Skip to main content
💡Part of Purple8 Hyper Graph · Belief graphs · Stance detection

Model what the world actually believes.

Sentiment analysis tells you the tone of a sentence. That's the easy part. An opinion engine tells you who believes what, who convinced them, which positions are gaining traction, and where consensus ends and controversy begins. Modelled as a living, queryable belief graph that you can traverse, analyse, and ask questions of in plain language.

The belief graph

Opinions as data, not text.

DocIntel extracts entities and relationships from documents automatically. The graph engine stores them as typed nodes and edges — queryable, traversable, analysable.

Node
Claim

A specific assertion — 'Drug X is effective for condition Y'

Node
Person / Organisation

An entity that holds, expresses, or influences beliefs

Node
Document

Source of the claim — paper, article, filing, transcript

Edge
SUPPORTS

Entity or document endorses a claim, with confidence weight

Edge
CONTRADICTS

Entity or document disputes a claim, with confidence weight

Edge
INFLUENCES

One entity shapes another's position — timestamped

Edge
CITES

Document references another document as evidence

Edge
AUTHORED_BY

Claim provenance — who said it, in which document, when

Built-in algorithms

Six ways to understand a belief graph.

📈

Temporal PageRank

Which claims are gaining influence right now? Time-decayed edge weights surface rising vs fading opinions.

🔵

Community Detection

Which belief clusters exist? Who agrees with whom? Reveals faction structure across stakeholders.

🌉

Betweenness Centrality

Who are the opinion brokers — entities that bridge different belief clusters and drive consensus or division?

🔗

Link Prediction

Which entities are likely to align or diverge in future? Predict coalition formation before it happens.

🔍

GraphRAG

Ask natural language questions over the entire belief graph: 'What does the scientific consensus say about X?'

🛤️

Shortest Path

How did opinion A reach entity B? Trace the influence chain step by step through the graph.

Use cases

Where belief graphs change decisions

🏛️

Policy & stakeholder mapping

Before publishing a regulation, map every stakeholder's documented position, who influences whom, and which coalitions exist. Know where opposition will come from before the consultation starts.

💼

M&A sentiment intelligence

Map the target company's board, investor base, and key customer positions. Identify who supports the deal, who will resist, and which advisors are most influential — before the first approach.

📰

Media & narrative tracking

Track how a claim propagates through media outlets, think tanks, and social networks over time. Identify the original source, the amplifiers, and the points where the narrative diverged.

⚖️

Legal discovery & position analysis

Map which documents support, contradict, or are silent on each contested claim. Betweenness centrality identifies the pivotal documents — the ones that connect opposing positions.

🔬

Scientific consensus detection

For any research claim, traverse the citation + SUPPORTS/CONTRADICTS graph to quantify true consensus vs manufactured controversy. Surface which papers are load-bearing for each position.

🏢

Brand & reputation intelligence

Model customer, analyst, and press beliefs about your brand as a graph. Track which criticisms are gaining traction, which advocates are most influential, and where reputational risk is concentrated.

FAQ

Common questions

What is an opinion engine?

An opinion engine models who believes what, how strongly, and why — across a population of entities (people, organisations, documents, claims). Rather than counting keyword mentions, it builds a typed graph of beliefs and their relationships, then applies graph algorithms (PageRank, community detection, betweenness centrality) to understand how opinions form, spread, and conflict.

How does Purple8 model opinions as a graph?

Entities (Person, Organisation, Document, Claim) become nodes. Relationships (SUPPORTS, CONTRADICTS, INFLUENCES, CITES, AUTHORED_BY) become typed edges with confidence weights and timestamps. DocIntel extracts these automatically from documents. The graph engine then runs temporal PageRank to show which claims are gaining influence, community detection to find belief clusters, and betweenness centrality to identify opinion brokers.

What is the difference between sentiment analysis and an opinion engine?

Sentiment analysis classifies tone (positive/negative/neutral) at the text level. An opinion engine models the structure of belief — who holds which position, which claims support or contradict each other, which entities are influencing others, and how opinion networks evolve over time. It answers 'what does this stakeholder group believe about X and who convinced them?' — not just 'is this tweet positive?'

Can the opinion engine track how views change over time?

Yes. All edges carry timestamps, and Purple8's temporal PageRank applies time-decay to edge weights — recent influences count more than old ones. You can snapshot the graph at any point and compare belief networks across time periods to see which claims gained traction, which influence paths emerged, and which positions were abandoned.

Start modelling beliefs, not just sentiment

Free to start. Graph algorithms, temporal PageRank, and GraphRAG work on day one.