How it works
We teach computers to answer your particular question.
This may sound like science fiction, but it really is what we do. Our software gives computers the ability to recognise features of text, images and video, and our human team shows them what to look for.
Defining the search
2 days – 1 week
You define the “category” in a standard brief. It can be wide (compact cars, horror movies) or narrow: the new Mini Cooper, a single film title. We run early probes on the web to augment the description, double‑check the findings with you, then our machine‑learning algorithms build an ontology for the category from both the brief and the content discovered.
Collecting & analysing
Continuous
Our proprietary focused crawlers use the ontology to separate mentions from opinions across roughly half a million online resources. Content is processed, classified and analysed: we calculate clusters, measure sentiment polarity and identify sentiment drivers: which aspect of the brand, product or person drives positive or negative opinion. Results are normalised by the buzzfactor.
Reporting
Daily, once trained
An online dashboard with key volume and sentiment metrics, with the actual data underneath it. Every comment, its URL, its buzzfactor. Alongside the data we provide our own analysis and conclusions and, where requested, graphical representations of public sentiment.
Learning, then listening
Clusters, and the ones nobody asked for.
Our team identifies examples of relevant comments that i‑sieve computers then analyse, literally learning how to distinguish relevant from irrelevant, positive from negative, interesting from unimportant. Relevant comments are assigned to one or more clusters: comments about the music on an ad, comments about the product’s effectiveness, and so on.
Sometimes the system finds resources that are clearly relevant but don’t fit any pre‑defined cluster. The team is alerted and can define a new one, even if it isn’t what was originally asked.
Work on a painkiller aimed at lower‑back and arthritis pain revealed a huge cluster of women in their mid‑30s using it for menstrual cramps.
Discovered by the ontology, not the brief.
Normalising the noise
A mention is not an opinion. A view is not an impact.
The buzzfactor
During collection a huge number of items are identified. Some are mere mentions with no opinion; others lie low in obscure corners of the web. We use every reference in building overall buzz measures and trends, but we apply the buzzfactor to normalise the data and select meaningful samples for deeper analysis. It estimates the size of the audience reached, and therefore the message’s influence.
Calculated from
- Recency
- Number of viewers
- Author’s reputation
- In‑links
- Number of comments
- Keyword frequency
- PageRank
The tubefactor
Views alone are not a robust metric for a clip’s impact. While collecting and delivering all relevant clips, we apply the tubefactor to normalise the data and quantify their influence. It goes one step further: following through on uploaders’ profiles to include a contributing factor derived from their channel’s reach.
Calculated from
- Recency
- Views
- Duration
- Comments
- Votes & ratings
- Channel views
- Subscribers
- Friends
Pinpointing influence
The higher the buzzfactor, the more impactful the resource, helping analysts rank sites, identify discussion hotspots for deeper monitoring, and find the power‑users in blogs and forums of the domain. The higher the tubefactor, the more impactful the clip. Using graph theory we derive special subgraphs (cliques) of these power‑users and obtain a solid group of influencers, ready for marketers to engage.
Why our approach holds up
The web is our database
Every project starts afresh, which shows in data quality, freshness and relevance. New sources are added as soon as they become available.
Fine accuracy, with alerts
The ontology approach allows fine accuracy, and content that passes the ontology but can’t be clustered alerts a human.
Terms users actually use
Strengthening the ontology with concepts discussed by users helps marketing teams qualify issues they never knew existed.
Languages at minimal cost
Professional services are needed only for the first few days to define the ontology, then for rudimentary checks at delivery.
Verifiable results
We deliver the exact data used by our classifiers. Clients can verify results and analyse the data further themselves.
Low maintenance
The master ontology is built at startup; keeping it fresh and efficient is mainly a matter of tuning.