当前位置：首页 > news >正文

基于物品的协同过滤推荐算法实现（Java电商平台）

news 来源：原创 2025/6/14 7:39:23

下面我将为你实现一个基于物品的协同过滤推荐算法，适用于Java电商平台。这个实现包括核心算法、相似度计算和推荐生成。

1. 数据模型

首先定义我们需要的数据模型：

public class Item {private String itemId;    // 商品IDprivate String name;      // 商品名称// 其他商品属性...// 构造方法、getter和setter
}public class User {private String userId;    // 用户IDprivate String username; // 用户名// 其他用户属性...// 构造方法、getter和setter
}public class UserBehavior {private String userId;    // 用户IDprivate String itemId;    // 商品IDprivate double preference; // 偏好分数（如评分、购买次数等）private long timestamp;   // 行为时间戳// 构造方法、getter和setter
}

2. 相似度计算

基于物品的协同过滤核心是计算物品之间的相似度：

public class ItemSimilarity {/*** 计算物品之间的余弦相似度* @param itemPrefMap 物品-用户偏好矩阵: Map<itemId, Map<userId, preference>>* @return 物品相似度矩阵: Map<itemId, Map<itemId, similarity>>*/public Map<String, Map<String, Double>> calculateCosineSimilarity(Map<String, Map<String, Double>> itemPrefMap) {Map<String, Map<String, Double>> similarityMatrix = new HashMap<>();// 获取所有物品列表List<String> itemIds = new ArrayList<>(itemPrefMap.keySet());// 计算每对物品之间的相似度for (int i = 0; i < itemIds.size(); i++) {String itemId1 = itemIds.get(i);Map<String, Double> userPrefs1 = itemPrefMap.get(itemId1);// 初始化相似度矩阵similarityMatrix.put(itemId1, new HashMap<>());for (int j = i; j < itemIds.size(); j++) {String itemId2 = itemIds.get(j);if (itemId1.equals(itemId2)) {// 相同物品相似度为1similarityMatrix.get(itemId1).put(itemId2, 1.0);continue;}Map<String, Double> userPrefs2 = itemPrefMap.get(itemId2);// 计算两个物品的共同用户Set<String> commonUsers = new HashSet<>(userPrefs1.keySet());commonUsers.retainAll(userPrefs2.keySet());if (commonUsers.isEmpty()) {// 没有共同用户，相似度为0similarityMatrix.get(itemId1).put(itemId2, 0.0);similarityMatrix.get(itemId2).put(itemId1, 0.0);continue;}// 计算余弦相似度的分子和分母double dotProduct = 0.0;double norm1 = 0.0;double norm2 = 0.0;for (String userId : commonUsers) {double pref1 = userPrefs1.get(userId);double pref2 = userPrefs2.get(userId);dotProduct += pref1 * pref2;norm1 += Math.pow(pref1, 2);norm2 += Math.pow(pref2, 2);}// 计算余弦相似度double similarity = dotProduct / (Math.sqrt(norm1) * Math.sqrt(norm2));// 对称矩阵，所以两个方向都存储similarityMatrix.get(itemId1).put(itemId2, similarity);// 确保对称矩阵的另一半也被填充if (!similarityMatrix.containsKey(itemId2)) {similarityMatrix.put(itemId2, new HashMap<>());}similarityMatrix.get(itemId2).put(itemId1, similarity);}}return similarityMatrix;}/*** 计算物品之间的改进余弦相似度（考虑用户平均评分）*/public Map<String, Map<String, Double>> calculateAdjustedCosineSimilarity(Map<String, Map<String, Double>> itemPrefMap,Map<String, Double> userAvgPrefMap) {// 实现类似上面，但在计算时减去用户平均评分// ...return similarityMatrix;}
}

3. 推荐引擎

基于物品相似度生成推荐：

public class ItemBasedRecommender {private Map<String, Map<String, Double>> itemSimilarityMatrix;private Map<String, Map<String, Double>> userItemPrefMap; // 用户-物品偏好矩阵public ItemBasedRecommender(Map<String, Map<String, Double>> itemSimilarityMatrix,Map<String, Map<String, Double>> userItemPrefMap) {this.itemSimilarityMatrix = itemSimilarityMatrix;this.userItemPrefMap = userItemPrefMap;}/*** 为用户推荐物品* @param userId 用户ID* @param topN 推荐数量* @return 推荐物品ID列表，按推荐分数降序排列*/public List<String> recommendItems(String userId, int topN) {// 获取用户历史行为物品Map<String, Double> userHistory = userItemPrefMap.getOrDefault(userId, new HashMap<>());// 存储物品的推荐分数Map<String, Double> recommendationScores = new HashMap<>();// 遍历用户历史行为物品for (Map.Entry<String, Double> entry : userHistory.entrySet()) {String historyItemId = entry.getKey();double historyPref = entry.getValue();// 获取与历史物品相似的物品Map<String, Double> similarItems = itemSimilarityMatrix.getOrDefault(historyItemId, new HashMap<>());// 计算推荐分数for (Map.Entry<String, Double> simEntry : similarItems.entrySet()) {String candidateItemId = simEntry.getKey();double similarity = simEntry.getValue();// 排除用户已经有过行为的物品if (!userHistory.containsKey(candidateItemId)) {double score = recommendationScores.getOrDefault(candidateItemId, 0.0);score += similarity * historyPref;recommendationScores.put(candidateItemId, score);}}}// 按推荐分数排序并返回topNreturn recommendationScores.entrySet().stream().sorted(Map.Entry.<String, Double>comparingByValue().reversed()).limit(topN).map(Map.Entry::getKey).collect(Collectors.toList());}
}

4. 数据预处理

在实际应用中，我们需要将原始用户行为数据转换为算法需要的格式：

public class DataPreprocessor {/*** 将用户行为列表转换为物品-用户偏好矩阵*/public Map<String, Map<String, Double>> convertToItemUserMatrix(List<UserBehavior> behaviors) {Map<String, Map<String, Double>> itemUserMatrix = new HashMap<>();for (UserBehavior behavior : behaviors) {String itemId = behavior.getItemId();String userId = behavior.getUserId();double preference = behavior.getPreference();itemUserMatrix.putIfAbsent(itemId, new HashMap<>());itemUserMatrix.get(itemId).put(userId, preference);}return itemUserMatrix;}/*** 将用户行为列表转换为用户-物品偏好矩阵*/public Map<String, Map<String, Double>> convertToUserItemMatrix(List<UserBehavior> behaviors) {Map<String, Map<String, Double>> userItemMatrix = new HashMap<>();for (UserBehavior behavior : behaviors) {String userId = behavior.getUserId();String itemId = behavior.getItemId();double preference = behavior.getPreference();userItemMatrix.putIfAbsent(userId, new HashMap<>());userItemMatrix.get(userId).put(itemId, preference);}return userItemMatrix;}/*** 计算每个用户的平均评分*/public Map<String, Double> calculateUserAveragePreference(List<UserBehavior> behaviors) {Map<String, Double> sumMap = new HashMap<>();Map<String, Integer> countMap = new HashMap<>();for (UserBehavior behavior : behaviors) {String userId = behavior.getUserId();double preference = behavior.getPreference();sumMap.put(userId, sumMap.getOrDefault(userId, 0.0) + preference);countMap.put(userId, countMap.getOrDefault(userId, 0) + 1);}Map<String, Double> avgMap = new HashMap<>();for (String userId : sumMap.keySet()) {avgMap.put(userId, sumMap.get(userId) / countMap.get(userId));}return avgMap;}
}

5. 完整使用示例

public class RecommendationDemo {public static void main(String[] args) {// 1. 模拟数据List<UserBehavior> behaviors = new ArrayList<>();behaviors.add(new UserBehavior("u1", "i1", 5.0, System.currentTimeMillis()));behaviors.add(new UserBehavior("u1", "i2", 3.0, System.currentTimeMillis()));behaviors.add(new UserBehavior("u1", "i3", 4.0, System.currentTimeMillis()));behaviors.add(new UserBehavior("u2", "i1", 4.0, System.currentTimeMillis()));behaviors.add(new UserBehavior("u2", "i3", 3.0, System.currentTimeMillis()));behaviors.add(new UserBehavior("u2", "i4", 4.5, System.currentTimeMillis()));behaviors.add(new UserBehavior("u3", "i2", 2.5, System.currentTimeMillis()));behaviors.add(new UserBehavior("u3", "i4", 4.0, System.currentTimeMillis()));behaviors.add(new UserBehavior("u3", "i5", 3.5, System.currentTimeMillis()));// 2. 数据预处理DataPreprocessor preprocessor = new DataPreprocessor();Map<String, Map<String, Double>> itemUserMatrix = preprocessor.convertToItemUserMatrix(behaviors);Map<String, Map<String, Double>> userItemMatrix = preprocessor.convertToUserItemMatrix(behaviors);Map<String, Double> userAvgPrefMap = preprocessor.calculateUserAveragePreference(behaviors);// 3. 计算物品相似度ItemSimilarity itemSimilarity = new ItemSimilarity();Map<String, Map<String, Double>> similarityMatrix = itemSimilarity.calculateCosineSimilarity(itemUserMatrix);// 4. 创建推荐引擎ItemBasedRecommender recommender = new ItemBasedRecommender(similarityMatrix, userItemMatrix);// 5. 为用户生成推荐String targetUserId = "u1";List<String> recommendations = recommender.recommendItems(targetUserId, 3);System.out.println("为用户 " + targetUserId + " 推荐的物品:");recommendations.forEach(System.out::println);}
}