This topic contains 5 replies, has 2 voices, and was last updated by Chickyky 1 year, 5 months ago.
Maximum category node in one parent node
Chickyky October 1, 2015 at 7:24 pm
I read the docs: “A max of 50 categories with the same parent”.
I wonder Root node first has to be a parent node?
I need your help, I need to classify more than 100 Route class, for example New York – Texas. What should I do to have best accurate?
Sorry about my English.
Thanks for your help.
Hi Chickyky, could you please explain a little bit more about your use case and your objectives? I’m asking this because maybe this is a more appropriate case for doing text extraction (vs text classification). Cheers!Chickyky October 4, 2015 at 7:51 pm
In my case:
My input is a question on a journey from one place to somewhere.
for example: “I want to bus travel from Texas to New York”
for answer that question, the output should be Texas – New York, not New York – Texas.
I used MonkeyLearn and service really good, I’ve classify with more than 100 classes and test on data trained, the exact results of about 60%, due to the opposite error (eg New York – Texas) approximately 30% and 10% totally wrong.
I think the current MonkeyLearn can do very well about the exact orientation of a certain pair journey (60% + 30%) but no good can subclass in order of appearance words (my personal opinion ).
So my journey class Texas – New York is different from the New Yor – Texas, so it have more than 100 layers Can you give me some advice in case of me and how to get the best accurate?
Thanks a lot.Chickyky October 7, 2015 at 2:40 am
are you here?
How many combinations of journeys do you have? We understand journeys as a combination of departures (“from”) and destinations (“to”).
In your case, text classification approach would be useful only if you have a limited number of journeys (from X to Y) possible. In this case, you could try splitting the problem in two different classifiers:
– a classifier that classifies a journey according to its departure (where the bus departs)
– a second classifier that classifies according to its destination.
This approach could only work if you have a limited number of journeys.
If you have a huge amount of journeys (N combinations of departures and destinations), it is more recommend to use a ‘text extraction’ approach. You should train a custom text extractor that learns to extract the departures from your input (questions) and another extractor that learns to extract the destinations.
You could eventually use a pipeline to get the output you need to answer the question (output Texas – New York).
We are working on integrating custom extractors on MonkeyLearn (right now users can only use public extractors), so our users can build their own extractors that fit their needs, it it will be available soon.
You can learn more about text extraction here: http://docs.monkeylearn.com/article/whats-extraction/Chickyky October 11, 2015 at 7:11 pm
Thanks your advance, I will try.
Hope you continue support me.