search

o19s / elasticsearch-learning-to-rank

Plugin to integrate Learning to Rank (aka machine learning for better relevance) with Elasticsearch
http://opensourceconnections.com/blog/2017/02/14/elasticsearch-learning-to-rank/
Apache License 2.0
1.48k stars 368 forks source link

Does LTR work for long-form questions #127

Closed bigrig2212 closed 6 years ago

bigrig2212 commented 6 years ago

Is the solution ideally suited to keyword based search optimization - or do you think it would work also with longer form questions (with a lot of chaff in them)?

ie: "Hi. I can't find the warranty information for my BlahX98. Can you send me the warranty stuff for it? Thank you for the help."

softwaredoug commented 6 years ago

It's honestly good for any ranking optimization. Spellchecking, autosuggest, recommendations, search... LTR is a pretty broad subgrouping under LTR that covers optimizing any ranking as a function of features.

The features would be quite different. They'd likely be more related to understanding the intend of the question. With lots of NLP and what-not.

The overall topic of 'question answering' is very broad. Taming Text has some introductory chapters on the topic.

rpedela commented 6 years ago

Are you wanting to understand the query or optimize the ranking of the query's search results?

bigrig2212 commented 6 years ago

I'm wanting to pick the best answer (or top 3 answers) out of a large number of possible answers. So, depends what you mean by "understand." I think that optimizing ranking would achieve the desired effect, given a great dataset of answers to start - and a reliable feedback loop (which I have).

(just ordered Taming Text. thx). Also just received my copy of Relevant Search. Excited to dive into it!

bigrig2212 commented 6 years ago

Also have a hunch that there may be some interesting opportunities for pre-processing using AWS comprehend API...

softwaredoug commented 6 years ago

Thanks for the question @bigrig2212, don't hesitate to reach out to the LTR team on Relevancy Slack for further questions: http://o19s.com/slack