This article provides answers to frequently asked questions about the Narratives AI product. Whether you're a seasoned user or just starting to explore Narratives, this FAQ article aims to give you a deeper understanding to help you get the most out of your Narrative AI experience. Here, you'll find answers to common questions about narrative structure, clustering, searching, and more, as well as helpful tips and insights to enhance your understanding of this innovative technology.
What makes a narrative?
A narrative is a belief/story held/shared by a particular audience. As a data mining tool a narrative is a cluster of statistically similar news articles and social media posts which describe an event, theme, story from the perspective of a particular audience, within a similar timeframe.
To grasp the concept of narratives in the Narrative AI product, it's essential to understand the hierarchical structure of information that forms a narrative. From specific to general:
Context Bundle:
In news: A single news article
In social media: a series of related social media posts, such as Tweets and replies under a parent Tweet, that form a unit of meaningful content. Context bundles are used to normalise content into smaller, meaningful units that can be clustered into daily topics.
Daily Topic: A cluster of semantically similar context bundles that share a common theme or focus. These are the building blocks of a narrative. .
Weekly Topic: A cluster of similar daily topics that emerge over a one-week period. These are the building blocks of a narrative.
Narrative: A system of topics appearing together over multiple days or weeks.
Theme: A grouping of related narratives that share a common theme or focus, revealing key topics and trends in a specific area of interest.
How do we cluster narratives?
Unlike other models that simply extract entities from the text of the content or categorise the content into topics, the Narratives AI model assesses content based on sematic similarity. By using semantic similarity clustering, we can match similar content based on meaning, so that similar conversations can be identified despite using different text to discuss the same subject. For example, in a discussion about obesity treatments, articles or social posts may mention words like “diet”, “weight-loss drugs”, “obesity medication” and “weight management”. Using semantic similarity matching, this content would be identified as similar even though the wording used is different.
How are topics and narratives linked over time?
Each day, an AI model assesses similarity of news and social content, then groups the content into daily topics. New topics clustered for subsequent days are compared to the previous day, and are then linked to the topics that they have the highest similarity to. At this point they form a a narratives. If a new topic is not similar enough to any of the previous daily topics, they will not be linked. Also, if an existing daily topic is not similar to any topics on the following day it will not be linked any. There is no limit to how long a narrative can run for, however, over time narratives may deviate.
How are weekly narratives created?
Weekly narratives are formed by comparing the similarity of daily topics in a seven day period (Monday to Sunday). Similar daily topics in the same week period will cluster together and form a weekly topic. Like daily narratives, weekly narratives are formed by linking the most similar topics over subsequent weeks.
How are themes extracted?
To ensure that you can always find the most prominent themes in any are of interest, themes are clustered at the time you conduct a search. We do this by analysing the centrality of the top 500 narratives (based on relevance) matching your search and then comparing this centrality to group similar narratives and create a theme. The top 10 themes are then displayed based on highest significance.
Why are there sometimes multiple narratives with a similar focus?
Narratives clusters similar content into distinct narratives based on subtle differences in perspective, tone, and emphasis, even if the overall topic is similar. For example, the below two narratives ran for a similar period of time and both focus on Israel’s invasion of Gaza:
Narrative 1
Title: Israel's Actions in Gaza Condemned by International Community
Summary: A growing chorus of international voices is condemning Israel's actions in Gaza, accusing the country of perpetuating a humanitarian catastrophe. From the UN to individual nations, leaders and organizations are speaking out against Israel's blockade and military actions in Gaza. The Czech MEP has gone so far as to accuse Israel of genocide, while the UN has decried the country's blocking of aid trucks. Even the UK's Cameron has called for more aid to be allowed into Gaza. The Gaza Strip has become a humanitarian crisis zone, with Israel's actions exacerbating the situation. The international community is increasingly critical of Israel's actions, calling for an end to the violence and a negotiated settlement to the conflict.
Narrative 2
Title: Israel's actions in Gaza are widely criticized and seen as a threat to regional security.
Summary: The international community is increasingly critical of Israel's actions in Gaza, with many countries and organizations accusing Israel of genocide and human rights violations. The International Court of Justice (ICJ) is set to hear a case brought by South Africa against Israel, with many experts and countries, including Indonesia and China, supporting the case. Israel's Prime Minister Netanyahu has been accused of inciting genocide, and even the UK's chief rabbi has acknowledged that Israel's actions in Gaza are not justified. The South African government has been vocal in its criticism, with many calling for Israel to comply with international law and respect the rights of Palestinians. The narrative emerging is one of widespread condemnation of Israel's actions, with many seeing them as a threat to regional security and stability.
In this case, the two narratives are clustered separately because they present different aspects of the international community's condemnation of Israel's actions in Gaza, with distinct:
Emphases: Narrative 1 focuses on the humanitarian crisis, while Narrative 2 emphasizes the legal and security implications.
Perspectives: Narrative 1 quotes political leaders and organizations, while Narrative 2 references legal entities and experts.
Key points: Narrative 1 highlights the blocking of aid trucks, while Narrative 2 mentions the International Court of Justice's involvement and accusations of genocide.
By clustering these narratives separately, we can provide a more nuanced understanding of the topic, showcasing the various facets of the international community's condemnation of Israel's actions in Gaza.
What data sources are Included?
News
Narratives uses content from over 40,000 global English print and online outlets. These include:
Top international, national, and business news sources, e.g. the New York Times, BBC, CNN, The Sydney Morning Herald, The Economist.
Top regional sources, e.g. Houston Chronicle, Scotsman Online, MIT Technology Review, Pharmaceutical Journal, Advertising Age.
A broad range of news sources of good editorial quality. Includes the following types of news sources:
Industry specific news sources such as PC Magazine, World of Concrete
Country specific news sources, e.g. New York Daily News, Canadian Press
Government department press releases, e.g. US Treasury, 10 Downing Street
International organizations, e.g. OECD, NATO
Dedicated sports news sources, e.g. ESPN.com, Sports Network
Social
Social media content is currently not availble in Narrative AI, but soon we will expand to include X (formerly Twitter) content in English, and will expand to include additional sources in the future.
How far back can I search?
We include data back to the 1st of January 2023, and you have access to search for anything from then to the current data.
How frequently are new narratives added?
For now, new weekly narratives are added each Wednesday covering content over the previous 7 day period Monday to Sunday.
How are narratives summarised?
Large language models are used to summarise a sample of the most representative content in a narrative. The samples are selected from the daily or weekly topics that are most representative of all the topics in the narrative. Sample content can be selected from any period, meaning longer narratives will often include summaries with a broader focus than summaries for short narratives.
We hope you enjoyed reading this article! 📚
If you have any questions or would like to learn more, please don't hesitate to reach out to our support team via live chat. 🚀