Many of us enjoy reading the news and staying late in current events. But every day the number of new stories can be huge.
You probably want to know who is involved in global events, where things are happening globally, and which organizations are talking about. But through each article it takes a lot of time to read – and you are probably busy. This is a place where the identification of the designated entity (NER) can help.
In this article, I show you how to make a news analyst that uses a transformer NER model to extract useful data directly from RSS feed.
Let’s go how it all works.
The table of content
What is the identity of the entity?
The designated entity identification is a device that helps you select the key terms in the text.
It label some parts of a phrase as a specific entity type – such as names, places or dates. It looks like that. Take this phrase:
“Apple CEO Tim Cook met with Executive of Goldman Sex in New York City.
Will identify a good nerve model:
“Tim Cook” – a Person
“apple” – a Organization
“Goldman sex” – a Organization
“New York City” – a Location
Such extraction turns non -imposed text into structural data. This makes it easy to find, count and analyze what is happening in the news.
What’s from hugging facial transformers?
Hugs the facial transformer There is a Library that gives you access to some of the latest NLP models there.
These models are widely trained on data. Instead of starting from the beginning, you have to use a model that already understands the identity of grammar, phrase structure, and entity.
The library provides a simple pipeline() The function that allows you to operate complex work like NER in just a few lines of code. You can find many models already trained on huggingface.co/models.
We will use this project that is fine for English NER.
How to build a news analyst
Let’s create a news analyst. Here is a Google Kolab Notebook If you want to try this hand.
You will need a couple of pairs of packages. Open your terminal or command prompt and run:
pip install feedparser transformers
These libraries will allow you to bring RSS feeds and analyze the text using pre -trained transformer models.
We will use feed parsar to get news articles. Here you bring and print how to summarize CNN’s RSS feed:
import feedparser
rss_url = "
feed = feedparser.parse(rss_url)
for entry in feed.entries(:5): # limit to first 5 articles
print(f"Title: {entry.title}")
print(f"Summary: {entry.summary}\n")
This code draws the title and summary of the latest articles.
Let’s now load a transformer model for NER.
Model DSlim/Bert-Basenr works well for English News Text:
from transformers import pipeline
ner_pipeline = pipeline("ner", model="dslim/bert-base-NER", aggregation_strategy="simple")
aggregation_strategy=”simple” The argument tells the pipeline to integrate a continuous token that forms the same designated entity (such as “Tim Cook”).
This model classifies each word/token in one of the entity categories: per (person), folk (location), org (organization), miscellaneous (miscellaneous), or o (outside any institution).
Give the model some time to download your Kolb notebook or in your local machine.
Let’s connect the Ner model with your feed. The script below draws the title of each article and runs nervous on it.
For the sake of simplicity, we are leaving the summary but if you want to add it, refresh ner_pipeline(title) to ner_pipeline(title+entry.summary).
for entry in feed.entries(:5):
title = entry.title
print(f"\nAnalyzing: {title}")
entities = ner_pipeline(title)
for ent in entities:
print(f"{ent('word')} ({ent('entity_group')})")
It prints organizations found in each article summarizing, which is classified in a variety.
For example, the first piece of text is:
Mexico is ready for retaliation by hurting US farmers
The answer is:
Mexico (LOC)
US (LOC)
There are both places. If we look at other examples, we can see the rating made by the Ner model such as:
iPhone (MISC)
America First (ORG)
India First (ORG)
Swiss (MISC)
Trump (PER)
Once you get out of companies, you can:
Count how often people or organizations appear.
Track the trends over time (for example, how often a particular person appears weekly).
Filters for articles mentioning some locations or companies.
Accuracy in the nerve
Getting structural data from the NER is powerful, but it is not perfect. Models can lose institutions, confuse the terms of misrepresentation, or similar names.
For example, “Amazon” depends on the context, in a sentence and can be tagged as an organization. This is normal because nerve models look for patterns, they really do not “understand” the meaning behind the text.
To get the highest price from Ner, think of it as a first -pass filter instead of the final answer. There are some practical ways to work with its output:
Find samples: When you analyze the trends over time, occasionally mistakes do not matter. For example, keeping track of which companies often appear in headlines give you useful insights even if some mentioned are wrongly classified.
Cross check with known lists or database: If you are monitoring the company’s names or products, compare the nerve results against a reference list of types of types or misunderstandings.
Pair the nerve with another technique: Add it to the analysis of emotions, the matching of key words, or the frequency count to make the data more reliable and viable.
Manually confirm the results of high stake: If your workflow includes decisions with legal, financial, or credibility effects, sample and review the accuracy verification sample and review.
By treating the nerve as a device for texture and filter instead of the absolute source of truth, you can expose trends, make dashboards, and faster the surface insights.
Other use matters
Nair is far ahead of analyzing news headlines. This is a basic tool to remove the meaning of a large amount of non -structured text.
Businesses use it to automatically highlight important details in consumer interactions. For example, support teams can immediately flag customer names, products, serial numbers, or auxiliary tickets and emails. This makes it easier to give priority to quick requests, the right team, and the problems that come up repeatedly without reading each message manually.
Legal firms and researchers relies heavily on the nerves to take a large amount of documents. Legal teams can remove the names of people, companies and places of places, filed in court, and regulatory updates to build maps between the searchable database or entities.
Educational researchers can do the same with scientific papers, accelerate literature studies, and expose samples in thousands of posts.
In finance, Ner Market is a powerful source for intelligence. Analysts use it to detect news, income reports, and analysts’ briefings for the mention of companies, stock tickets, currencies and commodities. By collecting this data, they can detect trends, evaluate the risk exposure, or spot the market -running events far faster than a manual review.
Social media and marketing teams also depend on the NER. By automatically identifying brands, rivals, or influence in tweets and posts, they can monitor the brand’s emotions, detect emerging trends, and react faster to the risk.
Recently. Wherever you are drowning in the text, whether it be consumer feedback, contracts, market reports, or social feeds, Ner can convert this non -structured dirt into structural, viable insights.
Conclusion
What we have made here is a small but powerful news analyst. By adding a direct data source (RSS feed) to facial transformers with a pre -trained nerve model, you can automatically remove who, and where to get out of news articles.
Remember that nerve models are not perfect. They make predictions based on samples, not understanding. It is up to you how to translate their output and handle mistakes.