Analyzes of texts put lights in two main types of information “facts and opinions”. Most current treatment methods of textual information aim to extract and use factual information, this is the case for example of research we do on the web. Analysis of opinions is concerned about feelings and emotions expressed in the texts, it has grown much today because of the space taken from the web in our society, and the very large volume of daily comments expressed by consumers with the advent of the Web 2.0 world.
What is opinion analysis? It identifies the orientation of an opinion expressed in a piece of text (blog, forum, comments, website, document sharing site, etc.). In other words, it determines whether a sentence or a document expresses a feeling positive, negative or neutral regarding a defined object. For example: "The movie was fabulous" is an expression of opinion while saying "the main actor of the film is Jean Dujardin" is the formulation of a factual matter. Opinion analysis may occur at different levels. At the word level: the film is entertaining and motivating; on the sentence level: the police (subject) hunt (verb) smuggling (object), or finally at the document level, that is to say a set of sentences: his early films were very good, but this one is worthless.
In fact an opinion can be characterized by a formula of five components, a quintuple: Oj, Fjk, Hi Tj, SQijkl, where Oj is a target object; Fjk a characteristic of the target object; Hi a bearer of opinion; til the time the opinion is expressed, and SOjkl the opinion orientation, of the bearer Hi, about the Fjk characteristic of the OJ object at Tl time. Using this formula we can structure an entire unstructured web document, highlighting all quintuples included in the text. Quintuples represent structured data that can be analyzed qualitatively or quantitatively, and visually represented with traditional tools of decision systems. All kinds of analyzes are possible. Analysis of opinions is not only to characterize the opinion of one person by words and phrases, but also for example to compare the opinions of different people or groups.
The first step is to delete sentences that contain only facts, keeping only those who express and define the polarity (positive, negative or neutral). Specifically you have adjectives (red, metallic) that indicate facts or positive feelings (honest, important, mature, big, patient), negative (harmful, hypocritical, ineffective) or subjective being neither positive nor negative (curious, strange, odd, perhaps, likely). It is the same for verbs, positive (praise, love), negative (blaming, criticizing), subjective (predict), or for the names: positive (pleasure, enjoyment), negative (pain, criticism) and subjective (prediction, impression).
Be careful with sequence words meaning, a sentence can be complicated and punctuation that is of great importance, can play tricks. “Not good at all” is different with “Not all good”. The sentence might express sentiment not in any word: “convinced my watch had stopped” or “got up and walk out”. We must consider that words or phrases can mean different things in different contexts and domains, or the subtlety of the expression of feelings when someone makes ironic statement.
Ultimately, however, the analysis of opinions and feelings is able to provide much information about the populations studied and warned marketers already know how take advantage. This is true of many Teradata Aster customers, like Barnes & Noble for example. If you want to go further on this subject you can usefully consult the following site: