Java CosineVectorSimilarity类代码示例

import org.grouplens.lenskit.vectors.similarity.CosineVectorSimilarity; //导入依赖的package包/类
/**
 * Get the cosine similarity between user-user
 * @param user1  user ID 1
 * @param user2  user ID 2
 * @return  the cosine similarity
 */
private double getUserUserSimilarity(long user1, long user2){
    CosineVectorSimilarity csim = new CosineVectorSimilarity();
    SparseVector uv1 = getUserRatingVector(user1);
    SparseVector uv2 = getUserRatingVector(user2);

    MutableSparseVector mv1  = uv1.mutableCopy();
    MutableSparseVector mv2  = uv2.mutableCopy();
    mv1.add(-uv1.mean());
    mv2.add(-uv2.mean());
    return csim.similarity(mv1,mv2);
}
 

import org.grouplens.lenskit.vectors.similarity.CosineVectorSimilarity; //导入依赖的package包/类
@Override
public SimpleItemItemModel get() {
	// Get the transposed rating matrix
	// This gives us a map of item IDs to those items' rating vectors
	Map<Long, ImmutableSparseVector> itemVectors = getItemVectors();

	// Get all items - you might find this useful
	LongSortedSet items = LongUtils.packedSet(itemVectors.keySet());
	// Map items to vectors of item similarities
	@SuppressWarnings("unused")
	Map<Long,MutableSparseVector> itemSimilarities = new HashMap<Long, MutableSparseVector>();

	// Compute the similarities between each pair of items
	// It will need to be in a map of longs to lists of Scored IDs to store in the model
	Map<Long, List<ScoredId>> neighborhoods = new HashMap<Long, List<ScoredId>>();


	// Compute the similarities between each pair of items
	CosineVectorSimilarity cosine = new CosineVectorSimilarity();

	for(long item : items){

		// get this item ratings
		ImmutableSparseVector itemRatings = itemVectors.get(item);

		// create the accumulator for this item
		TopNScoredItemAccumulator accumulator = new TopNScoredItemAccumulator(items.size() - 1);

		for(long neighbor : items){

			// skip itself
			if(item == neighbor) continue;

			ImmutableSparseVector neighRatings = itemVectors.get(neighbor);

			// cosine similarity
			double similarity = cosine.similarity(itemRatings, neighRatings);

			//accumulate positive similarities
			if(similarity >= 0.0){
				accumulator.put(neighbor, similarity);
			}
		}

		//get the final list of sorted neighbors
		List<ScoredId> similarities = accumulator.finish();

		// update the map of similarity
		neighborhoods.put(item, similarities);
	}

	// It will need to be in a map of longs to lists of Scored IDs to store in the model
	return new SimpleItemItemModel(neighborhoods);
}
 

import org.grouplens.lenskit.vectors.similarity.CosineVectorSimilarity; //导入依赖的package包/类
@Override
  public SimpleItemItemModel get() {
      // Get the transposed rating matrix
      // This gives us a map of item IDs to those items' rating vectors
      Map<Long, ImmutableSparseVector> itemVectors = getItemVectors();

      // Get all items - you might find this useful
      LongSortedSet items = LongUtils.packedSet(itemVectors.keySet());
      // Map items to vectors of item similarities
      //Map<Long,MutableSparseVector> itemSimilarities = new HashMap<Long, MutableSparseVector>();
Map<Long, List<ScoredId>> neighborhoods = new HashMap<Long, List<ScoredId>>();

// Computing the similarities between each pair of items
      // It will need to be in a map of longs to lists of Scored IDs to store in the model
for(Iterator outerIter = items.iterator(); outerIter.hasNext() ; ) {
	Long thisItemId = (Long) outerIter.next();

	TopNScoredItemAccumulator accumulator = new TopNScoredItemAccumulator(items.size()-1);

	// Calculate similiarity with other item one by one and
	for(Iterator innerIter = items.iterator(); innerIter.hasNext() ; ) {
		Long nghbrItemId = (Long) innerIter.next();

		if(thisItemId.equals(nghbrItemId)) continue;

		// cosine similarity
		double similarity = new CosineVectorSimilarity().similarity(itemVectors.get(thisItemId),
				itemVectors.get(nghbrItemId));

		//accumulate
		if (similarity > 0) {
			accumulator.put(nghbrItemId, similarity);
		}
	}
	//put in the final list of sorted neighbors
	List<ScoredId> similarities = accumulator.finish();
	neighborhoods.put(thisItemId, similarities);
}
return new SimpleItemItemModel(neighborhoods);
      //return new SimpleItemItemModel(Collections.EMPTY_MAP);
  }
 

import org.grouplens.lenskit.vectors.similarity.CosineVectorSimilarity; //导入依赖的package包/类
/**
    * Calculate User similarity
    * @param userVector, otherUserVector
    * @return The similarity vector 
    */
private double calculateUserSimilarity(SparseVector userVector, SparseVector otherUserVector) {
	
	MutableSparseVector userCopy = userVector.mutableCopy();
	MutableSparseVector otherUserCopy = otherUserVector.mutableCopy();
	
	userCopy.subtract(MutableSparseVector.create(userCopy.keyDomain(), userVector.mean()));
	otherUserCopy.subtract(MutableSparseVector.create(otherUserCopy.keyDomain(), otherUserCopy.mean()));
	
	CosineVectorSimilarity similarity = new CosineVectorSimilarity();
	
	return similarity.similarity(userCopy, otherUserCopy);
}