static List <UserRecom> Predict(List <UserBrandTable> ubt, List <int> bList, double[,] sim, int k_ub, int k_b, double h, List <BrandInfo> bi)
{
List <UserRecom> recommend = new List <UserRecom>();
int bsize = bList.Count;
int usize = ubt.Count;
List <BrandAndRating> bSim = new List <BrandAndRating>(); // 将相似度也看成一种给分
List <BrandAndRating> pbSim = new List <BrandAndRating>(); // 商品预测给分
List <int> havePrd = new List <int>();
// 对所有用户
for (int i = 0; i < usize; i++)
{
// 选取i用户的k_ub个给分高的商品
for (int k = 0; k < (k_ub < ubt[i].brands.Count ? k_ub : ubt[i].brands.Count); k++)
{
for (int j = 0; j < bsize; j++)
{
// 用户感兴趣的物品的相似物品
bSim.Add(new BrandAndRating(bList[j], sim[bList.IndexOf(ubt[i].brands[k].brand_id), j]));
}
bSim.Sort(); // 对相似度排序
int idx = bi.FindIndex(delegate(BrandInfo bi_) { return(bi_.brand_id == ubt[i].brands[k].brand_id); });
pbSim.Add(new BrandAndRating(ubt[i].brands[k].brand_id, ubt[i].brands[k].rating * (bi[idx].brand_click / (double)bi[idx].brand_count))); // 本商品
havePrd.Add(ubt[i].brands[k].brand_id);
// 选取k_b个相似商品
for (int l = 0; l < k_b; l++)
{
idx = bi.FindIndex(delegate(BrandInfo bi_) { return(bi_.brand_id == bSim[l].brand_id); });
// 本商品的不考虑，已算
if (!havePrd.Contains(bSim[l].brand_id))
{
pbSim.Add(new BrandAndRating(bSim[l].brand_id, ubt[i].brands[k].rating * bSim[l].rating * (bi[idx].brand_click / (double)bi[idx].brand_count)));// *2修正
havePrd.Add(bSim[l].brand_id);
}
}
bSim.Clear();
}
pbSim.Sort(); // 对预测给分排序
recommend.Add(new UserRecom(ubt[i].user_id));
//Console.WriteLine("用户:"+ubt[i].user_id);
for (int m = 0; m < pbSim.Count; m++)
{
recommend[recommend.Count - 1].Add(pbSim[m]);
//Console.Write(pbSim[m].brand_id+":"+pbSim[m].rating+",");
}
//Console.WriteLine();
havePrd.Clear();
pbSim.Clear();
}
DataRWHelper dh = new DataRWHelper();
// save
StreamWriter sw = new StreamWriter(dh.pathPrefix + "p.txt");
string line;
for (int i = 0; i < recommend.Count; i++)
{
line = "" + recommend[i].user_id + " ";
//sw.Write(recommend[i].user_id);
//sw.Write('\t');
for (int j = 0; j < recommend[i].brands.Count; j++)
{
if (recommend[i].brands[j].rating >= h)
{
//sw.Write(recommend[i].brands[j].brand_id);
line += recommend[i].brands[j].brand_id;
line += ":";
line += recommend[i].brands[j].rating;
line += ",";
}
}
// 没有推荐品牌的用户不输出
if (!line.Equals("" + recommend[i].user_id + " "))
{
//line.Remove(line.LastIndexOf(','));
sw.WriteLine(line.Substring(0, line.Length - 1));
}
}
sw.Close();
return(recommend);
}
static void Main(string[] args)
{
DataRWHelper dh = new DataRWHelper();
List <Dt> data = dh.ReadData(dh.TRAIN_DATA); // 加载测试数据
List <int> bList = new List <int>(); // 保存品牌表
List <UserBrandTable> ubt = new List <UserBrandTable>();
List <Dt> oneUserData = new List <Dt>(); // 保存一个用户的记录
List <BrandInfo> bi = new List <BrandInfo>();
for (int i = 0; i <= data.Count; i++)
{
if (i == data.Count) // 处理最后一个数据
{
ubt.Add(new UserBrandTable(oneUserData)); // 添加到用户商品表
oneUserData.Clear();
break;
}
int idx = bi.FindIndex(delegate(BrandInfo bi_) { return(bi_.brand_id == data[i].brand_id); });
if (idx < 0)
{
bi.Add(new BrandInfo(data[i].brand_id, 0, 0));
}
idx = bi.FindIndex(delegate(BrandInfo bi_) { return(bi_.brand_id == data[i].brand_id); });
bi[idx].brand_count++;
if (data[i].operation == 1)
{
bi[idx].brand_click++;
}
// 如果该表中存在用户而且加入的用户不同，则清理列表，加入新的数据
if (oneUserData.Count != 0 && oneUserData[0].user_id != data[i].user_id)
{
// 已经得到一个用户的全部操作信息
ubt.Add(new UserBrandTable(oneUserData)); // 添加到用户商品表
oneUserData.Clear();
}
oneUserData.Add(data[i]);
// 保存品牌列表
if (!bList.Contains(data[i].brand_id))
{
bList.Add(data[i].brand_id);
}
}
for (int i = 0; i < ubt.Count; i++)
{
ubt[i].brands.Sort(); // 对品牌评分排序
}
Console.WriteLine("用户的品牌评分完成！");
double[,] sim = new double[bList.Count, bList.Count]; // 物品相似矩阵
CalcSimilarity(sim, bList, ubt); // 计算相似矩阵
Console.WriteLine("计算相似度完成！");
//for (int i = 0; i < 10; i++) {
// for (int j = 0; j < 10; j++) {
// Console.Write(sim[i,j]+",");
// }
// Console.WriteLine();
//}
Console.WriteLine("开始预测...");
List <UserRecom> rcm = Predict(ubt, bList, sim, 3, 3, 4.0, bi);
Console.WriteLine("预测完毕");
// 清楚内存
sim = null;
data = null;
ubt = null;
GC.Collect();
// 检验...
Console.WriteLine("正在检验算法...");
List <Dt> testData = dh.ReadData(dh.TEST_DATA); // 读取测试数据
int hb = 0;
int pb = 0;
int bb = 0;
for (int i = 0; i < rcm.Count; i++)
{
for (int j = 0; j < rcm[i].brands.Count; j++)
{
// 跳过不符合条件的，没有考虑的
if (rcm[i].brands[0].rating < 4.0)
{
continue;
}
pb++; //
if (testData.FindIndex(delegate(Dt d) {
// 找到用户购买了该商品
return(d.user_id == rcm[i].user_id && d.brand_id == rcm[i].brands[j].brand_id && d.operation == 1);
}) >= 0)
{
hb++;
}
}
}
double p = hb / (double)pb;
Console.WriteLine("准确率:{0}", p);
hb = 0;
for (int i = 0; i < testData.Count; i++)
{
if (testData[i].operation == 1)
{
bb++;
}
}
double r = hb / (double)bb;
// so precision:
Console.WriteLine("召回率：{0}", r);
Console.WriteLine("F-score:{0}", 2 * p * r / (p + r));
}