How to implement sentiment analysis of text?

Question

PHP

It is necessary to determine the tone of comments on the site on a scale of points, taking into account phrases and turns.
For example: from 1 to 30 points is potentially negative, 30-60 is neutral, 60 and above is potentially positive.

At the moment, classification is carried out by reducing words to normal form, obtaining word forms and a Bayesian filter.

But such a system produces only fixed states and does not take into account phrases and phrases.
Theoretically, you can make many fixed states, but then training becomes difficult and resource-intensive.

Answer 1

as I understand it, you use the naive Bayes model, in which the probability of a positive assessment is calculated for a given set of words in a comment and multiplying by 100 you get points?

Answer 2

www.sasbi.ru/sentiment-analiz/ this article will be useful for you.

Answer 3

According to the "correct" - full syntactic and, if possible, semantic analysis (this is how we implement it). However, with a “budget” implementation, I would at least try to take into account a couple of points:
1) Anaphora resolution (attempts to guess which word the pronouns “this”, “he”, “it”, etc. refer to will allow you to get a better result in more complex cases than “blue sky is good”
2) Correct binding of negations (I wrote it and realized that it is not trivial to do it without a syntax tree, but you can at least try to take into account the main cases). This will make it possible to understand where the object is denied, where is the characteristic