Machine News Writing: Origins, Technologies, and Future Prospects

Introduction

Machine news writing originated in the United States and has been adopted by Chinese media platforms such as Tencent Dreamwriter and Toutiao's XiaomingBot, primarily for data‑driven news like finance and sports. These systems automate the entire workflow from data collection to text or multimedia generation and publishing.

Origins and Development

The practice began with financial news; in 2006 Thomson Reuters used a program to produce economic reports in 0.3 seconds, sparking both excitement and criticism about the lack of “soul” in machine‑written articles.

Despite early doubts, AI advances have expanded capabilities. Companies like Automated Insights (AutoIns) now provide automated market analyses, millions of sports summaries, and integrated textual descriptions for data‑visualization tools.

Main Technical Overview

Machine writing can be divided into two major types: (1) data‑to‑text generation (e.g., stock analysis, sports match reports) and (2) extractive summarization of large text corpora for news cards.

Template‑based generation remains dominant: experts or algorithms extract standard structures and sentence patterns from historical articles, then fill them with up‑to‑date data. Challenges include selecting key data points, avoiding repetitive phrasing, and pairing text with appropriate images or GIFs.

Neural text generation, by contrast, learns writing styles from massive corpora without explicit templates, but currently produces short, less‑coherent sentences.

WritingBot (Sohu) Case Study

During the 2018 World Cup, WritingBot generated around 70 match reports, producing a complete article with text and images within one second of a goal. The pipeline involved building a 50,000‑article football database, extracting terminology and sentence templates, real‑time data capture, and automatic publishing.

The system identifies key events (goals, cards, halftime) and selects from multiple template sentences to create varied, accurate descriptions. It also uses OCR to locate relevant video frames and generate animated GIFs for highlighted moments.

Financial News Generation

Financial bots produce daily market summaries, stock‑specific alerts, and aggregated news. They monitor market movements, detect events such as limit‑up/limit‑down stocks, and generate personalized reports for users. The workflow mirrors sports reporting: data collection, analysis, template filling, and distribution.

Hotspot News Summarization

Hotspot news uses extractive summarization: topics are captured from internal and third‑party platforms, related articles are retrieved, and multi‑document summarization produces concise news briefs for push notifications.

Future Outlook

Machine writing will improve in quality and creativity as knowledge graphs and advanced language models mature. Applications will expand beyond media into B2B reporting, data visualization, and personalized content generation. However, fully creative long‑form writing remains beyond current capabilities, keeping human journalists essential for deep analysis and storytelling.