var critics = new Dictionary<string, Dictionary<string, float>>
{
["Lisa Rose"] = new Dictionary<string, float>
{
["Lady in the Water"] = 2.5f,
["Snakes on a Plane"] = 3.5f,
["Just My Luck"] = 3.0f,
["Superman Returns"] = 3.5f,
["You, Me and Dupree"] = 2.5f,
["The Night Listener"] = 3.0f
},
["Gene Seymour"] = new Dictionary<string, float>
{
["Lady in the Water"] = 3.0f,
["Snakes on a Plane"] = 3.5f,
["Just My Luck"] = 1.5f,
["Superman Returns"] = 5.0f,
["The Night Listener"] = 3.0f,
["You, Me and Dupree"] = 3.5f
},
["Michael Phillips"] = new Dictionary<string, float>
{
["Lady in the Water"] = 2.5f,
["Snakes on a Plane"] = 3.0f,
["Superman Returns"] = 3.5f,
["The Night Listener"] = 4.0f
},
["Claudia Puig"] = new Dictionary<string, float>
{
["Snakes on a Plane"] = 3.5f,
["Just My Luck"] = 3.0f,
["The Night Listener"] = 4.5f,
["Superman Returns"] = 4.0f,
["You, Me and Dupree"] = 2.5f
},
["Mick LaSalle"] = new Dictionary<string, float>
{
["Lady in the Water"] = 3.0f,
["Snakes on a Plane"] = 4.0f,
["Just My Luck"] = 2.0f,
["Superman Returns"] = 3.0f,
["The Night Listener"] = 3.0f,
["You, Me and Dupree"] = 2.0f
},
["Jack Matthews"] = new Dictionary<string, float>
{
["Lady in the Water"] = 3.0f,
["Snakes on a Plane"] = 4.0f,
["The Night Listener"] = 3.0f,
["Superman Returns"] = 5.0f,
["You, Me and Dupree"] = 3.5f
},
["Toby"] = new Dictionary<string, float>
{
["Snakes on a Plane"] = 4.5f,
["You, Me and Dupree"] = 1.0f,
["Superman Returns"] = 4.0f
}
};
interface ISimilar
{
double Calc(Dictionary<string, Dictionary<string, float>> prefs, string person1, string person2);
}
public class SimilarDistance : ISimilar
{
public double Calc(Dictionary<string, Dictionary<string, float>> prefs, string person1, string person2)
{
//同前...略
}
}
public class SimilarPerson : ISimilar
{
public double Calc(Dictionary<string, Dictionary<string, float>> prefs, string person1, string person2)
{
//同前...略
}
}
使用也很簡單：
ISimilar simiCaculator = new SimilarPerson();
var similar = simiCaculator.Calc(critics, "Lisa Rose", "Gene Seymour");
Console.WriteLine(similar);
查找偏好最相近的評論者
有了上面的評分函數，尋找偏好相近的評論者就是很簡單的事了。直接上函數：
public readonly Dictionary<string, Dictionary<string, float>> Critics = new Dictionary<string, Dictionary<string, float>>
{
//初始偏好資料 - 略
};
//n:為傳回最相似比對者數目
//similar:相似度函數
public List<KeyValuePair<double,string>> TopMatches(Dictionary<string, Dictionary<string, float>> prefs,
string person, int n = 5, ISimilar similar = null)
{
if(similar == null)
similar = new SimilarPerson();
var dic = prefs.Where(p => p.Key != person)
.Select(p => p.Key)
.ToDictionary(other => similar.Calc(prefs, person, other), other => other);
var sortDic = new SortedDictionary<double, string>(dic, new SimilarComparer());
return sortDic.Take(n).ToList();
}
class SimilarComparer : IComparer<Double>
{
public int Compare(double x, double y)
{
return y.CompareTo(x);
}
}
使用下面的代碼可以進行測試：
Tester c = new Tester();
var result = c.TopMatches(c.Critics, "Toby", 3);
Console.WriteLine(string.Join(Environment.NewLine, result.Select(kvp=>$"{kvp.Key}, {kvp.Value}")));
//利用所有他人評價值權重，為某人提供建議
public List<KeyValuePair<double, string>> GetRecommendations(Dictionary<string, Dictionary<string, float>> prefs,
string person, ISimilar similar = null)
{
if (similar == null)
similar = new SimilarPerson();
var totals =new Dictionary<string,double>();
var simSums = new Dictionary<string,double>();
foreach (var other in prefs)
{
//不和自己比較
if(other.Key == person)
continue;
var sim = similar.Calc(prefs, person, other.Key);
//忽略相似度小于等于0的情況
if (sim <= 0) continue;
foreach (var kvp in prefs[other.Key])
{
//隻對自己未看過的電影評價
if (!prefs[person].ContainsKey(kvp.Key))
{
//相似度 x 評價值
if(!totals.ContainsKey(kvp.Key)) totals.Add(kvp.Key,0);
totals[kvp.Key] += prefs[other.Key][kvp.Key]*sim;
//相似度之和
if(!simSums.ContainsKey(kvp.Key)) simSums.Add(kvp.Key,0);
simSums[kvp.Key] += sim;
}
}
}
var avgVal = totals.ToDictionary(t => t.Value/simSums[t.Key], t => t.Key);
var rankings = new SortedDictionary<double, string>(avgVal,new SimilarComparer());
return rankings.ToList();
}
測試代碼：
Tester c = new Tester();
var result = c.GetRecommendations(c.Critics, "Toby");
Console.WriteLine(string.Join(Environment.NewLine, result.Select(kvp=>$"{kvp.Key}, {kvp.Value}")));
public Dictionary<string, Dictionary<string, float>> TransformPrefs(
Dictionary<string, Dictionary<string, float>> prefs)
{
var result = new Dictionary<string, Dictionary<string, float>>();
foreach (var personKvp in prefs)
{
foreach (var itemKvp in personKvp.Value)
{
if(!result.ContainsKey(itemKvp.Key))
result.Add(itemKvp.Key,new Dictionary<string, float>());
//将人員和物品對調
result[itemKvp.Key].Add(personKvp.Key, prefs[personKvp.Key][itemKvp.Key]);
}
}
return result;
}
下面的測試代碼，可以得到與《Superman Returns》最相近的影片：
Tester c = new Tester();
var movies = c.TransformPrefs(c.Critics);
var result = c.TopMatches(movies, "Superman Returns");
Console.WriteLine(string.Join(Environment.NewLine, result.Select(kvp=>$"{kvp.Key}, {kvp.Value}")));
注意：結果中的負值表示，喜歡Superman Returns的人，有存在不喜歡Just My Luck的傾向。
Tester c = new Tester();
var movies = c.TransformPrefs(c.Critics);
var result = c.GetRecommendations(movies, "Just My Luck");
Console.WriteLine(string.Join(Environment.NewLine, result.Select(kvp=>$"{kvp.Key}, {kvp.Value}")));
public Dictionary<string, List<KeyValuePair<double,string>>> CalculateSimilarItems(
Dictionary<string, Dictionary<string, float>> prefs, int n)
{
//字典key為物品，value為與這個物品最為相近的前n個物品
var result = new Dictionary<string, List<KeyValuePair<double,string>>>();
var itemPrefs = TransformPrefs(prefs);
var c = 0;
foreach (var itemPref in itemPrefs)
{
//顯示運作進度
++c;
if(c%100==0) Console.WriteLine($"{c} / {itemPref.Value.Count}");
//尋找最為相近的物品
var sources = TopMatches(itemPrefs, itemPref.Key, n, new SimilarDistance());
result.Add(itemPref.Key,sources);
}
return result;
}
測試一下這個函數：
Tester c = new Tester();
var result = c.CalculateSimilarItems(c.Critics, 10);
Console.WriteLine(JsonConvert.SerializeObject(result));
public List<KeyValuePair<double, string>> GetRecommendedItems(
Dictionary<string, Dictionary<string, float>> prefs,
Dictionary<string, List<KeyValuePair<double, string>>> itemMatch,
string user)
{
var userRatings = prefs[user];
var scores= new Dictionary<string,double>();
var totalSim = new Dictionary<string,double>();
//周遊目前使用者評分的産品
foreach (var ratingKvp in userRatings)
{
//循環目前物品相近的物品
foreach (var simiKvp in itemMatch[ratingKvp.Key])
{
//如果該使用者已經對目前物品評價過，則忽略
if(userRatings.ContainsKey(simiKvp.Value))
continue;
//評價值與相似度的權重值和
if(!scores.ContainsKey(simiKvp.Value))
scores.Add(simiKvp.Value,0);
scores[simiKvp.Value] += simiKvp.Key*ratingKvp.Value;
//全部相似度之和
if(!totalSim.ContainsKey(simiKvp.Value))
totalSim.Add(simiKvp.Value, 0);
totalSim[simiKvp.Value] += simiKvp.Key;
}
}
//将每個合計值除以權重和，求出平均值
var avgVal = scores.ToDictionary(t => t.Value / totalSim[t.Key], t => t.Key);
//按最高值排序，傳回評分結果
var rankings = new SortedDictionary<double, string > (avgVal, new SimilarComparer());
return rankings.ToList();
}
下面的代碼可以測試上面的函數：
Tester c = new Tester();
//再生産場景中，下面的值應該被計算并存儲
var simiItems = c.CalculateSimilarItems(c.Critics, 10);
var recommended = c.GetRecommendedItems(c.Critics, simiItems, "Toby");
Console.WriteLine(JsonConvert.SerializeObject(recommended));
public List<KeyValuePair<double,string>> TopMatches(Dictionary<string, Dictionary<string, float>> prefs,
string person, int n = 5, ISimilar similar = null)
{
if(similar == null)
similar = new SimilarPerson();
var dicKey = prefs.Where(p => p.Key != person).Select(p => p.Key).ToList();
var dic = new Dictionary<double, string>(dicKey.Count);
foreach (var pi in dicKey)
{
var score = similar.Calc(prefs, person, pi);
if(!dic.ContainsKey(score))
dic.Add(score, pi);
}
var sortDic = new SortedDictionary<double, string>(dic, new SimilarComparer());
return sortDic.Take(n).ToList();
}
