Organizations consume huge amounts of digital text. If big data technologies make it possible to better manage these volumes, they focus mostly only on structured data.
Today, organizations produce and consume huge amounts of documents and digital texts. While big data technologies can better manage these huge volumes of data, they mostly focus on structured data only. However, a text is not a structured data: to know what it contains and to extract the essence, it must be read in full. If it is a text of a few lines, it can be done manually, but what about when it is necessary to treat thousands or millions?
Automatically generate actionable metadata
For a text to be exploitable without the need to read it, it should be transposed into metadata and put them in fields more manipulable, like Excel lines. The challenge lies in the quality and interest of the metadata extracted from the text by tools called semantic annotators. Their role is to process and analyze large masses of unstructured content to derive key elements and key indicators. But, how does it work?
Who are we talking about? Named entities
First of all, it’s about detecting the named entities in order to know who we are talking about: a company, an organization, a place, a company, a person, an event, a product, etc.
Business intelligence and sentiment analysis: how does it work?
In a second step, it is necessary to extract the concepts to know the subject of the article in which our entity is quoted. Concepts are such phrases as “public health policy” or “attack wave”. There are two types: hot concepts (which are frequently spoken on the web) and categorized concepts (classified in particular categories such as “economy” “policy” or “environment”).
How is this information related to each other?
Then, the interest of the analysis is to be able to establish relations between two entities, linked by a phrase in a sentence. This may be a merger-acquisition relationship between two companies or the appointment of a person in a corporation.
How are we talking about it? The feelings expressed
In the immensity of big data, individuals are sensors like the others, but they disseminate subjective information: we can therefore learn about them and especially the feelings they experience. The most powerful annotators are able to detect the feelings expressed in a sentence, turning them into three levels of metadata. First, by extracting the nature of the feeling: it can be the anger of an unsatisfied customer, the joy of a promoted employee, a suggestion of improvement emanating from a consumer or even the intention of unsubscribing a customer. dissatisfied customer … Then we observe the force with which the feeling is expressed by graduating from -3 (very negative) to +3 (very positive). And this, especially by analyzing the adverbs used (“a lot”, “very strongly” etc.) Finally, the annotators make it possible to know the theme on which the sensor expresses itself.
What areas of application for sentiment analysis? The voice of the customer
The main area in which it is interesting to use sentiment analysis is the voice of the customer, to know what customers and prospects say products or brand on the internet (forums, social networks, blogs …). Indeed, bypassing the mass of web data through an annotator, it is possible to extract all the feelings expressed on a service, a product or a brand. Technologies are able, thanks to machine learning, to learn the domain’s semantic universe, to extract quantified data (intensity of feelings etc.) and thus to help the prediction (what works today Which products are on the downward slope?).
The voice of the collaborator
Sentiment analysis is also an excellent tool for knowing the voice of employees on HR topics, expressed spontaneously during annual interviews or on corporate social networks, for example. The analysis of sentiment raises strong issues here because with the help of insights, trends, and prediction, we can for example guard against resignations, it is attrition.
The dirty intelligence
There is a growing trend to link data from the company’s CRM to news data from a massive web crawl. For good reason, by combining these two sources of information, a company can point to the relevant elements that will help salespeople to adapt their speech to the expectations of their prospects. For example, if you know that the target company has opened a subsidiary, salespeople can use this information in their approach.