machine learning - Too small RMSE. Recommender systems -

sorry, i'am newbie @ recommender systems, wrote few lines of code using apache mahout lib. well, dataset pretty small, 500x100 8102 cells known.

so, dataset subset of yelp dataset "yelp business rating prediction" competition. take top 100 commented restaurants, , take 500 active customers.

i created svdrecommender , evaluated rmse. , result 0.4... why small? maybe don't understand , dataset not sparse, tried larger , more sparse dataset , rmse become smaller (about 0.18)! explain me such behaviour?

datamodel model = new filedatamodel(new file("datamf.csv")); final ratingsgdfactorizer factorizer = new ratingsgdfactorizer(model, 20, 200); final factorization f = factorizer.factorize();   recommenderbuilder builder = new recommenderbuilder() {             public recommender buildrecommender(datamodel model) throws tasteexception {                 //build here whatever existing or customized recommendation algorithm                 return new svdrecommender(model, factorizer);             }         };   recommenderevaluator evaluator = new rmsrecommenderevaluator();         double score = evaluator.evaluate(builder,                 null,                 model,                 0.6,                 1);  system.out.println(score);  

rmse calculated looking @ predicted ratings versus hidden ground-truth. sparse dataset may have few hidden ratings predict, or algorithm may not able predict many hidden ratings because there's no correlation other ratings. means though rmse low ("better"), coverage low because aren't predicting many items.

there's issue: rmse dataset dependent. on movielens ratings dataset has star ratings 0.5 5.0 stars, rmse of 0.9 common. on dataset 0.0 1.0 points, i've observed rmse of around 0.2. @ properties of dataset , see if 0.4 makes sense.