Generating Articles From Time Series Data

Data analysisNatural language generationSEO

Challenge

In a crowded digital content marketplace, it is difficult and/or expensive for data-driven businesses to attract SEO traffic because they produce very little “original content”. Two clients approached us independently at the same time with this very problem - a Spanish deal aggregator and a U.S.-based cryptocurrency portal.

Solution

We created Graphomania, a NLG system that uncovers stories hidden in time series data and tells them in the form of natural language summaries. It can generate hundreds of such stories per second and requires no human maintenance.

Input Data

Bitcoin chart

Example Summary

BTC Stabilized Around $750

After the steep fall of last month, Bitcoin price has stabilized around $7,500 this week. Today’s maximum was $7,536.62 with the market cap of $0.127T before a minor correction appeared around 2:00 AM UTC. Since then, BTC was traded for approximately $7,450 and closed the day even lower, at $7,437.43, with -1.11% day-over-day drop.

According to LongForecast, it is still not the best time to buy for long-term traders as Bitcoin price is projected to fall even more until May next year, when it’s expected to gain traction again.

Features

Time series analysis

Time Series Analysis

Advanced analysis of time series data: trends, peaks, events. This part of the system can be optionally extended by domain-specific plugins.

Smart templates

Smart Templates

Simple-to-use natural language templates with if/then branching and synonymic enrichment to ensure maximum variation of the generated content.

Timex engine

Timex Engine

Unique time expression engine that generates the most appropriate expression based on the story context.

System Architecture

System architecture

Results

5000+

summaries

generated every day as of June 2018

221%

organic traffic increase

on average after integrating Graphomania

Read More Case Studies