recommendy

##Usage

data_hander = JSONDataHandler("http://your_data_resource")
similarity_metric = pearson_corellation
racommender = Recommender(data_handler, similarity_metric)

• To get item based recommendations:

      recommender.item_based_recommendations("<subject_to_get_recommendations_for>")

• For content based recommendations:

      recommender.content_based_recommendations("<subject_to_get_recommendations_for>")

##Similarity metrics

There are three similarity metrics currently built in the recommender:

• euclidean distance
• pearson corellation coefficient
• tanimoto score

Euclidean distance and pearson corellation coefficient can be used when calculating similarities based on given scores, for example when users rate movies.

Tanimoto score is best suited when similarities are calculated based on the presence or absence of a property. For example if you have a web store and you just indicate wheather an item was bought or not and want to make recommendations based on items bought.

You are free to implement whatever other similarity algorithm most suits you and pass it to the recommender. It should take two maps, formatted like {: } as arguments and always return a number between 0 and 1.

##Data Providers The recommender relies on a data provider to feed it necessairy data in order to make recommendations. Check the data handler interface to see what methods each data provider must implement. There are currently two abstract data handlers - XMLDataHandler and JSONDataHandler, but as long as it implements the data handler interface, you can pass to the recommender whatever classes you find most suitable.

##Precomputing similarities

As your data set grows computing similarities between items may become slower and slower. That's why the recommender class has a calculate_similarities method, that when called will compute similarities for us. It's advised that similarities are precomputed and stored via the data handler class so that there would be no need to compute them each time recommendations are required.