Вопрос

I have a CSV file like this:

typeA,typeB
typeA,typeC
typeA,typeC
typeA,typeB

Here, typeA, typeB and typeC are 3 different types of entities. Consider types B and C to be two different types of items and consider type A to be the users.

I can build a model by feeding this CSV file into Myrrix. This file has two types only, B (the "B" items from the former CSV file are in here as users) and D. Now, suppose I have another CSV file like this:

typeB,typeD
typeB,typeD
typeB,typeD
typeB,typeD

Here, I want to get recommendations of typeD for typeB. Must I have a separate Myrrix instance for this? Is it possible to have two models and choose which to recommend from? Doing this with Apache Mahout is easy I guess, but I wouldn't be able to get the joy of real-time setPreference and model refreshing like in Myrrix then I suppose?

I would appreciate any idea to solve this.

Это было полезно?

Решение

Yes, these are conceptually two different domains. One is A-to-B-or-C, and the other is B-to-D. This might be users-to-articles-or-posts, and articles-to-tags. D doesn't exist in the first one, but is exactly what the second recommends.

The separation does keeps things simple. You can, for example, host both models within the same machine or even JVM potentially. For example, consider running two servers on two ports. You can choose which one you want by accessing one or the other service. Or you could simply embed and use two instances of ServerRecommender directly.

It's not different than Mahout in this regard, no.

Лицензировано под: CC-BY-SA с атрибуция
Не связан с StackOverflow
scroll top