Opinion MiningIdentify opinions and emotions with semantic analysis
Why analyze conversations?
Digital mobiles and devices today give us unprecedented access to information, products and services through a number of digital channels. Companies need to create a brand experience without interruption and clearing communicating their brand through the numerous touchpoints and channels that are available to their users. In addition, companies need to be able to profit from analysis the wealth of information available in order to understand and interpret and take advantage of the market opportunities ahead of their competitors.
It is therefore essential that brands can monitor their brand reputation online to assess the right actions in the shortest time. Company initiatives must also take account of all aspects social customer care and manage it in real time.
The content generated and shared on social media are invaluable resources for understanding users’ opinions and to have an accurate vision of users’ emotional levels.
Semantic analysis of social content
The Semantic Engine developed by CELI (and integrated in Blogmeter) monitors mood changes through technologies of Natural Language Processing and categorizes web opinions through Sentiment Analysis and Opinion Mining.
The process of Natural Language Processing created by out Sentiment Analysis tool is defined by three stages: analysis of portions of text such as punctuation, the examination of the tone of the messages through the analysis of the text and finally the categorisation of the documents according to their polarity of positive, negative and mixed.
According to their polarity the Semantic Analysis engine attributes a score which matches the intensity of the opinion described in the document (High, Medium, Low).
If on one hand the fluidity and variety of language lends it to be the main means of expression of our state of mind, on the other this state of mind is complex and difficult to interpret automatically.
In order to confront these complexities our team has developed two different language resources: Sentiment lexicons and Syntactic-Semantic rules. The former consists of a wide collection of words and multi-word expressions with information on the weight of their polarity (positive or negative) and on expressed emotions (joy, sadness, fear, anger etc.).
To guarantee the maximum precision of analysis, the lexical resources integrate a system able to select the correct polarity of language expression according to Semantic dominion of application and the context of language use. In this way, for example, words such as fast and slow in a musical sense will have no polar weight, “the song started slowly”, will have relevance in a technological dominion, “the pc that I bought is slow”.
The automatic understanding of complex emotions happens through the Syntactic-Semantic rules capable of interpreting language phenomena such as the phrase “there is nothing positive”, the reversal of polarity “without beauty”, the quantification “I don’t feel well”, the polarity in some use of words which are normally neutral “price reduction” and “price increase” and vice versa, the cancellation of polarity in common usage such as extraordinary, “extraordinary meeting” and “extraordinary opening”.