Deep sentiment and semantic analysis
Yes, quantitative insights are very important, but they are not enough. We need to know more about the target audience real opinion and feedback such that we can use it in our next campaigns (or even to rectify a running campaign). Here comes the qualitative insights.
Firstly, to be able to analyze textual posts written by social network users, it is mandatory to have a technology that is capable to extract the various meanings that will exist in the users textual contribution. Such technologies are classified into two main categories as follows…
- Sentiment analyzer… which is an entry level technology able to speculate the textual tone either to be positive or negative and accordingly calculate the overall sentiment of a group of posts.
- Semantic analyzer… Semantics are a much broader technology. It does not stop on sentiment but goes much more further. A good semantic analyzer should be able to identify the subject, time and place of textual streams (if exist) plus the object/objects that are mentioned. Based on the object type -e.g. a human being, a mobile phone, a company …etc.- it can then extract many useful indicators about its features, needs, feelings, actions and many other dimensions.
To illustrate some samples of the semantic analysis, we shall hereafter use some findings from Ramadan 2017 annual reports.
Firstly, let us have a look on the Most mentioned products’ features for commercial advertisements… When talking about product features, audience have talked mainly about service and price. There talks about these two features were with negative tone. Comes after this, the advertisement offering with a positive tone. Then followed by healthy food, taste, weight, sound, color, quality and reputation.
And when digging down more details about the Associated meanings that comes with each feature for the top 3 features, the main trends with each was as follows… 4.6% of the users’ contributions on commercial advertisements have talked about the service that is provided by the advertiser. And 4.2% of the talks were about prices. On the other hand, 2.8% percent were posts that appreciated the advertised offering and asking about the ability to continue this offer beyond Ramadan.
We may apply the same techniques on drama and charity fund raising.
As another use model, we can identify the Most offensive posts by volume or degree of offensiveness. In the holy month of Ramadan, users’ contribution with offensive (impolite) wording did not represent more than 0.24% of the whole users’ contribution. Such offensive posts were distributed over the various sample sectors as shown above.
Just think of the above and you can imagine thousands of very important insights that can be concluded from the implementation of semantic analysis on qualitative insights.