个性化推荐算法系统（2）：基于二分图的个性化推荐召回算法personal rank（MovieLens数据集电影推荐）

#! /usr/bin/env python # -*- coding: utf-8 -*- # __author__ = "Q1mi" # Date: 2019/6/15 ''' personal rank main algo ''' from __future__ import division import numpy as np import read import operator import 矩阵personal_rank as mat_util from scipy.sparse.linalg import gmres def personal_rank(graph, root, alpha, iter_num, recom_num=10): ''' param graph:user item graph root: the recommend for the fixed user alpha: the probability to go random walk iter_num: iteration num recon_num: recommend item num :return: a dict,key:itemid , value:PR ''' rank = {} # 所有顶点对于root顶点的PR值 rank = {i: 0 for i in graph} rank[root] = 1 recon_result ={} # 输出 for iter_dex in range(iter_num): tmp_rank = {} # 记录该迭代轮次下，其他顶点对root顶点的PR值 tmp_rank = {i: 0 for i in graph} for out_userid, out_dict in graph.items(): # out_point是userid，out_dict是userid评分高于4.5分的电影字典{'item_10': 1} # 代表着二分图中用户指向物品的箭头 for inner_itemid, value in graph[out_userid].items(): # graph[out_point]:{'item_1196': 1} tmp_rank[inner_itemid] += round(alpha*rank[out_userid]/len(out_dict), 4) if inner_itemid == root: tmp_rank[inner_itemid] += round(1-alpha,4) if tmp_rank == rank: print('out'+str(iter_dex)) break rank = tmp_rank # 排序 right_num = 0 for zuhe in sorted(rank.items(), key=operator.itemgetter(1), reverse=True): point_item, pr_score = zuhe[0], zuhe[1] if len(point_item.split('_')) < 2: # 若这个顶点不是item顶点就过滤掉 continue if point_item in graph[root]: # 若这个item被当前的user行为过，就过滤 continue recon_result[point_item] = pr_score right_num += 1 if right_num > recom_num: break return recon_result def personal_rank_matrix(graph,root,alpha,recom_num=10): ''' 矩阵形式的 :param graph:user and item graph :param root: the fix user to recommend :param alpha: probability to random walk :param recom_num:item num :return: a dict:{key:item,value:pr_score} ''' m, vertex, address_dict = mat_util.graph_to_m(graph) # address_dict 所有顶点的行号 if root not in address_dict: return {} pr_score_dict = {} # 未排序pr值 pr_recom_dic = {} # 已排序pr值 mat_all = mat_util.mat_all_point(m,vertex,alpha) # A index = address_dict[root] initial_r0_list = [[0] for i in range(len(vertex))] print(initial_r0_list) initial_r0_list[index] = [1] r0_array = np.array(initial_r0_list) res = gmres(mat_all, r0_array, tol=1e-8)[0] # r = A*r0 # 存储、排序 pr值 for i in range(len(res)): point = vertex[i] if len(point.strip().split('_')) < 2: continue if point in graph[root]: continue pr_score_dict[point] = round(res[i], 3) for zuhe in sorted(pr_score_dict.items(), key=operator.itemgetter(1), reverse=True)[:recom_num]: point, pr_score = zuhe[0], zuhe[1] pr_recom_dic[point] = pr_score return pr_recom_dic def get_one_recom(): ''' give one fixed user , recommend result :return: ''' user = "1" alpha = 0.8 graph = read.get_graph_from_data('ratings.csv') iter_num = 100 pr_recom_result = personal_rank(graph, user, alpha, iter_num, 100) return pr_recom_result ''' item_info = read.get_item_info('movies.csv') # {itemid:[title,genre]} for itemid in graph[user]: pure_itemid = itemid.split('_')[1] print(item_info[pure_itemid]) print('--------------------------推荐结果-----------------------------') for itemid in recom_result: pure_itemid = itemid.split('_')[1] print(item_info[pure_itemid]) print(recom_result[itemid]) ''' def test_one_user_by_matrix(): ''' give one fixed user , recommend result,by matrix :return: ''' user = "1" alpha = 0.8 graph = read.get_graph_from_data('ratings.csv') pr_recom_resuit = personal_rank_matrix(graph, user, alpha, 100) return pr_recom_resuit if __name__ == "__main__": recom_resuit_base = get_one_recom() recom_resuit_matrix = test_one_user_by_matrix() num = 0 for i in recom_resuit_base: if i in recom_resuit_matrix: num += 1 print(num)