Each web page found by the Brandwatch Crawler goes through a series of automated analysis processes:
Language detection: a Natural Language Processing algorithm recognises the language used in the page; this improves the accuracy of the other analysis processes, and means that customers can filter their data by language
Title and Content Extraction: the title and the main content of the page is extracted, so that navigation text and adverts do not interfere with the other analysis processes
Query matching: the queries (brands, competitors, etc) are matched against the content and title of each page; with each match Brandwatch has effectively found a new query mention
Sentiment analysis: smart sentiment classifiers analyse the mentions and recognise their tone (positive, neutral, negative) towards each query
Recurring phrase identification: new recurring phrases are continuously identified within each query’s recent mentions, highlighting new or growing trends amongst the query’s online coverage; this particular mechanism is much smarter than a simple ‘word cloud’, as it identifies the grammatical role of each word, and joins words into common phrases, resulting in more relevant combinations
The sentiment analysis is by far the most involved step of the whole analysis. Brandwatch developed its own Natural Language Processing framework for sentiment analysis, based on the regular retraining of specialised classifiers, for various combinations of languages and subjects.
This means that sentiment can be reliably measured over time. Combined with recurring phrase identification, it allows our customers to effectively measure new trends and themes and their tonality amongst their queries’ online coverage.
How It Works > Analysing Data
Each web page found by the Brandwatch Crawler goes through a series of automated analysis processes:
The sentiment analysis is by far the most involved step of the whole analysis. Brandwatch developed its own Natural Language Processing framework for sentiment analysis, based on the regular retraining of specialised classifiers, for various combinations of languages and subjects.
This means that sentiment can be reliably measured over time. Combined with recurring phrase identification, it allows our customers to effectively measure new trends and themes and their tonality amongst their queries’ online coverage.