The module is designed to support the prototype development and evaluation of user recommendation system. The current version includes following algorithm:
-
Nearest Neigbour User Recommendation System (Distance-based): utilize the user profile information only, with customized or generic distance metrics.
-
Clustering-based User Recommendation System: utilize the user profile information only, with clustering algorithm to improve the efficiency of searching suggestions.
-
Learning Distance Metrics-based User Recommendation System: utilize both user profile information and existing social connections, requiring more computational resource.
- UserRecommenderMixin
- NNUserRecommender
- PairwiseDistMatrix
- Conduct an experiment of applying Nearest Neigbour-based recommendation system:
from user_recommender_framework.network_simulator import *
from user_recommender_framework.user_recommender import *
# load data for experimentation
# 1. user_ids: all users ids
# 2. user_profiles: user profiles
# 3. user_connections: list of user connections
# 4. init_user_connections: sample user connections which experiment starts with
# ...
# prepare components for experiment
nnu_recommender = NNUserRecommender(user_ids, user_profiles, init_user_connections)
evaluator = SocialNetworkEvaluator()
evaluator.load_ref_user_connections(user_connections)
# user behavior simulator
user_clicker = UserClickSimulator()
# setup experiment
experimentor = UserRecSysExpSimulator(name="MyExperiment")
experimentor.load_recommender(nnu_recommender)
experimentor.load_evaluator(evaluator)
experimentor.load_clicker(user_clicker)
# set the number of suggestions for each user at each iteration
experimentor.set_recommendation_size(5)
# start experiment, experiment results will be exported automatically
experimentor.run()
# to retore experiment to status before .run()
experimentor.reset()When the experiment is running, .run() method will report the progress of experimenting in console.
