How to rank a lot of reviews like "A bad"; "A is better than B"; "A and B are bad, D is better"; “C and A are terrible, B and D are so-so, A is beautiful”?

Question

Algorithms

And tell me the tool in which it is easiest to do this

Answer 1

I'm not sure it's possible to make a full rating based on such reviews...
Reviews should be placed on the scale. To do this, you need to evaluate each verbal characteristic: "good", "bad", "terrible", "excellent", "so-so", "not very".
With the help of comparison, better/worse place on the scale and then perform statistical processing and aggregate the results.

Answer 2

Fuzzy logic (Fyzzy Logic) is excellent for solving such problems. She just operates with such concepts.

Answer 3

In general, such problems are solved, including at all sorts of Olympiads. Google in the direction of combinatorics. Completely without Google, I would solve the problem as follows:
1. For each of the product groups (a and b, for example), I would receive an average review (b is better in 75% of cases), and then I would simply sort it .. And it would be better to use c 75% of cases, but b is better in 60%, so the top looks like this: a, c, b.
Eniway, there will be artifacts (but better b, better a, but b better c), if you do not give weight to reviews and do not build a model of the estimated parameters and their importance (functionality, price, ergonomics, etc.)

Answer 4

1. We divide reviews into direct ones ("A is excellent, B is bad") and relative ("A is better than B").
2. We assign some ratings to direct reviews.
For example: terrible 0, bad 1, so-so 2, normal 3, good 4, excellent 5.
We calculate the average for each product.
3. Similarly, we assign certain coefficients to relative reviews.
Something like this: much better than x2, better than 1.5, slightly better than 1.25, slightly worse than 0.8, worse than 0.6, much worse than 0.4.
We calculate the average for each pair of goods.
4. We put everything together with some weight coefficients: a direct assessment of the product itself and the assessment of all competitors / analogues, multiplied by relative coefficients.
That is, in essence, the algorithm is quite simple, but in practice it will take a lot of time to run in and fine-tune the coefficients.