SwipeNews

SwipeNews is an online news aggregator service that compares different kinds of news.

Through the application of machine learning text processing and natural language text processing, SwipeNews makes it convenient for people to read how key news stories are covered by sources with different perspectives than their own.

The client came up with an idea to create a website that would collect different information and would group it regarding the subject of each text. We offered to validate his idea by creating a prototype. After seeing the first touches of his idea, the client was satisfied with what he was offered so he decided to proceed with the project and we started to build an MVP.

Location: UK
Industry: Media & PR

Product: news aggregator

The scope of our work: back-end

Solutions: ML, Natural Language Text Processing

Web site: swipenews.com

Sloboda Studio helped with every step of the project, from helping to define the scope and setting out the idea to execution and project management. I was very satisfied with the team's attitude and the result has given the limited budget. Great communication and ownership throughout the project.

Founder of SwipeNews

Client Goal:

To create a news website in order to provide the most objective picture of events.

The volume of truncated news grows, distracting readers from analyzing key issues more deeply and forming their own view. That’s why the client wanted to make sure that the readers wouldn’t rely on politicians to present an unbiased view and the media which is increasingly setting the political agenda rather than informing.

SwipeNews was created to appeal to those people who are interested in seeing a fuller spectrum of views and perspectives, to show news from different angles.

Solutions

How does it work?

The website contains sources. Each source has its own feed. Every source and feed includes their own status. Feeds are divided into topics for the purpose of further news filtering. There are 24 sources of 86 feeds (RSS feeds) in total.

When entering the system the news are downloaded in full (title, short description, full text) and edited (stop words, punctuation marks, decreased to normal form). Statistical evaluation and selection of simple (from one word) and complex trends (several words) are carried out.

Then a table of similarity between news, with the formation of a similarity coefficient based on keywords, abbreviations, trends, and complex trends. Trend Abbreviations is built.

After that the dynamic clustering with the ability to view a more general structure (modified for the needs of this service single link clusterizer) takes place.

#1 Challenge:

Text perception

Since there was a large block of unstructured textual information, it was difficult for the user to pick out the most important information and form an opinion around it.

Solution:

We created a search and to group news on various topics. This was achieved by using NLP for systematization and clustering of a large flow of textual information.

#2 Challenge

Server overload and processing speed

This issue was caused by the large volume of received articles and the analytics based on them.

Since there was a large block of unstructured textual information, it was difficult for the user to pick out the most important information and form an opinion around it.

Solution:

We limited the amount of analyzed information to cut off articles that are too long. To do so we used K-means clustering algorithms: basically, there are a few primary news stories and the rest are secondary ones. The scoring algorithm calculates the distance between the news and cuts off where the distance is too large.

The similarity assessment is carried out according to the following factors: word, abbreviation, trends, trending phrases, and complex combinations, which are used to process words or phrases. From the previous parameters, we form a common similarity index between articles. Finally, we form a chart from the news that has passed the limit.

Project Stages

#1 Stage
#2 Stage
#3 Stage
#4 Stage

Analysis of the customer’s business model
Architecture planning

Design improvement
Building a scalable architecture
News processing
Connecting GNIP API
Basic scoring algorithm

Enterprise API
Scoring algorithm improvement
UX analysis

Scalable infrastructure
Voice devices
Optimization of response time
Scoring algorithm 2.0
Increasing the number of data sources and topics covered

Success Stories

Scale your team with us