clj-recommendation

A clojure recommendation library. Written for educational purposes but might be useful to someone.

In beta and changing frequently at the moment.

Install as:

[clj-recommendation "0.1.33"]

Require:

(:require [clj-recommendation :as rec])

Usage

The library defines a few data structures that can act as recommenders. In general when querying the query should be in the same format as the underlying data structure used by the recommender. In theory it should be easy to produce recommenders with different underlying data (i.e. databases) by programming at the protocols defined in the library.

Recommenders

The function knn-recommender will return a knnRecommender record when supplied with data, a distance calculation method keyword and the number of nearest neighbours to use in recommendations (k; default 5). The data should be a hash of hashes and attribute values should be numerical. Allowable distance calculation keywords are :pearson, :cosine, :euclidean and :manhattan. The functions nearest-neighbours and recommend can then be used to get nearest neighbours and recommendations:

user> music-example
{"Bill" {"Blues Traveler" 2.0, "Broken Bells" 3.5, "Deadmau5" 4.0,
"Phoenix" 2.0, "Slightly Stoopid" 3.5, "Vampire Weekend" 3.0}, "Chan" 
{"Blues Traveler" 5.0, "Broken Bells" 1.0, "Deadmau5" 1.0,
 "Norah Jones" 3.0, "Phoenix" 5, "Slightly Stoopid" 1.0}, ...}
user> (def nh (knn-recommender music-example :pearson))
#'clj-recommendation.core-test/nh
user> (nearest-neighbours nh {"query" {"Blues Traveler" 3.5, "Broken Bells" 2.0,
"Norah Jones" 4.5, "Phoenix" 5.0, "Slightly Stoopid" 1.5, "The Strokes" 2.5,
"Vampire Weekend" 2.0}})
(["Veronica" 0.829385976226374] ["Chan" 0.8197822947299414] ["Jordyn"
0.7639748605475432] ["Hailey" 0.42008402520840293] ["Sam" 0.2818672605010608])
user> (recommend nh {"query" {"Blues Traveler" 3.5, "Broken Bells" 2.0,
"Norah Jones" 4.5, "Phoenix" 5.0, "Slightly Stoopid" 1.5, "The Strokes" 2.5,
"Vampire Weekend" 2.0}})
"Deadmau5"
user> (recommendations nh {"query" {"Broken Bells" 4.0, "Deadmau5" 1.0,
 "Norah Jones" 4.0, "The Strokes" 4.0, "Vampire Weekend" 1.0}})
(["Phoenix" 5.0] ["Slightly Stoopid" 4.5])
>user

Both nearest-neighbour and recommendations return list of vectors with the first element of each vector corresponding to the name of the neighbour or recommendation and the second element the distance score or the predicted value of the recommendation based on the query attribute values, respectively. recommend returns a single value.

For item-based recommendations similarity matrix records can be defined using weighted-slope1-sim-matrix or adj-cosine-sim-matrix. Both accept a nested hash of hashes of names or classes and a hash of attributes. Attribute values should be numerical. The helper function invert-hash can be used to invert a hash for use in a similarity matrix. The recommend and predict functions can then be used to provide recommendations and predictions based on item similarity deviations:

user> (def wsm (weighted-slope1-sim-matrix (invert-hash music-example)))
#'clj-recommendation.core-test/wsm
user> (predict wsm "Blues Traveler" {"query" {"Broken Bells" 4.0,
 "Deadmau5" 1.0, "Norah Jones" 4.0, "The Strokes" 4.0, "Vampire Weekend" 1.0}})
3.526315789473684
user> (recommend wsm {"query" {"Broken Bells" 4.0, "Deadmau5" 1.0, "Norah Jones"
 4.0, "The Strokes" 4.0, "Vampire Weekend" 1.0}})
"Phoenix"
user> (recommendations recm {"query" {"Broken Bells" 4.0, "Deadmau5" 1.0, "Norah Jones"
 4.0, "The Strokes" 4.0, "Vampire Weekend" 1.0}})
(["Phoenix" 3.8541666666666665] ["Blues Traveler" 3.526315789473684]
 ["Slightly Stoopid" 2.8541666666666665])
user>

License

Copyright © 2017 Jason Mulvenna

Distributed under the Eclipse Public License either version 1.0 or (at your option) any later version.