Hi all,
I have a user-item-rating data set (1 is low and 5 is high rating) and would like to obtain top-N recommendations. Currently, I am using the EALSRecommender. Here are my two questions:
1) Since I'm doing ranking, do I need to convert ratings into rankings for each user? I don't think so, but I want to make sure.
2) I am using K-fold cross-validation and AveragePrecision@10 to measure the performance of the model. Does KCV make sense when ranking items or should I use some other approach (e.g., LOOCV)? Also, should I use NDCG instead of AP? (I read somewhere that AP is more appropriate when having binary ratings)
Thanks in advance and thanks to @SunYatong for answering my questions on Github :-)
I am trying to evaluate the diversity of both, the slim and the plsa recommenders. however, the only result that I saw till now is 0.0. I wonder if there are special settings for diversity evaluator? bellow is the setting file.
Thanks in advance.
dfs.data.dir=../data
dfs.result.dir=../result
convertor
load data and splitting data
into two (or three) set
dfs.data.dir=C:\data
data.input.path=movielens/ml-100k
data.input.path=filmtrust/ratings
data.column.format=UIR
data.model.splitter=ratio
data.splitter.ratio=rating
data.model.format=text
data.splitter.trainset.ratio=0.2
set the random seed for reproducing the results (split data, init parameters and other methods using random)
rec.random.seed=1
binarize threshold mainly used in ranking
data.convert.binarize.threshold=-1.0
evaluation the result or not
rec.eval.enable=true
if this algorithm is ranking only true or false
rec.recommender.isranking=true
rec.recommender.ranking.topn=50
rec.iterator.maximum=100
rec.pgm.burn-in=100
rec.pgm.samplelag=100
rec.recommender.class=slim
rec.neighbors.knn.number=50
rec.similarity.class=pcc
rec.recommender.similarities=user
rec.eval.classes=precision,recall,auc,diversity
Hi all,
I wonder what tools and algorithms of Librec should I use if I want to implement a multi criteria similarity search for a bike recommendation system?
Assume I have a database where all bikes have 3 attributes: Category, Weight, Price.
The user sets his preferences to: Mountain Bike, 30kg, $200
I would like to get as a recommendation result the following 2 bikes:
- Mountain Bike, 25kg, $180
- Mountain Bike, 35kg, $250
So, how should I go about it using Librec?
Thanks in advance!
I have been playing around with SVDPlusPlus and I have noticed that it is able to make predictions on an item that has never been seen (was not rated by any user in the training set, because it had been rated only by one user in the movielens 1M dataset, and that single rating was assigned to the test fold so it does not appear in the training folds). Could you please explain how do you predict a rating that has never appeared in the training set?
