using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
namespace SlopeOne
{
public class Rating
{
public float Value { get; set; }
public int Freq { get; set; }
public float AverageValue
{
get { return Value / Freq; }
}
}
public class RatingDifferenceCollection : Dictionary<string, Rating>
{
private string GetKey(int Item1Id, int Item2Id)
{
return (Item1Id < Item2Id) ? Item1Id + "/" + Item2Id : Item2Id + "/" + Item1Id ;
}
public bool Contains(int Item1Id, int Item2Id)
{
return this.Keys.Contains<string>(GetKey(Item1Id, Item2Id));
}
public Rating this[int Item1Id, int Item2Id]
{
get {
return this[this.GetKey(Item1Id, Item2Id)];
}
set { this[this.GetKey(Item1Id, Item2Id)] = value; }
}
}
public class SlopeOne
{
public RatingDifferenceCollection _DiffMarix = new RatingDifferenceCollection(); // The dictionary to keep the diff matrix
public HashSet<int> _Items = new HashSet<int>(); // Tracking how many items totally
public void AddUserRatings(IDictionary<int, float> userRatings)
{
foreach (var item1 in userRatings)
{
int item1Id = item1.Key;
float item1Rating = item1.Value;
_Items.Add(item1.Key);
foreach (var item2 in userRatings)
{
if (item2.Key <= item1Id) continue; // Eliminate redundancy
int item2Id = item2.Key;
float item2Rating = item2.Value;
Rating ratingDiff;
if (_DiffMarix.Contains(item1Id, item2Id))
{
ratingDiff = _DiffMarix[item1Id, item2Id];
}
else
{
ratingDiff = new Rating();
_DiffMarix[item1Id, item2Id] = ratingDiff;
}
ratingDiff.Value += item1Rating - item2Rating;
ratingDiff.Freq += 1;
}
}
}
// Input ratings of all users
public void AddUerRatings(IList<IDictionary<int, float>> Ratings)
{
foreach(var userRatings in Ratings)
{
AddUserRatings(userRatings);
}
}
public IDictionary<int, float> Predict(IDictionary<int, float> userRatings)
{
Dictionary<int, float> Predictions = new Dictionary<int, float>();
foreach (var itemId in this._Items)
{
if (userRatings.Keys.Contains(itemId)) continue; // User has rated this item, just skip it
Rating itemRating = new Rating();
foreach (var userRating in userRatings)
{
if (userRating.Key == itemId) continue;
int inputItemId = userRating.Key;
if (_DiffMarix.Contains(itemId, inputItemId))
{
Rating diff = _DiffMarix[itemId, inputItemId];
itemRating.Value += diff.Freq * (userRating.Value + diff.AverageValue * ((itemId < inputItemId) ? 1 : -1));
itemRating.Freq += diff.Freq;
}
}
Predictions.Add(itemId, itemRating.AverageValue);
}
return Predictions;
}
public static void Test()
{
SlopeOne test = new SlopeOne();
Dictionary<int, float> userRating = new Dictionary<int, float>();
userRating.Add(1, 5);
userRating.Add(2, 4);
userRating.Add(3, 4);
test.AddUserRatings(userRating);
userRating = new Dictionary<int, float>();
userRating.Add(1, 4);
userRating.Add(2, 5);
userRating.Add(3, 3);
userRating.Add(4, 5);
test.AddUserRatings(userRating);
userRating = new Dictionary<int, float>();
userRating.Add(1, 4);
userRating.Add(2, 4);
userRating.Add(4, 5);
test.AddUserRatings(userRating);
userRating = new Dictionary<int, float>();
userRating.Add(1, 5);
userRating.Add(3, 4);
IDictionary<int, float> Predictions = test.Predict(userRating);
foreach (var rating in Predictions)
{
Console.WriteLine("Item " + rating.Key + " Rating: " + rating.Value);
}
}
}
}