Title: Opinion Mining for Business Intelligence Applications
Authors: Horacio Saggion - University of Sheffield


Work on opinion mining has been fuelled by the development of
evaluation programs such as the Text Retrieval Conference (TREC) 2006
track on blog mining for opinion retrieval or the National Institute
of Informatics Test Collection for Information Retrieval (NTCIR)
Workshop on Evaluation of Information Access
Technologies
and the new Text Analysis
Conference (TAC) with a track
on opinion summarization.


Opinion mining consists of several different problems, such as
determining whether each segment of text (sentence, paragraph, or
section) is ``opinionated'' or not; identifying the opinion-holder
(the person or organization who expresses the opinion); and determining
 how positive or
negative each opinion is.  For business intelligence, it is also
useful to classify each opinion according to the aspect of the
business or transaction described: e.g., product quality, ordering, or
integrity.

Opinion analysis helps to assess the limitations of particular products
and then exploit this information in the development of improved
products or services.  It also helps enterprises understanding their
customers as well as plan for future products and services.  


Given the abundance of reviews on the World Wide Web about companies, products, and services 
especially with the more recent proliferation of blogs and other
Web 2.0 services, one application is to identify for a given entity
its features and then identify what is being said about them (positive
or negative statements), in order to compile summaries of opinions
about particular entities or features.  This information is then
compiled in order to produce a textual summary together with
statistics about what has been said.  Opinion summaries are useful
instruments in competitive intelligence for example, because they help
assess the limitations of particular products and then exploit this
information in the development of improved produces or services by the
producer or its competitors.

The work to be presented in the paper has been carried out in the context of
the business intelligence
 Musing project, and in particular with respect to a target application whose
goals is to compute and track the reputation of business entities. One of the core components
of such application is a natural language processing application which
identifies opinions in text about a given entity. 
The paper will describe experiments carried out for the identification
of opinions on Web sources. We will describe in detail the problem, the data sources used,  the linguistic processing
 applied to the data for  the  derivation of usefull
document representations, the machine learning experiments carried out, and the results.