i-sieve Technologies - Sentiment Analysis
Sentiment analysis is the process of determining the opinion held by an individual about a particular topic. Related terms include opinion mining, natural language processing, computational linguistics and information extraction.
There are many different approaches to the task but i-sieve Technologies' patented approach has proved to be particularly effective. Our system is exceptionally good at identifying relevant comments made by individuals and organisations in all sorts of situations including:
- Twitter posts
- Social Networks (where publicly accessible)
- Professional media
- Discussion forums
- YouTube Comments
i-sieve provides statistically sound quantitative data about attitudes to your company from across the web. In addition, it:
- tells you what people are saying about you;
- tells you why they are saying it;
- tells you how they perceive your brand compared with your competitors.
Unlike ordinary buzz monitoring services, i-sieve does not work from a database of pre-selected sources, instead we extract data directly from the Web, how ever many relevant blog entries, forums, tweets or comments we find. Our analysis uses a combination of human and machine intelligence to offer remarkable classification accuracy across enormous amounts of data (we routinely achieve well over 90% accuracy). Comments expressing sentiment about a particular brand are assigned a Buzz Factor: an easy to understand, mathematically-sound measure of its importance and influence.
i-sieve uses technology developed at the National Centre for Scientific Research in Athens: the birthplace of modern mathematics and logic. Close ties with this world-class research centre means our search methods and analysis algorithms define the state of the art.
Through our dashboard you have direct access to the data as it is collected. You can see:
- the comment that triggered the classifier in isolation;
- the comment on the actual page live on the web where it can be found;
- any comment from the i-sieve team;
- the comment's Buzz Factor;
- high quality graphs and charts based on the data as it is collected.
Just A Hint
Sorry, but we're not going to tell you exactly how we do it — that's as secret as the Coca-Cola recipe — but we don't mind giving you a few hints.
Our computers have the ability to recognise patterns in video, audio and text. For each new question asked, our (human) team trains the system by inputting examples of content that are relevant and that fit into different clusters, i.e. specific areas of interest. From these examples, the system builds up an ontology, that is, a specific classification system. A well known example of an ontology is that of the natural world:
- there is class of thing called an animal;
- sub classes of animal include arthropoda, chordata and echinodermata;
- sub classes of chordata include the mammals and sub classes of the mammals include the primates.
The relationships are as important as the terms, so we can construct statements like:
- all primates are mammals;
- not all mammals are primates;
- an animal cannot be both a mammal and an arthropod.
Combined with highly efficient web crawlers (automated systems that go from one website to another) and spiders (automated systems that look deeply into individual websites), we are able to identify the relevant comments as they appear online. The crawlers also take note of the number of links that are made to a particular comment and this is an important factor in assessing the relative influence of a given individual or publication: what we call the Buzz Factor.
It is this combination of sentiment analysis and web technologies that allows i-sieve to deliver so much more than a mere count of the mentions of a particular term or the relative number of positive and negative comments made about a given topic.