How to include categorical features in feature set about elasticsearch-learning-to-rank HOT 5 CLOSED

The reason to include this is because I do not want the model to rank 2 documents of the same category one after the other.. The training set is also provided in the same way .. So make to the model learn this I want to include this categorical feature.

from elasticsearch-learning-to-rank.

small world moment here: https://stackoverflow.com/questions/75958254/including-categorical-features-in-elasticsearch-ltr, but I still think one-hot encoding is the path forward here. This notebook has an example with a two category feature (is a movie genre 'Drama' or 'Science Fiction' : https://github.com/o19s/hello-ltr/blob/main/notebooks/elasticsearch/tmdb/tale-of-two-queries%20(ES).ipynb

from elasticsearch-learning-to-rank.

Haha that's me only even there :) Wanted to make sure I get a reply soon so just posted on both platforms..

from elasticsearch-learning-to-rank.

Hi!
I think this is interesting. I've been following the code of the "tale of two queries(ES)" example, and Im trying to figure out the following:
In the date_genre_judgments.py - file, the synthesize() function:

Seems the function only looks for the applicable genre value in the first value of the genres-field,
Is the genres list assumed to be sorted by the "importance" of the genres as appearing in the list, so that a list ["Drama", "Scince Fiction", "Comedy"] Would imply the main genre of the movie is "Drama"?
Or should the function open for "Drama" anywhere in the list?

Michael

from elasticsearch-learning-to-rank.

@mich123prem Yes, the function assumes that the first genre in a list is the main genre of a movie, just like in a list of actors the main actors are listed first. It is, after all, just a function to synthesize some judgments quickly in a demo, training or proof of concept phase, so don't take that as a general assumption.

from elasticsearch-learning-to-rank.