Illusion as a way to fight in the internet.

[Плотников Юрий Михайлович]

ROMAN, thanks, interesting stuff!

 

[Матушкин Андрей Николаевич]

Thanks again!

 

[Марат]

Thank!

 

[Турабян Владимир]

THANK!

 

[Ющук Евгений Леонидович]

But recently, the emphasis in promoting such systems has been on the ability to identify negative and positive. And thus giving the user a completely new unique tool. Here on this tool you need to dwell in more detail. How does he work? And also - by a set of keywords. Those. we decided that the word "pid ####" when mentioning together with the name of our product "VasKomProm" is a negative message about the product, so it will be so. Even the message "..." McWafen "is still" pid #### "because" VasKomProm "is a very decent product, but why do not they sell it?" (example real only names changed). But the mention is rather positive and this is a simple example. And then there is humor, hints, sarcasm, which machines are simply not able to understand. And here, the spacing between words and other settings will not help. How to believe such a division into negative and positive? This is a direct misinformation of the user. So do not flatter yourself at the expense of the capabilities of systems in the field of "positive-negative."

There is no misinformation here. Neither direct nor indirect. I will explain.

Firstly, far from always even one person can understand another. Yes, even look at this forum - from time to time you have to ask again what the person had in mind. In this regard, the machine identification of negative and positive is neither better nor worse than human.

Secondly, Internet resources, which are evaluated in terms of negative and positive, are divided into two unequal size categories. The first are significant, widely visited, authoritative resources. There are not many of them.
The second - "extras". In the "crowd" the opinion of each individual does not matter, but the quantity is important. When it is small, it does not affect the situation; when it becomes large, it does.

Based on this, the conclusion is obvious: the machine can look at significant resources only for the fact of mentioning the object, but not for evaluating the negative and positive. Such an assessment is already made by man.
The car can look at the "mass" also with a view to assessing the negative and positive.

How to avoid the problem of sarcasm / humor and the problem that the negative word refers to another object, which is also mentioned in the text?

In fact, there is no problem from a practical point of view.
Refined sarcasm is so rare in comparison with more direct statements that in the case of the "extras" it can be safely ignored. Most likely, he will even be balanced by the reverse statements, also not caught by the machine, as a result, the specific gravity of the negative and the positive will not change significantly. But even this fluctuation is so insignificant that it does not matter.
Negative / positive related to the object of study is caught (for example, in IQBuzz) due to the parameter "distance from the object." That is, for example, we look at the negative only in five words from the object. Naturally, some of the objects will not fall into the selection. So what? When do you assess the extent of the flood in the Far East in order to understand whether water is arriving or retreating, trying to measure it with an accuracy of a glass? It is important?

So in the case of definition systems speakers negative and positive - exactly the same.

 

[Хагани]

thanks interesting