(一)基于 Mahout 实现 User CF
1、相似度的计算
Similarity是计算两个用户或者两个物品之间的相似度的,归结到数学上就是计算向量的距离。Mahout 中提供了基本的相似度的计算,它们都实现了UserSimilarity 这个接口,实现用户相似度的计算,包括下面这些常用的:
- PearsonCorrelationSimilarity:基于皮尔逊相关系数计算相似度
- EuclideanDistanceSimilarity:基于欧几里德距离计算相似度
- TanimotoCoefficientSimilarity:基于 Tanimoto 系数计算相似度
- UncerteredCosineSimilarity:计算 Cosine 相似度
以PearsonCorrelationSimilarity为例,做一下讲解
首先看继承关系:
public final class PearsonCorrelationSimilarity extends AbstractSimilarity abstract class AbstractSimilarity extends AbstractItemSimilarity implements UserSimilarity public abstract class AbstractItemSimilarity implements ItemSimilarity //实际上AbstractSimilarity 这个抽象类实现了ItemSimilarity和UserSimilarity这两个接口 //那么作为他的子类,PearsonCorrelationSimilarity 既可以计算物品的相似度也可以计算用户的相似度
下边分析一下相似度具体是怎么实现的:
//用于计算两个用户之间的相似度 @Override public double userSimilarity(long userID1, long userID2) throws TasteException { DataModel dataModel = getDataModel(); PreferenceArray xPrefs = dataModel.getPreferencesFromUser(userID1);//user1的评分列表 PreferenceArray yPrefs = dataModel.getPreferencesFromUser(userID2);//user2的评分列表 int xLength = xPrefs.length(); int yLength = yPrefs.length(); if (xLength == 0 || yLength == 0) { return Double.NaN; } long xIndex = xPrefs.getItemID(0);//user1的当前item的ID long yIndex = yPrefs.getItemID(0);//user2的当前item的ID int xPrefIndex = 0;//user1的评分列表中的当前index int yPrefIndex = 0;//user2的评分列表中的当前index double sumX = 0.0; double sumX2 = 0.0; double sumY = 0.0; double sumY2 = 0.0; double sumXY = 0.0; double sumXYdiff2 = 0.0; int count = 0; boolean hasInferrer = inferrer != null; boolean hasPrefTransform = prefTransform != null; //遍历user1和user2的评分列表 while (true) { int compare = xIndex < yIndex ? -1 : xIndex > yIndex ? 1 : 0; if (hasInferrer || compare == 0) { double x; double y; if (xIndex == yIndex) { // Both users expressed a preference for the item if (hasPrefTransform) { x = prefTransform.getTransformedValue(xPrefs.get(xPrefIndex)); y = prefTransform.getTransformedValue(yPrefs.get(yPrefIndex)); } else { x = xPrefs.getValue(xPrefIndex); y = yPrefs.getValue(yPrefIndex); } } else { // Only one user expressed a preference, but infer the other one's preference and tally // as if the other user expressed that preference if (compare < 0) { // X has a value; infer Y's x = hasPrefTransform ? prefTransform.getTransformedValue(xPrefs.get(xPrefIndex)) : xPrefs.getValue(xPrefIndex); y = inferrer.inferPreference(userID2, xIndex); } else { // compare > 0 // Y has a value; infer X's x = inferrer.inferPreference(userID1, yIndex); y = hasPrefTransform ? prefTransform.getTransformedValue(yPrefs.get(yPrefIndex)) : yPrefs.getValue(yPrefIndex); } } //下边这一大堆乱七八糟的东西都是计算向量距离需要用的一些变量 sumXY += x * y; sumX += x; sumX2 += x * x; sumY += y; sumY2 += y * y; double diff = x - y; sumXYdiff2 += diff * diff; count++; } if (compare <= 0) { if (++xPrefIndex >= xLength) {//user1评分列表的index加一 if (hasInferrer) { // Must count other Ys; pretend next X is far away if (yIndex == Long.MAX_VALUE) { // ... but stop if both are done! break; } xIndex = Long.MAX_VALUE; } else { break; } } else { xIndex = xPrefs.getItemID(xPrefIndex);//获取user1评分列表中当前的item的ID } } if (compare >= 0) { if (++yPrefIndex >= yLength) {//user2评分列表的index加一 if (hasInferrer) { // Must count other Xs; pretend next Y is far away if (xIndex == Long.MAX_VALUE) { // ... but stop if both are done! break; } yIndex = Long.MAX_VALUE; } else { break; } } else { yIndex = yPrefs.getItemID(yPrefIndex);//获取user2评分列表中当前的item的ID } } } // "Center" the data. If my math is correct, this'll do it. double result; if (centerData) { double meanX = sumX / count; double meanY = sumY / count; // double centeredSumXY = sumXY - meanY * sumX - meanX * sumY + n * meanX * meanY; double centeredSumXY = sumXY - meanY * sumX; // double centeredSumX2 = sumX2 - 2.0 * meanX * sumX + n * meanX * meanX; double centeredSumX2 = sumX2 - meanX * sumX; // double centeredSumY2 = sumY2 - 2.0 * meanY * sumY + n * meanY * meanY; double centeredSumY2 = sumY2 - meanY * sumY; result = computeResult(count, centeredSumXY, centeredSumX2, centeredSumY2, sumXYdiff2); } else { //这个computeResult()函数是具体计算相似度的abstract函数 //那么AbstractSimilarity的子类去实现这个函数 result = computeResult(count, sumXY, sumX2, sumY2, sumXYdiff2); } if (similarityTransform != null) { result = similarityTransform.transformSimilarity(userID1, userID2, result); } if (!Double.isNaN(result)) { result = normalizeWeightResult(result, count, cachedNumItems); } return result; }
物品之间的相似度计算跟用户的基本上大同小异,就不写了。
2、邻居用户
根据建立的相似度计算方法,找到邻居用户。这里找邻居用户的方法,包括两种:“固定数量的邻居”和“相似度门槛邻居”计算方法,Mahout 提供对应的实现:
- NearestNUserNeighborhood:对每个用户取固定数量 N 的最近邻居
- ThresholdUserNeighborhood:对每个用户基于一定的限制,取落在相似度门限内的所有用户为邻居
下边以NearestNUserNeighborhood为例,看一下固定数量的最近邻居是怎么获取的:
NearestNUserNeighborhood类中的函数: @Override public long[] getUserNeighborhood(long userID) throws TasteException { DataModel dataModel = getDataModel();//得到数据源 UserSimilarity userSimilarityImpl = getUserSimilarity();//计算相似度的方法 TopItems.Estimator<Long> estimator = new Estimator(userSimilarityImpl, userID, minSimilarity); //SamplingLongPrimitiveIterator的作用是看是否要求取样,取样率<1.0的时候,对所有用户进行取样 LongPrimitiveIterator userIDs = SamplingLongPrimitiveIterator.maybeWrapIterator(dataModel.getUserIDs(), getSamplingRate()); return TopItems.getTopUsers(n, userIDs, null, estimator); } //这里也实现了一个评估器Estimator类: private static final class Estimator implements TopItems.Estimator<Long> { //主要功能函数,计算两用户的相似度,可以设置一个阀值(最小的,大于此值才要) @Override public double estimate(Long userID) throws TasteException { if (userID == theUserID) { return Double.NaN; } //计算userID用户与我们这个特定的user的相似度 double sim = userSimilarityImpl.userSimilarity(theUserID, userID); return sim >= minSim ? sim : Double.NaN; } } //最后获取邻居的最重要的实现部分: TopItems.getTopUsers(n, userIDs, null, estimator) public static long[] getTopUsers(int howMany, LongPrimitiveIterator allUserIDs, IDRescorer rescorer, Estimator<Long> estimator) throws TasteException { Queue<SimilarUser> topUsers = new PriorityQueue<SimilarUser>(howMany + 1, Collections.reverseOrder()); boolean full = false; double lowestTopValue = Double.NEGATIVE_INFINITY; while (allUserIDs.hasNext()) { long userID = allUserIDs.next(); if (rescorer != null && rescorer.isFiltered(userID)) { continue; } double similarity; try { //得到相似度 similarity = estimator.estimate(userID); } catch (NoSuchUserException nsue) { continue; } double rescoredSimilarity = rescorer == null ? similarity : rescorer.rescore(userID, similarity); if (!Double.isNaN(rescoredSimilarity) && (!full || rescoredSimilarity > lowestTopValue)) { //将该用户插入到优先队列中,就是个最小堆 topUsers.add(new SimilarUser(userID, rescoredSimilarity)); if (full) { topUsers.poll(); } else if (topUsers.size() > howMany) { full = true; topUsers.poll(); } lowestTopValue = topUsers.peek().getSimilarity(); } } int size = topUsers.size(); if (size == 0) { return NO_IDS; } List<SimilarUser> sorted = Lists.newArrayListWithCapacity(size); sorted.addAll(topUsers); Collections.sort(sorted); long[] result = new long[size]; int i = 0; for (SimilarUser similarUser : sorted) { result[i++] = similarUser.getUserID(); } return result; }