基于Q-Learning算法实现的论文推荐系统python源码(带数据和论文).zip

import csv import time import agent as ag import pandas as pd import numpy as np import other as ot from intro_paper import con_sim_matrix from user_in import dis_cf_matrix, user_paper, find_suit_user from article_list import art_array Q = np.random.random((7, 7)) # Q表初始化 initial the Q table # 给action表赋值 initial the Action table action = [[1, 1, 1], [1, 2, 3], [1, 3, 2], [2, 1, 3], [2, 3, 1], [3, 1, 2], [3, 2, 1]] if __name__ == "__main__": time_start = time.time() filename = r'C:\Users\严书航\Desktop\data.csv' data = [] with open(filename, "r", encoding='utf-8', errors='ignore') as f: csv_reader = csv.reader(f) header = next(csv_reader) for row in csv_reader: data.append(row) data = np.array(data) data1 = data[:, 1] # 标题数据 titles data2 = data[:, 4] # 摘要数据 abstracts data3 = pd.read_csv("user-info.csv", delimiter=',') title_mat = con_sim_matrix(data1) abs_mat = con_sim_matrix(data2) user_mat, item_mat = dis_cf_matrix(data3) time_end = time.time() print("<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<") print('time cost: ', time_end - time_start, 's') user = find_suit_user(200, data3) # 挑选样本用户 select the sample users print(len(user)) number = 20 # 每次推荐文章数 the number of paper recommand everytime turn = 25 # 推荐次数 the recommendation times recall_list = np.zeros((1, turn)) precision_list = np.zeros((1, turn)) out = pd.DataFrame() for x in user: docs_all = user_paper(x, data3) # 提取读者所有文章 get the paper that the user has read doc_history = list(docs_all[:number]) # 选择作为历史数据的部分 choose part of them as historial data len_doc_all = len(docs_all) paperlist = art_array(title_mat, abs_mat, item_mat, doc_history) # 根据用户历史计算得到的各论文得分 get the score of paper by historical data doc_r = ag.recomender(a=0, action=action, docs=paperlist, history=doc_history) # 用平均分方法得到第一次推荐 achieve the first recommandation by mean a = 0 # action初始值 the initial number of Action count = 0 # 推荐成功篇数计数 counter for the number of successful recomandation recall_user = [] # 初始化该用户召回率 initialize the recall rate precision_user = [] # 初始化该用户准确率 initialize the accuracy rate doc_except = list(doc_history) for i in range(turn): doc_s = ag.user(doc_r, docs_all) # 用户选择论文 user choose the paper #print(doc_s) state = ag.state(doc_s, paperlist) # 选择的论文状态 the chosen paper's status reply = len(doc_s) # 选择的论文数 the number of chosen paper count += reply recall = count/len_doc_all precision = count/((i+1)*number) recall_user.append(recall) precision_user.append(precision) Q = ag.renewQ(state, a, reply, Q) # 更新Q表 renew the Q table if len(doc_s) != 0: doc_history = doc_history + doc_s # 更新用户喜爱的论文 renew the paper that the user likes doc_except = doc_except + list(doc_r) # 不推荐的文章集合 the collection of paper that user dislikes paperlist = art_array(title_mat, abs_mat, item_mat, doc_history[-number:]) # 根据用户历史计算得到的各论文得分 calculate the score a = ag.agent(state, Q) doc_r = ag.recomender(a=a, action=action, docs=paperlist, history=doc_except, m=number) recall_list = recall_list + np.array(recall_user) precision_list = precision_list + np.array(precision_user) # print(recall_list) time_end = time.time() print("<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<") print('time cost: ', time_end - time_start, 's') recall_list = recall_list[0] / len(user) precision_list = precision_list[0] / len(user) out['recall'] = recall_list out['pre'] = precision_list print(Q) # 参照实验 control experiment colum = ['x1', 'x2', 'x3', 'mix'] for z in colum: recall_list = np.zeros((1, turn)) precision_list = np.zeros((1, turn)) for x in user: docs_all = user_paper(x, data3) # 提取读者所有文章 get the paper doc_history = list(docs_all[:number]) # 选择作为历史数据的部分 select historical part len_doc_all = len(docs_all) paperlist = art_array(title_mat, abs_mat, item_mat, doc_history) # 根据用户历史计算得到的各论文得分 caculate the score recall_user = [] # 初始化该用户召回率 initialize the recall rate precision_user = [] # 初始化该用户准确率 initialize the accuracy rate for j in range(turn): num = number * (j + 1) doc_rr = ot.recomender(paperlist, z, doc_history, num) sn = ot.select_num(doc_rr, docs_all) recall = sn / len_doc_all precision = sn / num recall_user.append(recall) precision_user.append(precision) recall_list = recall_list + np.array(recall_user) precision_list = precision_list + np.array(precision_user) recall_list = recall_list[0] / len(user) precision_list = precision_list[0] / len(user) out['recall_{}'.format(z)] = recall_list out['pre_{}'.format(z)] = precision_list out.to_csv('pre_recall.csv') # 输出结果 output the results