locationtagger 0.0.1

Detect & Extract locations from text or URL and find relationships among locations

locationtagger

version 0.0.1

Detect and extract locations (Countries, Regions/States & Cities) from text or URL. Also, find relationships among countries, regions & cities.

---

About Project

In the field of Natural Lauguage Processing, many algorithms have been derived for different types of syntactic & semantic analysis of the textual data. NER (Named Entity Recognition) is one of the best & frequently needed tasks in real-world problems of text mining that follows some grammer-based rules & statistical modelling approaches. An entity extracted from NER can be a name of person, place, organization or product. locationtagger is a further process of tagging & filter out place names (locations) amongst all the entities found with NER.

Approach followed is given below in the picture;

https://github.com/kaushiksoni10/locationtagger/blob/master/locationtagger/data/diagram.jpg?raw=true

---

Install and Setup

(Environment: python >= 3.5)

Install the package using pip -

pip install locationtagger

But before we install the package, we need to install some useful libraries given below,

nltk

spacy

newspaper3k

pycountry

After installing these packages, there are some important nltk & spacy modules that need to be downloaded using commands given in /locationtagger/bin/locationtagger-nltk-spacy on IPython shell or Jupyter notebook.

---

Usage

After proper installation of the package, import the module and give some text/URL as input;

Text as input

import locationtagger

text = "Unlike India and Japan, A winter weather advisory remains in effect through 5 PM along and east of a line from Blue Earth, to Red Wing line in Minnesota and continuing to along an Ellsworth, to Menomonie, and Chippewa Falls line in Wisconsin."

entities = locationtagger.find_locations(text = text)

Now we can grab all the place names present in above text,

entities.countries

['India', 'Japan']

entities.regions

['Minnesota', 'Wisconsin']

entities.cities

['Ellsworth', 'Red Wing', 'Blue Earth', 'Chippewa Falls', 'Menomonie']

Apart from above places extracted from the text, we can also find the countries where these extracted cities, regions belong to,

entities.country_regions

{'United States': ['Minnesota', 'Wisconsin']}

entities.country_cities

{'United States': ['Ellsworth', 'Red Wing', 'Blue Earth', 'Chippewa Falls', 'Menomonie']}

Since "United States" is a country but not present in the text still came from the relations to the cities & regions present in the text, we can find it in other_countries,

entities.other_countries

['United States']

If we are really serious about the cities we got in the text we can find which regions in the world it may fall in,

entities.region_cities

{'Maine': ['Ellsworth'], 'Minnesota': ['Red Wing', 'Blue Earth'], 'Wisconsin': ['Ellsworth', 'Chippewa Falls', 'Menomonie'], 'Pennsylvania': ['Ellsworth'], 'Michigan': ['Ellsworth'], 'Illinois': ['Ellsworth'], 'Kansas': ['Ellsworth'], 'Iowa': ['Ellsworth']}

And obviously, we'll put these regions in other_regions since they are not present in original text,

entities.other_regions

['Maine', 'Minnesota', 'Wisconsin', 'Pennsylvania', 'Michigan', 'Illinois', 'Kansas', 'Iowa']

Whatever words nltk & spacy both grabbed from the original text as named entity , most of them are stored in cities, regions & countries. But the remaining words (not recognized as place name) will be stored in other.

entities.other

['winter', 'PM', 'Chippewa']

URL as Input

Similarly, It can grab places from urls too,

URL = 'https://edition.cnn.com/2020/01/14/americas/staggering-number-of-human-rights-defenders-killed-in-colombia-the-un-says/index.html'
entities2 = locationtagger.find_locations(url = URL)

outputs we get: countries;

entities2.countries

['Switzerland', 'Colombia']

regions;

entities2.regions

['Geneva']

cities;

entities2.cities

['Geneva', 'Colombia']

Now, if we want to check how many times a place has been mentioned or most common places which have been mentioned in the whole page of the URL, we can have an idea about what location that page is talking about;

hence, most commonly mentioned countries;

entities2.country_mentions

[('Colombia', 3), ('Switzerland', 1), ('United States', 1), ('Mexico', 1)]

and most commonly mentioned cities;

entities2.city_mentions

[('Colombia', 3), ('Geneva', 1)]

---

Credits

locationtagger uses data from following source for country, region & city lookups,

GEOLITE2 free downloadable database

Apart from famous nlp libraries NLTK & spacy, locationtagger uses following very useful libraries;

pycountry

newspaper3k