Skip to content

Latest commit

 

History

History
31 lines (26 loc) · 3.02 KB

Readme.md

File metadata and controls

31 lines (26 loc) · 3.02 KB
Raw

Python application saving extracted named entities of crawled news text corpora into a database:

ner

Example

Typical mortgage payment could be 30% higher in 5 years, Bank of Canada says the average price figure can be misleading because it is easily skewed by sales in large expensive markets such as Toronto and Vancouver. So it calculates another number, known as the House Price Index (HPI), which it says is a better gauge of the market because it adjusts for the volume and types of housing. Heaps Estrin, president and CEO of Toronto-based real estate, says the slowdown in Toronto is mostly happening in the suburbs, where prices jumped the most during the pandemic as buyers sought more space

Entity Label doc_id
Bank of Canada ORG 3
Toronto LOC 3
Vancouver LOC 3
HPI ORG 3
Heaps Estrin PER 3
Toronto LOC 3

Our Work

  • First, we implement a crawler.NewsCrawler class, crawling news from CBC News website of different categories
  • Then, we use one of the three designed NER models, namely ners.BertNER, ners.LstmNER, and ners.SpacyNER (sorted here from higher to lower accuracy performance), to extract named entities of each crawled news text one by one in main.py
    • In ners.BertNER, we use the pretrained model dslim/bert-base-NER of transformers library, which is a fine-tuned version of BERT for NER task
    • In ners.LstmNER, we use our custom trained model LSTM under lstm_ner.lstm_ner, that is further explained in lstm_ner/Readme.md
    • In ners.SpacyNER, we use Spacy pipeline for NER
  • After deriving named entities out of a crawled text (using previous two steps), finally, we use database.NewsDatabase class, implementing logics to create and insert into two tables of interest, document table (with columns id, title, text) and named-entity table (with columns id, entity, start_char, end_char, label, doc_id, the entity column of which would be one of the three inferred types, namely PER for Person, ORG for Organization, and LOC for Location), by connecting Python to PostgreSQL database using psycopg2 connector library

Run

In a console with an activated python environment, under current directory:

  • Enter pip install -r requirements.txt to install dependencies
    • As prerequisites, make sure to have PostgreSQL server installed (equipped with the "postgres" user having password "postgres"), Selenium webdriver added to the path, and spacy data downloaded
  • Enter python main.py to start crawling news text corpora, and put their extracted named entities into the database