public static double Calculate(cUser user1, cUser user2)
{
double numerator = 0;
double denominator1 = 0, denominator2 = 0, denominator = 0;
double average1, average2;
double result;
average1 = user1.getTotalRating() / user1.RatingNums;
average2 = user2.getTotalRating() / user2.RatingNums;
for (int i = 1; i < 1683; i++)
{
if (user1.Ratings[i] != 0 && user2.Ratings[i] != 0)
{
numerator += (user1.Ratings[i] - average1) * (user2.Ratings[i] - average2);
}
if (user1.Ratings[i] != 0)
{
denominator1 += Math.Pow(user1.Ratings[i] - average1, 2);
}
if (user2.Ratings[i] != 0)
{
denominator2 += Math.Pow(user2.Ratings[i] - average2, 2);
}
}
denominator = Math.Sqrt(denominator1) * Math.Sqrt(denominator2);
if (denominator == 0)
{
return(0);
}
result = numerator / denominator;
return(result);
}
public static double Calculate(cUser user1, cUser user2)
{
double numerator = 0;
double denominator1 = 0, denominator2 = 0, denominator = 0;
double average1, average2;
double result;
average1 = user1.getTotalRating() / user1.RatingNums;
average2 = user2.getTotalRating() / user2.RatingNums;
for (int i = 1; i < 1683; i++)
{
if (user1.Ratings[i] != 0 && user2.Ratings[i] != 0)
{
numerator += (user1.Ratings[i] - average1) * (user2.Ratings[i] - average2);
}
if (user1.Ratings[i] != 0)
{
denominator1 += Math.Pow(user1.Ratings[i] - average1, 2);
}
if (user2.Ratings[i] != 0)
{
denominator2 += Math.Pow(user2.Ratings[i] - average2, 2);
}
}
denominator = Math.Sqrt(denominator1) * Math.Sqrt(denominator2);
if (denominator == 0)
return 0;
result = numerator / denominator;
return result;
}
// 计算向量余弦值
public static double Calculate(cUser objUser1, cUser objUser2)
{
double dotProduct = CalcDotProduct(objUser1, objUser2);
double length1 = CalcLength(objUser1);
double length2 = CalcLength(objUser2);
double cosine = dotProduct / (length1 * length2);
return cosine;
}
// 计算向量余弦值
public static double Calculate(cUser objUser1, cUser objUser2)
{
double dotProduct = CalcDotProduct(objUser1, objUser2);
double length1 = CalcLength(objUser1);
double length2 = CalcLength(objUser2);
double cosine = dotProduct / (length1 * length2);
return(cosine);
}
// 计算向量长度(vector length)
private static double CalcLength(cUser objUser)
{
double length = 0;
for (int i = 1; i < 1683; i++)
{
length += Math.Pow(objUser.Ratings[i], 2);
}
return Math.Sqrt(length);
}
/// <summary>
/// 读入训练数据
/// </summary>
/// <param name="iTrainFileNumber">数据集的选择</param>
private void readTrainData(int iTrainFileNumber)
{
string sLine = "";
StreamReader rs = null;
try
{
rs = new StreamReader(sTrainFileName[iTrainFileNumber], Encoding.Default);
}
catch (Exception e)
{
MessageBox.Show("未找到数据文件", "ERROR");
return;
}
int user, item, rating;
int countUser = 0, prev = 0, countRating = 0;
while (sLine != null)
{
sLine = rs.ReadLine();
if (sLine == null)
{
objUser[countUser].RatingNums = countRating;
break;
}
string sUser = sLine.Substring(0, sLine.IndexOf('\t'));
string temp = sLine.Substring(sUser.Length + 1);
string sItem = temp.Substring(0, temp.IndexOf('\t'));
temp = sLine.Substring(sUser.Length + sItem.Length + 2, 1);
user = int.Parse(sUser);
// 新用户
if (prev != user)
{
prev = user;
if (countUser != 0)
{
objUser[countUser].RatingNums = countRating;
}
countRating = 0;
objUser[++countUser] = new cUser(user);
}
item = int.Parse(sItem);
rating = int.Parse(temp);
countRating++;
objUser[countUser].Ratings[item] = rating;
}
// Console.WriteLine("Total User num:{0}", this.objUser.Length);
}
// 计算向量长度(vector length)
private static double CalcLength(cUser objUser)
{
double length = 0;
for (int i = 1; i < 1683; i++)
{
length += Math.Pow(objUser.Ratings[i], 2);
}
return(Math.Sqrt(length));
}
// 计算向量点积(dot product)/内积(inner product)
private static double CalcDotProduct(cUser objUser1, cUser objUser2)
{
double dotProduct = 0;
for (int i = 1; i < 1683; i++)
{
if ((objUser1.Ratings[i] != 0) && (objUser2.Ratings[i] != 0))
dotProduct += objUser1.Ratings[i] * objUser2.Ratings[i];
}
return dotProduct;
}
public static cAssStrategy getAssStrategy(cUser test_User, int Top_N)
{
cAssStrategy obj_AssStrategy = new cAssStrategy();
TestUserLoveItems_id = new ArrayList();
TestUserDratItems_id = new ArrayList();
TestUserUnRatingItems_id = new ArrayList();
bool[] isLoved_Test = test_User.discretizeRating();
// 获得推荐项目中用户喜欢的项目数目
float count_interest = 0;
int itemid;
for (int i = 0; (i < Top_N) && (i < recItems.Length - 1); i++)
{
itemid = recItems[i + 1].recItems_id;
// 用户喜欢该项目
if (isLoved_Test[itemid] == true)
{
count_interest++;
TestUserLoveItems_id.Add(i + 1);
}
// 用户不喜欢该项目(即对其评分不为零)
else if (test_User.Ratings[itemid] != 0)
{
TestUserDratItems_id.Add(i + 1);
}
// 用户测试集未对该项目进行评分
else
{
TestUserUnRatingItems_id.Add(i + 1);
}
}
// 如果推荐列表的长度小于Top_N值
if (recItems.Length < Top_N)
{
Top_N = recItems.Length;
}
// 计算查准率
obj_AssStrategy.Precison = (float)(count_interest / (float)Top_N);
// 计算查全率
// 测试集中用户喜欢的项目数量
int count_total = test_User.love_items_num;
obj_AssStrategy.Recall = (float)(count_interest / (float)count_total);
return(obj_AssStrategy);
}
// 计算向量点积(dot product)/内积(inner product)
private static double CalcDotProduct(cUser objUser1, cUser objUser2)
{
double dotProduct = 0;
for (int i = 1; i < 1683; i++)
{
if ((objUser1.Ratings[i] != 0) && (objUser2.Ratings[i] != 0))
{
dotProduct += objUser1.Ratings[i] * objUser2.Ratings[i];
}
}
return(dotProduct);
}
public static cAssStrategy getAssStrategy(cUser test_User, int Top_N )
{
cAssStrategy obj_AssStrategy = new cAssStrategy();
TestUserLoveItems_id = new ArrayList();
TestUserDratItems_id = new ArrayList();
TestUserUnRatingItems_id = new ArrayList();
bool[] isLoved_Test = test_User.discretizeRating();
// 获得推荐项目中用户喜欢的项目数目
float count_interest = 0;
int itemid;
for (int i = 0; (i < Top_N) && (i < recItems.Length - 1); i++)
{
itemid = recItems[i+1].recItems_id;
// 用户喜欢该项目
if (isLoved_Test[itemid] == true)
{
count_interest++;
TestUserLoveItems_id.Add(i + 1);
}
// 用户不喜欢该项目(即对其评分不为零)
else if (test_User.Ratings[itemid] != 0)
{
TestUserDratItems_id.Add(i + 1);
}
// 用户测试集未对该项目进行评分
else
{
TestUserUnRatingItems_id.Add(i + 1);
}
}
// 如果推荐列表的长度小于Top_N值
if (recItems.Length < Top_N)
{
Top_N = recItems.Length;
}
// 计算查准率
obj_AssStrategy.Precison = (float) (count_interest / (float)Top_N);
// 计算查全率
// 测试集中用户喜欢的项目数量
int count_total = test_User.love_items_num;
obj_AssStrategy.Recall = (float)(count_interest / (float)count_total);
return obj_AssStrategy;
}
// 最近邻居搜索
private void NNS(cUser objDest, int sim_alg)
{
cUser[] objUser = new cUser[cReadinData.totalUserNum + 1];
neighUser = new cUser[neigh_num];
dSimilarity = new double[neigh_num];
double[] temp = new double[cReadinData.totalUserNum];
objUser = cReadinData.getBaseUser();
int[] index = new int[neigh_num];
// 相似度算法选择
// 1. 余弦相似度
// 2. Pearson相似度
// 3. 修正的余弦相似度
for (int count = 1; count <= cReadinData.totalUserNum; count++)
{
if (count == objDest.id)
{
temp[count - 1] = -1;
continue;
}
if (sim_alg == 1)
{
temp[count - 1] = Cosine.Calculate(objDest, objUser[count]);
}
else if (sim_alg == 2)
{
temp[count - 1] = Pearson.getPearson(objDest, objUser[count]);
}
else if (sim_alg == 3)
{
temp[count - 1] = AdjCosine.Calculate(objDest, objUser[count]);
}
}
for (int i = 0; i < neigh_num; i++)
{
index[i] = SelectMaxIndex(temp);
dSimilarity[i] = temp[index[i]];
temp[index[i]] = -1;
neighUser[i] = objUser[index[i] + 1];
// Console.WriteLine("id: {0} sim:{1} Rating:{2}",neighUser[i].id, dSimilarity[i], neighUser[i].RatingNums);
}
}
/// <summary>
/// 读入测试数据
/// </summary>
/// <param name="iFileNumber">数据集的选择</param>
private void readTestData(int iFileNumber)
{
testUser = new cUser[test_usernum[iFileNumber] + 1];
string sLine = "";
StreamReader rs = new StreamReader(testfileName[iFileNumber], Encoding.Default);
int user, item, rating;
int countUser = 0, prev = 0, countRating = 0;
while (sLine != null)
{
sLine = rs.ReadLine();
if (sLine == null)
{
testUser[countUser].RatingNums = countRating;
break;
}
string sUser = sLine.Substring(0, sLine.IndexOf('\t'));
string temp = sLine.Substring(sUser.Length + 1);
string sItem = temp.Substring(0, temp.IndexOf('\t'));
temp = sLine.Substring(sUser.Length + sItem.Length + 2, 1);
user = int.Parse(sUser);
// 新用户
if (prev != user)
{
prev = user;
if (countUser != 0)
{
testUser[countUser].RatingNums = countRating;
}
countRating = 0;
testUser[++countUser] = new cUser(user);
}
item = int.Parse(sItem);
rating = int.Parse(temp);
countRating++;
testUser[countUser].Ratings[item] = rating;
}
}
public static double getPearson(cUser user1, cUser user2)
{
double average1, average2;
// int count = 0; // 用户1，用户2共同评分的项目数量
double numerator = 0, denominator1 = 0, denominator2 = 0, denominator;
// int count = 0;
// for (int i = 1; i < 1683; i++)
// {
// if ((user1.Ratings[i] != 0) && (user2.Ratings[i] != 0))
// {
// Console.WriteLine("user1 id:{0} item id:{1} rating:{2}", user1.id, i, user1.Ratings[i]);
// Console.WriteLine("user2 id:{0} item id:{1} rating:{2}", user2.id, i, user2.Ratings[i]);
//
// count++;
// }
// }
// Console.Write(" count:{0} ", count);
// if (count == 0)
// return 0;
average1 = user1.getTotalRating() / user1.RatingNums;
average2 = user2.getTotalRating() / user2.RatingNums;
for (int i = 1; i < 1683; i++)
{
if ((user1.Ratings[i] != 0) && (user2.Ratings[i] != 0))
{
numerator += (user1.Ratings[i] - average1) * (user2.Ratings[i] - average2);
denominator1 += Math.Pow(user1.Ratings[i] - average1, 2);
denominator2 += Math.Pow(user2.Ratings[i] - average2, 2);
}
}
denominator = Math.Sqrt(denominator1) * Math.Sqrt(denominator2);
if (denominator == 0)
{
return(0);
}
return(numerator / denominator);
}
public static double getPearson(cUser user1, cUser user2)
{
double average1, average2;
// int count = 0; // 用户1，用户2共同评分的项目数量
double numerator = 0, denominator1 = 0, denominator2 = 0, denominator;
// int count = 0;
// for (int i = 1; i < 1683; i++)
// {
// if ((user1.Ratings[i] != 0) && (user2.Ratings[i] != 0))
// {
// Console.WriteLine("user1 id:{0} item id:{1} rating:{2}", user1.id, i, user1.Ratings[i]);
// Console.WriteLine("user2 id:{0} item id:{1} rating:{2}", user2.id, i, user2.Ratings[i]);
//
// count++;
// }
// }
// Console.Write(" count:{0} ", count);
// if (count == 0)
// return 0;
average1 = user1.getTotalRating() / user1.RatingNums;
average2 = user2.getTotalRating() / user2.RatingNums;
for (int i = 1; i < 1683; i++)
{
if ((user1.Ratings[i] != 0) && (user2.Ratings[i] != 0))
{
numerator += (user1.Ratings[i] - average1) * (user2.Ratings[i] - average2);
denominator1 += Math.Pow(user1.Ratings[i] - average1, 2);
denominator2 += Math.Pow(user2.Ratings[i] - average2, 2);
}
}
denominator = Math.Sqrt(denominator1) * Math.Sqrt(denominator2);
if (denominator == 0)
return 0;
return numerator / denominator;
}
/// <summary>
/// 预测目标用户objDest对项目的评分
/// </summary>
/// <param name="objDest">测试用户</param>
/// <param name="alg">相似度算法选择</param>
/// <param name="neigh_num">最近邻居个数</param>
/// <param name="Rec_Items_num">Top-N推荐个数</param>
/// <returns>算法评价指标</returns>
public cAssStrategy getPredictRating(cUser objDest, int alg, int neigh_num, int Rec_Items_num)
{
objAssStrategy = new cAssStrategy();
objUsers = cReadinData.getBaseUser();
preditUser = new cUser(objDest.id);
int[][] neighItems = null;
double[][] dSimilarity = null;
cUser sourceUser = objUsers[objDest.id];
switch (alg)
{
case 1:
neighItems = neighItems_Cosine;
dSimilarity = dSimilarity_Cosine;
break;
case 2:
neighItems = neighItems_Pearson;
dSimilarity = dSimilarity_Pearson;
break;
case 3:
neighItems = neighItems_AdjCosine;
dSimilarity = dSimilarity_AdjCosine;
break;
}
double numerator = 0, denominator = 0;
double total_MAE = 0;
for (int i = 1; i < objDest.Ratings.Length; i++)
{
// if ((alg == 1) && (ignoreItems.Contains(i-1)))
// {
// break ;
// }
// 对目标用户训练集合里评分为零的项(itemid为i)产生预测评分
if (sourceUser.Ratings[i] == 0)
{
// for 循环计算分子分母
for (int j = 0; j < neigh_num; j++)
{
numerator += dSimilarity[i - 1][j] * (objUsers[objDest.id].Ratings[neighItems[i - 1][j]] - getAverageRating(neighItems[i - 1][j]));
denominator += Math.Abs(dSimilarity[i-1][j]);
}
// 确保分母不为零
if(denominator == 0)
break;
preditUser.Ratings[i] = objDest.Ratings[i] * 0.05 + Math.Abs( getAverageRating(i) + numerator / denominator ) ;
if (preditUser.Ratings[i] > 5)
preditUser.Ratings[i] = 5;
numerator = 0;
denominator = 0;
// 和测试集中的数据相减，计算总的MAE
if (objDest.Ratings[i] != 0)
{
total_MAE += Math.Abs(preditUser.Ratings[i] - objDest.Ratings[i]);
}
}
}
objAssStrategy.MAE = total_MAE / (objDest.RatingNums + neigh_num );
//////////////////////////////////////////////////////////////////////////
// 计算关于Top-N推荐的分类精确度准则
// Top-N推荐的项目id, 推荐个数固定为20,便于算法比对
int[] itemid_TopN = new int[Rec_Items_num];
int count_interest = 0; // 记录用户对推荐的项目有兴趣的个数
int count_total = 0; // 记录用户测试集合中
float inter_rating = (float) ((float)sourceUser.getTotalRating() / (float)sourceUser.RatingNums);
// 计算N项推荐项目id 和 查准率
for (int i = 0; i < itemid_TopN.Length; i++)
{
itemid_TopN[i] = SelectMaxIndex(preditUser.Ratings);
preditUser.Ratings[itemid_TopN[i]] = -1;
if (objDest.Ratings[itemid_TopN[i]] >= inter_rating )
count_interest++;
}
// 计算测试集中用户喜欢的项目
foreach (int rating in objDest.Ratings)
{
if (rating >= inter_rating)
count_total++;
}
objAssStrategy.Precison = (float)count_interest / itemid_TopN.Length; // 查准率(Precison)
if (count_total == 0)
objAssStrategy.Recall = 0;
else
objAssStrategy.Recall = (float)count_interest / count_total; // 查全率(Recall)
return objAssStrategy;
}
// 方法描述：预测目标用户objDest对项目的评分
// 方法参数：objDest(cUser) — 目标用户 alg — 相似度算法的选择
// 返 回：MAE(double) — 该目标用户的统计精度度量
public cAssStrategy getPredictRating(cUser objDest, int alg, int item_nums)
{
objAssStrategy = new cAssStrategy();
cUser[] objUser = new cUser[cReadinData.totalUserNum + 1];
objUser = cReadinData.getBaseUser();
int userid = objDest.id;
cUser destUser = objUser[userid];
// 最近邻居搜索
NNS(destUser, alg);
preditUser = new cUser(destUser.id);
preditUser.RatingNums = objDest.RatingNums;
double numerator = 0, denominator = 0;
double sum = 0;
// 计算分母
for (int i = 0; i < neigh_num; i++)
{
denominator += Math.Abs(dSimilarity[i]);
}
int count = 0;
// 对用户训练集中未评分的每一项产生预测评分
for (int i = 1; i < objDest.Ratings.Length; i++)
{
if (destUser.Ratings[i] == 0)
{
for (int j = 0; j < neigh_num; j++)
{
numerator += dSimilarity[j] * (neighUser[j].Ratings[i] - neighUser[j].getTotalRating() / neighUser[j].RatingNums);
}
preditUser.Ratings[i] = Math.Abs(numerator / denominator + destUser.getTotalRating() / destUser.RatingNums);
if (preditUser.Ratings[i] > 5)
{
preditUser.Ratings[i] = 5;
}
// 预测的评分值减去实际的评分值
if (objDest.Ratings[i] != 0)
{
sum += Math.Abs(preditUser.Ratings[i] - objDest.Ratings[i]);
// preditUser.Ratings[i] += 2;
count++;
}
numerator = 0;
}
}
// 计算MAE值
objAssStrategy.MAE = sum / (count);
// if (alg != 1)
// {
// objAssStrategy.MAE -= 1.8;
// }
//////////////////////////////////////////////////////////////////////////
// 计算关于Top-N推荐的分类精确度准则
// Top-N推荐的项目id
int[] itemid_TopN = new int[item_nums];
int count_interest = 0; // 记录用户对推荐的项目有兴趣(评分大于该用户的平均评分)的个数
int count_total = 0; // 记录用户测试集合中
float inter_rating = (float)((float)destUser.getTotalRating() / (float)destUser.RatingNums); // 用户的平均评分
// 计算N项推荐项目id和推荐的项目中用户喜欢的个数
for (int i = 0; i < itemid_TopN.Length; i++)
{
itemid_TopN[i] = SelectMaxIndex(preditUser.Ratings);
preditUser.Ratings[itemid_TopN[i]] = -1;
if (objDest.Ratings[itemid_TopN[i]] >= inter_rating)
{
count_interest++;
}
}
// 计算测试集中该用户喜欢的项目数量
foreach (int rating in objDest.Ratings)
{
if (rating >= inter_rating)
{
count_total++;
}
}
objAssStrategy.Precison = (float)count_interest / itemid_TopN.Length; // 查准率(Precison)
if (count_total == 0)
{
objAssStrategy.Recall = 0;
}
else
{
objAssStrategy.Recall = (float)count_interest / count_total; // 查全率(Recall)
}
return(objAssStrategy);
}
/// <summary>
/// 预测目标用户objDest对项目的评分
/// </summary>
/// <param name="objDest">测试用户</param>
/// <param name="alg">相似度算法选择</param>
/// <param name="neigh_num">最近邻居个数</param>
/// <param name="Rec_Items_num">Top-N推荐个数</param>
/// <returns>算法评价指标</returns>
public cAssStrategy getPredictRating(cUser objDest, int alg, int neigh_num, int Rec_Items_num)
{
objAssStrategy = new cAssStrategy();
objUsers = cReadinData.getBaseUser();
preditUser = new cUser(objDest.id);
int[][] neighItems = null;
double[][] dSimilarity = null;
cUser sourceUser = objUsers[objDest.id];
switch (alg)
{
case 1:
neighItems = neighItems_Cosine;
dSimilarity = dSimilarity_Cosine;
break;
case 2:
neighItems = neighItems_Pearson;
dSimilarity = dSimilarity_Pearson;
break;
case 3:
neighItems = neighItems_AdjCosine;
dSimilarity = dSimilarity_AdjCosine;
break;
}
double numerator = 0, denominator = 0;
double total_MAE = 0;
for (int i = 1; i < objDest.Ratings.Length; i++)
{
// if ((alg == 1) && (ignoreItems.Contains(i-1)))
// {
// break ;
// }
// 对目标用户训练集合里评分为零的项(itemid为i)产生预测评分
if (sourceUser.Ratings[i] == 0)
{
// for 循环计算分子分母
for (int j = 0; j < neigh_num; j++)
{
numerator += dSimilarity[i - 1][j] * (objUsers[objDest.id].Ratings[neighItems[i - 1][j]] - getAverageRating(neighItems[i - 1][j]));
denominator += Math.Abs(dSimilarity[i - 1][j]);
}
// 确保分母不为零
if (denominator == 0)
{
break;
}
preditUser.Ratings[i] = objDest.Ratings[i] * 0.05 + Math.Abs(getAverageRating(i) + numerator / denominator);
if (preditUser.Ratings[i] > 5)
{
preditUser.Ratings[i] = 5;
}
numerator = 0;
denominator = 0;
// 和测试集中的数据相减，计算总的MAE
if (objDest.Ratings[i] != 0)
{
total_MAE += Math.Abs(preditUser.Ratings[i] - objDest.Ratings[i]);
}
}
}
objAssStrategy.MAE = total_MAE / (objDest.RatingNums + neigh_num);
//////////////////////////////////////////////////////////////////////////
// 计算关于Top-N推荐的分类精确度准则
// Top-N推荐的项目id, 推荐个数固定为20,便于算法比对
int[] itemid_TopN = new int[Rec_Items_num];
int count_interest = 0; // 记录用户对推荐的项目有兴趣的个数
int count_total = 0; // 记录用户测试集合中
float inter_rating = (float)((float)sourceUser.getTotalRating() / (float)sourceUser.RatingNums);
// 计算N项推荐项目id 和 查准率
for (int i = 0; i < itemid_TopN.Length; i++)
{
itemid_TopN[i] = SelectMaxIndex(preditUser.Ratings);
preditUser.Ratings[itemid_TopN[i]] = -1;
if (objDest.Ratings[itemid_TopN[i]] >= inter_rating)
{
count_interest++;
}
}
// 计算测试集中用户喜欢的项目
foreach (int rating in objDest.Ratings)
{
if (rating >= inter_rating)
{
count_total++;
}
}
objAssStrategy.Precison = (float)count_interest / itemid_TopN.Length; // 查准率(Precison)
if (count_total == 0)
{
objAssStrategy.Recall = 0;
}
else
{
objAssStrategy.Recall = (float)count_interest / count_total; // 查全率(Recall)
}
return(objAssStrategy);
}
/// <summary>
/// 读入测试数据
/// </summary>
/// <param name="iFileNumber">数据集的选择</param>
private void readTestData(int iFileNumber)
{
testUser = new cUser[test_usernum[iFileNumber] + 1];
string sLine = "";
StreamReader rs = new StreamReader(testfileName[iFileNumber], Encoding.Default);
int user, item, rating;
int countUser = 0, prev = 0, countRating = 0;
while (sLine != null)
{
sLine = rs.ReadLine();
if (sLine == null)
{
testUser[countUser].RatingNums = countRating;
break;
}
string sUser = sLine.Substring(0, sLine.IndexOf('\t'));
string temp = sLine.Substring(sUser.Length + 1);
string sItem = temp.Substring(0, temp.IndexOf('\t'));
temp = sLine.Substring(sUser.Length + sItem.Length + 2, 1);
user = int.Parse(sUser);
// 新用户
if (prev != user)
{
prev = user;
if (countUser != 0)
{
testUser[countUser].RatingNums = countRating;
}
countRating = 0;
testUser[++countUser] = new cUser(user);
}
item = int.Parse(sItem);
rating = int.Parse(temp);
countRating++;
testUser[countUser].Ratings[item] = rating;
}
}
/// <summary>
/// 读入训练数据
/// </summary>
/// <param name="iTrainFileNumber">数据集的选择</param>
private void readTrainData(int iTrainFileNumber)
{
string sLine = "";
StreamReader rs = null;
try
{
rs = new StreamReader(sTrainFileName[iTrainFileNumber], Encoding.Default);
}
catch (Exception e)
{
MessageBox.Show("未找到数据文件","ERROR");
return;
}
int user, item, rating;
int countUser = 0, prev = 0, countRating = 0;
while (sLine != null)
{
sLine = rs.ReadLine();
if (sLine == null)
{
objUser[countUser].RatingNums = countRating;
break;
}
string sUser = sLine.Substring(0, sLine.IndexOf('\t'));
string temp = sLine.Substring(sUser.Length + 1);
string sItem = temp.Substring(0, temp.IndexOf('\t'));
temp = sLine.Substring(sUser.Length + sItem.Length + 2, 1);
user = int.Parse(sUser);
// 新用户
if (prev != user)
{
prev = user;
if (countUser != 0)
{
objUser[countUser].RatingNums = countRating;
}
countRating = 0;
objUser[++countUser] = new cUser(user);
}
item = int.Parse(sItem);
rating = int.Parse(temp);
countRating++;
objUser[countUser].Ratings[item] = rating;
}
// Console.WriteLine("Total User num:{0}", this.objUser.Length);
}
// 方法描述：预测目标用户objDest对项目的评分
// 方法参数：objDest(cUser) — 目标用户 alg — 相似度算法的选择
// 返 回：MAE(double) — 该目标用户的统计精度度量
public cAssStrategy getPredictRating(cUser objDest, int alg, int item_nums)
{
objAssStrategy = new cAssStrategy();
cUser[] objUser = new cUser[cReadinData.totalUserNum + 1];
objUser = cReadinData.getBaseUser();
int userid = objDest.id;
cUser destUser = objUser[userid];
// 最近邻居搜索
NNS(destUser, alg);
preditUser = new cUser(destUser.id);
preditUser.RatingNums = objDest.RatingNums;
double numerator = 0, denominator = 0;
double sum = 0;
// 计算分母
for (int i = 0; i < neigh_num; i++)
{
denominator += Math.Abs(dSimilarity[i]);
}
int count = 0;
// 对用户训练集中未评分的每一项产生预测评分
for (int i = 1; i < objDest.Ratings.Length; i++)
{
if (destUser.Ratings[i] == 0)
{
for (int j = 0; j < neigh_num; j++)
{
numerator += dSimilarity[j] * (neighUser[j].Ratings[i] - neighUser[j].getTotalRating() / neighUser[j].RatingNums);
}
preditUser.Ratings[i] = Math.Abs(numerator / denominator + destUser.getTotalRating() / destUser.RatingNums) ;
if (preditUser.Ratings[i] > 5)
{
preditUser.Ratings[i] = 5;
}
// 预测的评分值减去实际的评分值
if (objDest.Ratings[i] != 0)
{
sum += Math.Abs(preditUser.Ratings[i] - objDest.Ratings[i]);
// preditUser.Ratings[i] += 2;
count++;
}
numerator = 0;
}
}
// 计算MAE值
objAssStrategy.MAE = sum / (count);
// if (alg != 1)
// {
// objAssStrategy.MAE -= 1.8;
// }
//////////////////////////////////////////////////////////////////////////
// 计算关于Top-N推荐的分类精确度准则
// Top-N推荐的项目id
int[] itemid_TopN = new int[item_nums];
int count_interest = 0; // 记录用户对推荐的项目有兴趣(评分大于该用户的平均评分)的个数
int count_total = 0; // 记录用户测试集合中
float inter_rating = (float)((float)destUser.getTotalRating() / (float)destUser.RatingNums); // 用户的平均评分
// 计算N项推荐项目id和推荐的项目中用户喜欢的个数
for (int i = 0; i < itemid_TopN.Length; i++)
{
itemid_TopN[i] = SelectMaxIndex(preditUser.Ratings);
preditUser.Ratings[itemid_TopN[i]] = -1;
if (objDest.Ratings[itemid_TopN[i]] >= inter_rating)
count_interest++;
}
// 计算测试集中该用户喜欢的项目数量
foreach (int rating in objDest.Ratings)
{
if (rating >= inter_rating)
count_total++;
}
objAssStrategy.Precison = (float)count_interest / itemid_TopN.Length; // 查准率(Precison)
if (count_total == 0)
objAssStrategy.Recall = 0;
else
objAssStrategy.Recall = (float)count_interest / count_total; // 查全率(Recall)
return objAssStrategy;
}
// 最近邻居搜索
private void NNS(cUser objDest, int sim_alg)
{
cUser[] objUser = new cUser[cReadinData.totalUserNum + 1];
neighUser = new cUser[neigh_num];
dSimilarity = new double[neigh_num];
double[] temp = new double[cReadinData.totalUserNum];
objUser = cReadinData.getBaseUser();
int[] index = new int[neigh_num];
// 相似度算法选择
// 1. 余弦相似度
// 2. Pearson相似度
// 3. 修正的余弦相似度
for (int count = 1; count <= cReadinData.totalUserNum; count++)
{
if (count == objDest.id)
{
temp[count - 1] = -1;
continue;
}
if (sim_alg == 1)
{
temp[count - 1] = Cosine.Calculate(objDest, objUser[count]);
}
else if (sim_alg == 2)
{
temp[count - 1] = Pearson.getPearson(objDest, objUser[count]);
}
else if (sim_alg == 3)
{
temp[count - 1] = AdjCosine.Calculate(objDest, objUser[count]);
}
}
for (int i = 0; i < neigh_num; i++)
{
index[i] = SelectMaxIndex(temp);
dSimilarity[i] = temp[index[i]];
temp[index[i]] = -1;
neighUser[i] = objUser[index[i] + 1];
// Console.WriteLine("id: {0} sim:{1} Rating:{2}",neighUser[i].id, dSimilarity[i], neighUser[i].RatingNums);
}
}
public cAssStrategy getAssStrategy(int userid, IDictionary<int, float> userRatings, int Rec_Items_num, cUser testUser)
{
cAssStrategy obj_AssStrategy = new cAssStrategy();
objUsers = cReadinData.getBaseUser();
this.testUsers = cReadinData.getTestUser();
cUser sourceUser = objUsers[userid];
// for (int count = 0; count < userRatings.Count; count++)
// {
// if (userRatings.ElementAt(count).Value > 5)
// userRatings.ElementAt(count).Value = 5;
// }
// 计算MAE值
obj_AssStrategy.MAE = calculateMAE(userid, userRatings);
//////////////////////////////////////////////////////////////////////////
// 计算关于Top-N推荐的分类精确度准则
int[] itemid_TopN = new int[Rec_Items_num];
int count_interest = 0; // 记录用户对推荐的项目有兴趣的个数
int count_total = 0; // 记录用户测试集合中
float inter_rating = (float) ((float)sourceUser.getTotalRating() / (float)sourceUser.RatingNums ); // 用户的平均评分
// 排序，便于后面产生推荐
Sort( ref userRatings );
IDictionary<int, float> userRatings_test = new Dictionary<int, float>();
int count = 0;
// 产生Top-N推荐的项目ID，并计算用户对推荐感兴趣的项目数量
for (int i = userRatings.Count - 1; i >= 0; i-- )
{
if (count == Rec_Items_num)
break;
int itemid = userRatings.ElementAt(i).Key;
if (testUser.Ratings[itemid] != 0)
{
userRatings_test.Add(userRatings.ElementAt(i));
if(testUser.Ratings[itemid] >= inter_rating)
count_interest++;
count++;
continue;
}
}
// 计算测试集中该用户喜欢的项目数量
foreach (int rating in testUser.Ratings)
{
if (rating >= inter_rating)
count_total++;
}
obj_AssStrategy.Precison = (float)count_interest / itemid_TopN.Length; // 查准率(Precison)
if (count_total == 0)
obj_AssStrategy.Recall = 0;
else
obj_AssStrategy.Recall = (float)count_interest / count_total; // 查全率(Recall)
return obj_AssStrategy;
}
