private static Tuple <int, double> SelectNextBook(List <int> currentlySelectedBookIDs,
List <Tuple <int, double> > currentlyAvailableBookIDAndWeightsForSelection,
double lambda, BooksSimilarityModel model)
{
if (currentlyAvailableBookIDAndWeightsForSelection.Count == 0)
{
return(null);
}
double theBestSimilarity = Double.MinValue;
Tuple <int, double> theBestBookIDAndWeight = null;
foreach (Tuple <int, double> bookIDAndWeightI in currentlyAvailableBookIDAndWeightsForSelection)
{
int bookIDI = bookIDAndWeightI.Item1;
double relevancyI = bookIDAndWeightI.Item2;
double similarityMaxI = model.GetSimilarityMax(
currentlySelectedBookIDs, bookIDI);
double similarityI =
lambda * relevancyI - (1 - lambda) * similarityMaxI;
if (similarityI > theBestSimilarity)
{
theBestSimilarity = similarityI;
theBestBookIDAndWeight = bookIDAndWeightI;
}
}
return(theBestBookIDAndWeight);
}
public static List <int> Recommend(List <int> recCoreOfBookIDs,
List <Tuple <int, double> > bookIDsAvailableForAdding,
BooksSimilarityModel model, double lambda, int howMany)
{
// iterative addition available books to rec. core
List <int> currentlySelectedBookIDs = recCoreOfBookIDs;
while (currentlySelectedBookIDs.Count() < howMany)
{
Tuple <int, double> nextBookIDAndWeight = SelectNextBook(
currentlySelectedBookIDs, bookIDsAvailableForAdding,
lambda, model);
if (nextBookIDAndWeight == null)
{
return(currentlySelectedBookIDs);
}
currentlySelectedBookIDs.Add(nextBookIDAndWeight.Item1);
bookIDsAvailableForAdding.Remove(nextBookIDAndWeight);
}
return(currentlySelectedBookIDs.Take(howMany).ToList());
}
/// <summary>
/// Recommender algorithm Maximal Marginal Relevance (MMR) searchs books similar to one other,
/// at the same time strives for diversity of returned books.
/// Algorithm is based od ContentBasedBookSimilarity recommender. In the first step, the algorithm
/// requires a list of all similar books to one other. Then it select the first 10 (CORE_SIZE)
/// books as a core of resulting recommendation list. The other 90 (CANDIDATES_SIZE) books from
/// the list are taken as candidates. The algorithm then iteratively adds one candidate to the
/// result list. Another element of the resulting list is selected to maximize diversity
/// (minimizing the similarity of books in the resulting list).
/// The similarity of books is measured by Jaccard's similarity to the authors and the genres.
/// /// </summary>
/// <param name="bookId">Id of book on which will the recommendation be based</param>
/// <param name="userId">Id of the signed in user</param>
/// <param name="howMany">How many books to return</param>
/// <returns>Maximal Marginal Relevance List of books (bookIDs) based on results of
/// RecommenderBookSimilar algorithm
/// /// </returns>
public static List <int> Recommend(int bookId, string userId = null,
double lambda = 0.2, int howMany = 6)
{
List <Tuple <int, int> > bookIDsAndTheirQuantitiesAll =
RecommenderContentBasedBookSimilarity.RecommendWeightedList(bookId, userId);
// creating core and candidates of books
List <Tuple <int, int> > recCoreWList =
bookIDsAndTheirQuantitiesAll.Take(CORE_SIZE).ToList();
List <Tuple <int, int> > recCandidatesWList =
bookIDsAndTheirQuantitiesAll.Skip(CORE_SIZE).Take(CANDIDATES_SIZE).ToList();
// creating similarity model of books
var db = new BookRecommenderContext();
db.ChangeTracker.QueryTrackingBehavior = Microsoft.EntityFrameworkCore.QueryTrackingBehavior.NoTracking;
List <int> boodIdsForModel = new List <int>();
boodIdsForModel.AddRange(recCoreWList.Select(b => b.Item1).ToList());
boodIdsForModel.AddRange(recCandidatesWList.Select(b => b.Item1).ToList());
BooksSimilarityModel model = new BooksSimilarityModel(boodIdsForModel);
model.CountSimilarity(db);
// run diversity enhanced recommender
List <int> recCoreList = recCoreWList.Select(b => b.Item1).ToList();
List <Tuple <int, double> > recCandidatesNormWList =
NormBookIDsAndTheirQuantities(recCandidatesWList);
return(RecommenderDiversityEnhanced.Recommend(recCoreList,
recCandidatesNormWList, model, lambda, howMany).ToList());
}
