Google has unveiled research on a novel algorithm capable of generating “coherent” articles by using content from both your website and those of your competitors. This new algorithm creates original material to address user queries without redirecting them to external sites.
### How Does the Paraphrasing Algorithm Operate?
The algorithm functions by summarizing web content, using a process that extracts information and discards irrelevant parts, similar to the technology behind featured snippets. Known as “extractive summaries,” this method condenses content to its most essential sentences. Additionally, Google’s algorithm employs an “abstractive summary,” akin to paraphrasing.
However, a downside of abstractive summaries is that nearly one-third contain inaccuracies. More information on extractive summaries can be found in research about fact-aware neural abstractive summarization.
Google’s latest research combines the strengths of both extractive and abstractive methods, refining important facts from web documents before paraphrasing. This results in new articles similar to a bespoke version of Wikipedia. The details of this algorithm are discussed in a paper titled, “Generating Wikipedia by Summarizing Long Sequences.”
### Google’s Approach Explained:
Google has demonstrated that English Wikipedia articles can be generated using multidocument summarization from various sources. This process involves collecting data from multiple web pages and using extractive summarization to identify crucial information. The extracted details are then synthesized using a neural abstractive model, crafting natural sentences and paragraphs to form articles.
According to Google, these newly created articles can endure human scrutiny, producing fluent and coherent text based on the extracted facts.
### Featured Snippets as a Starting Point:
Featured snippets exemplify extractive summarization, where a webpage is distilled to a few sentences that directly answer a query. There exists a related Google algorithm, used for Google Voice, known as Sentence Compression by Deletion with LSTMs.
### Is Google Summarizing Your Content?
The algorithm’s purpose is to extract and summarize information from multiple documents, which can include books, open-source databases, and public web pages such as your content. The research employs Wikipedia topics as search queries, using search results to generate new articles. Additionally, it tests article generation using only references cited by Wikipedia.
This method allows Google to generate web content without linking to original websites, facilitating an answer without directing users to external links.
### Google’s Independence from External Content:
The research concludes that Google can effectively generate content by summarizing existing materials, offering user answers without needing to visit the original sites. The research claims success in multi-document summarization, using freely available documents, including competitors’ web pages.
### Future Potential for Voice Assistants:
Although it’s unclear if Google will utilize this algorithm for creating content, it is an ideal fit for voice assistant searches, enabling Google Voice Assistant to respond naturally, much like a personal conversation. This reflects Google’s ongoing ambition to offer advanced voice interaction, similar to futuristic concepts from popular media.
This research outlines Google’s potential for creating informative content derived from numerous web pages.
