city][start:29] Rzym[city][start:86] Krym[city][start:98] Pacanowie[city][start:106] Gdańsku[city][start:163]
Chrząszcz[animal][start:0] kozy[animal][start:116]
Chrząszcz
kozy

QA ( question answer machine )

example/qa.py

from datetime import datetime
from prosecco import Prosecco, Condition, EnglishWordNormalizer, SuffixStemmer


messages = """Whats the time ?
How long boil egg?
100 miles to km
30,3 celsius to farenheit"""

# create conditions
question = ("what", "whats", "how")
measure = ("celsius", "farenheit", "mile", "km", "kilometer", "time", "long")
cooking = ("boil","cook", "fry")
food = ("egg",)
conditions = [
    Condition(lemma_type="question", compare=question, normalizer=EnglishWordNormalizer(), lower=True),
    Condition(lemma_type="measure", compare=measure,
              normalizer=EnglishWordNormalizer(),
              stemmer=SuffixStemmer(language="en"),
              lower=True),
    Condition(lemma_type="cooking", compare=cooking, normalizer=EnglishWordNormalizer(), lower=True),
    Condition(lemma_type="food", compare=food,
              normalizer=EnglishWordNormalizer(),
              stemmer=SuffixStemmer(language="en"),
              lower=True),
    Condition(lemma_type="number", compare=r"\d+([\.\,]\d+)?", regex=True, until_character=" "),
]

def printer(data):
    print("Robot : ", data)

# time condition
def resolve_time(p):
    printer(datetime.now())

# cooking condition
def resolve_cooking(p):
    if check_condition(p.get_lemmas("cooking|food"), ["boil", "egg"]):
        printer("""
Hard for 9-15 minutes.
Soft for 6-8 minutes.""")
        return True

def resolve_measure(p):
    measures = p.get_lemmas("measure")
    fr = measures[0]
    to = measures[1]
    numbers = p.get_lemmas("number")
    if len(numbers) == 0:
        printer("No number for conversion provided")
        return True
    value = float(numbers[0].sentence.replace(",", "."))
    if fr.condition == "mile" and to.condition == "km":
        printer(value / 0.62137119)
        return True
    elif fr.condition == "km" and to.condition == "mile":
        printer(value * 0.62137119)
        return True
    elif fr.condition == "celsius" and to.condition == "farenheit":
        print(9/5 * value + 32)
        return True
    elif fr.condition == "farenheit" and to.condition == "celsius":
        print((value - 32) * 5/9)
        return True
    return False

def check_condition(lemmas, conditions):
    for l in lemmas:
        for c in conditions:
            if l.condition == c:
                conditions.remove(c)
    return len(conditions) == 0

def resolve(p, m):
    if len(p.get_lemmas("question")) > 0:
        if check_condition(p.get_lemmas("measure"), ["time"]):
            resolve_time(p)
            return True
        elif len(p.get_lemmas("cooking")) > 0:
            return resolve_cooking(p)
    elif len(p.get_lemmas("measure")) > 0:
        return resolve_measure(p)
    return False

for m in messages.split('\n'):
    print("Question : ", m)
    p = Prosecco(conditions=conditions)
    p.drink(m)
    if not resolve(p, m):
        print("Unsupported resolver : ", p.lemmas)

Output

Question :  Whats the time ?
Robot :  2019-08-13 20:38:06.948720
Question :  How long boil egg?
Robot :  
Hard for 9-15 minutes.
Soft for 6-8 minutes.
Question :  100 miles to km
Robot :  160.93440057946685
Question :  30,3 celsius to farenheit
86.53999999999999