基于Pandas+余弦相似度+大数据智能护肤品推荐系统-机器学习算法应用(含Python工程源码)+测试数据集

import pandas as pd import numpy as np import math import itertools # 整体的推荐算法 class UserCF(): def __init__(self, df_fake, sub2_product): self.frame = df_fake self.product = sub2_product # self.screened_product_path = r'D:\work\dataclinic\fake\df_product1046.csv' #读取预筛选后的产品集 # self._init_data() # def _init_data(self): # self.frame = pd.read_csv(self.frame_path) # self.product = pd.read_csv(self.product_path,encoding='GB18030') # #self.screened_product_path = pd.read_csv(self.product_path,encoding='GB18030') def screen(self, need): self.frame = self.frame[(self.frame['诉求'].isin([need]))] def vec_purchase(self): # 提取购买记录并拉直 g = self.frame['购买记录'] g2 = self.frame['购买记录2'] g3 = self.frame['购买记录3'] wxid = list(self.frame['wechatid']) s = pd.Series(wxid, index=g) s2 = pd.Series(wxid, index=g2) s3 = pd.Series(wxid, index=g3) pin = pd.concat([s, s2, s3], axis=0) dict_pin = {'wechatid': pin.values, '购买记录': pin.index, } df2 = pd.DataFrame(dict_pin) # 拉直后的dataframe（wechat id ：购买记录） self.frame_p = df2[~(df2['购买记录'].isin([-1]))] # print(self.frame_p) # 计算肤质向量(T区油，U区油，敏感肌，痘痘肌)的余弦相似度 def cosine_skin(self, target_user_id, other_user_id): # 数据预处理 target_skin = [] other_skin = [] cols = ['T区油', 'U区油', '敏感肌', '痘痘肌'] for col in cols: target_skin.append((self.frame[self.frame['wechatid'] == target_user_id][col].values[0]) * 2 - 1) for col in cols: other_skin.append((self.frame[self.frame['wechatid'] == other_user_id][col].values[0]) * 2 - 1) # 计算余弦相似度 nume = sum(np.multiply(np.array(target_skin), np.array(other_skin))) # 分子 deno = sum(np.array(target_skin) ** 2) * sum(np.array(other_skin) ** 2) # 分母 cosine = nume / math.sqrt(deno) return cosine # 计算购买记录余弦相似度 def cosine_purchase(self, target_user_id, other_user_id): target_items = self.frame_p[self.frame_p['wechatid'] == target_user_id]['购买记录'] items = self.frame_p[self.frame_p['wechatid'] == other_user_id]['购买记录'] union_len = len(set(target_items) & set(items)) if union_len == 0: return 0.0 product = len(target_items) * len(items) cosine = union_len / math.sqrt(product) return cosine # 计算加权平均相似度并排序 def get_top_n_users(self, target_user_id, top_n): # 提取所有其他用户 other_users_id = [i for i in set(self.frame_p['wechatid']) if i != target_user_id] # 计算与所有其他用户的购买相似度 sim_purchase_list = [self.cosine_purchase(target_user_id, other_user_id) for other_user_id in other_users_id] # 计算与所有其他用户的肤质相似度 sim_skin_list = [self.cosine_skin(target_user_id, other_user_id) for other_user_id in other_users_id] # 加权平均（现在是各占50%） sim_list = list((np.array(sim_purchase_list) + np.array(sim_skin_list)) / 2) sim_list = sorted(zip(other_users_id, sim_list), key=lambda x: x[1], reverse=True) return sim_list[:top_n] # 提取候选产品表 def get_candidates_items(self, target_user_id): target_user_item = set(self.frame_p[self.frame_p['wechatid'] == target_user_id]['购买记录']) other_user_item = set(self.frame_p[self.frame_p['wechatid'] != target_user_id]['购买记录']) candidates_item = other_user_item - target_user_item # 寻找候选推荐品标准：目标用户没有使用过的（必要性存疑） candidates_item = list(candidates_item & set(self.product['ind'].values)) # 候选推荐品必须属于上一步预筛选出的项目(目前使用全产品表代替筛选后产品表) return candidates_item # 计算用户兴趣程度 def get_top_n_items(self, top_n_users, candidates_items, top_n): top_n_user_data = [self.frame_p[self.frame_p['wechatid'] == k] for k, _ in top_n_users] interest_list = [] for ind in candidates_items: tmp = [] for user_data in top_n_user_data: if ind in user_data['购买记录'].values: tmp.append(1) else: tmp.append(0) interest = sum([top_n_users[i][1] * tmp[i] for i in range(len(top_n_users))]) interest_list.append((ind, interest)) interest_list = sorted(interest_list, key=lambda x: x[1], reverse=True) return interest_list[:top_n] # 输入wxid，需求 默认推荐产品数为10件 输出有序推荐产品 ##################################@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@############3 def calculate(self, target_user): top_n = self.product.shape[0] target_user_id = target_user.wxid need = target_user.诉求 self.screen(need) self.vec_purchase() top_n_users = self.get_top_n_users(target_user_id, top_n) candidates_items = self.get_candidates_items(target_user_id) top_n_items = self.get_top_n_items(top_n_users, candidates_items, top_n) # 重构数据格式 返回完整推荐产品信息 productlist = [top_n_items[i][0] for i in range(len(top_n_items))] product_rec = self.product[(self.product['ind'].isin(productlist))] product_rec['InterestRate'] = [top_n_items[i][1] for i in range(len(top_n_items))] return product_rec # 组合的推荐算法 class CombRating(): def __init__(self,user, pro_withrate, fformula): self.user = user self.product = pro_withrate self.fformula=fformula # 第一个for 找到用户的诉求是哪一种 四个属性全部对上才行 # 第二个for 找到组合中应当有的产品类型，水 乳 霜 祛痘凝胶 洁面 def find_kind(self): #print(self.fformula) n_formula = self.fformula.shape[0] for i in range(n_formula): if (self.user.诉求 == self.fformula.诉求[i]) \ and (self.user.T区油 == self.fformula.T区油[i]) \ and (self.user.U区油 == self.fformula.U区油[i]) \ and (self.user.敏感肌 == self.fformula.敏感肌[i]): i_formula = i break # 此处利用总共的产品种类解决数字的问题 但是我觉得数据形式变化才能根本解决问题 毕竟在一个表里面 不好分割两边 # 寻找第一个是产品类型的列 并记录此前经过的列数 form_list = [] total_pro_type = ['水', '乳', '霜', '祛痘凝胶', '洁面'] type_number = 0 for j in range(len(self.fformula.columns)): if self.fformula.columns[j] in total_pro_type: break else: type_number = type_number + 1 # 再找到所有需要的产品种类 for j in range(type_number, len(self.fformula.columns)): if (self.fformula.loc[i_formula][j] == 1): form_list.append(self.fformula.columns[j]) return form_list def outer_multiple(self, form_list): ddict={} for i in range(len(form_list)): ddict[form_list[i]] = list(self.product[self.product.剂型 == form_list[i]].ind) # print(ddict) dd = [] for i in itertools.product(*ddict.values()): dd.append(i) comb_pd = pd.DataFrame(dd) # 为DF的每一列添加名称 column_name = [] for i in range(len(comb_pd.columns)):