As recently reported, numero presented a paper at AI-2009, the conference of the British Computer Society Specialist Group on Artificial Intelligence (BCS-SGAI) at Peterhouse College, Cambridge.
Entitled ‘Learning to Improve e-mail Classification with numero interactive’, the paper describes how the numero interactive platform uses a small number of manually classified messages to create a sophisticated statistical model from which to classify new messages. Whilst the paper focuses on e-mail communication, the same approach is used for all message forms, including those submitted by phone, fax, letter and web form.
Copies of the paper – which contains both a high level and detailed technical overview – are available from numero, free of charge. To request the paper in electronic format please use the contact form opposite or contact us here.
From an Artificial Intelligence perspective, the statistical classification service detailed in the paper represents just one of numero’s capabilities, as we have a number of NLP services that deliver some rather clever functionality including:
- Entity extraction: identifies names, addresses, phone numbers, order numbers etc contained within messages
- Language identifier: identifies the language used within a message, with support for 40 languages as standard
- CMS/knowledgebase: identifies content from a knowledgebase that is relevant to the content of an inbound message
- Spell-checker: through the use of smart algorithms we can support large dictionaries without a significant drain on system resources and ensure no delay in look-up time
- Profanity checker: identifies profanities within messages (typically applied to restrict outbound misuse), including all variants and derivatives
A well defined development roadmap exploits the reuseability of our proprietary AI technology and algorithms to bring fresh functionality for customers. Developments include:
- A sentiment analysis service that establishes whether a message is positive or negative
- An FAQ service that establishes whether a message can be automatically answered by information within a knowledgebase
- A content-sensitive spell checker
The numero paper presented at AI-2009 is a great place to find out more – request your copy using the form opposite.