标签：predictionio e-commerce recommendation 买了又买-物相似
原来的算法模型中，没有考虑买了又买的物品，对于钢材行业，用户的需求是一致的，经常买了又买，所以增加此功能
a 模型训练中增加如下代码
//buy item for users
JavaPairRDD<String, String> buyItemForUser = data.getBuyEvents().mapToPair(new PairFunction<UserItemEvent, Tuple2<String, String>, Integer>() {
@Override
public Tuple2<Tuple2<String, String>, Integer> call(UserItemEvent buyEvent) throws Exception {
return new Tuple2<>(new Tuple2<>(buyEvent.getUser(), buyEvent.getItem()), 1);
}
}).mapToPair(new PairFunction<Tuple2<Tuple2<String, String>, Integer>, String, Integer>() {
@Override
public Tuple2<String, Integer> call(Tuple2<Tuple2<String, String>, Integer> element) throws Exception {
return new Tuple2<>(element._1()._1()+":::"+element._1()._2(), element._2());
}
}).reduceByKey(new Function2<Integer, Integer, Integer>() {
@Override
public Integer call(Integer integer, Integer integer2) throws Exception {
return integer + integer2;
}
}).mapToPair(new PairFunction<Tuple2<String, Integer>, String, String>() {
@Override
public Tuple2<String, String> call(Tuple2<String, Integer> element) throws Exception {
String temp[]=element._1().split(":::");
if(temp.length==2){
return new Tuple2<>(temp[0], temp[1]);
}
return new Tuple2<>("", "");
}
});
b 增加预测算法
private List<ItemScore> buyItemForUser(Model model, Query query){
logger.info("start to add buy item for the user");
final JavaRDD<ItemScore> matchedUser = model.getUserBuyItem().filter(new Function<Tuple2<String, String>, Boolean>() {
@Override
public Boolean call(Tuple2<String, String> userIndex) throws Exception {
return userIndex._1().equals(query.getUserEntityId());
}
}).map(new Function<Tuple2<String,String>,ItemScore>() {
@Override
public ItemScore call(Tuple2<String, String> arg0) throws Exception {
return new ItemScore(arg0._2(),10);
}
});
return matchedUser.collect();
}
c topItemsForUser 按照你的业务逻辑出来两者的排序规则基于物的相似性，也是我们期望，加入以下代码
//根据事件查找物品的属性
private List<Set<String>> getRecentProductCategory(Query query, Model model) {
try {
List<Set<String>> result = new ArrayList<>();
List<Event> events = LJavaEventStore.findByEntity(
ap.getAppName(),
"user",
query.getUserEntityId(),
OptionHelper.<String>none(),
OptionHelper.some(ap.getSimilarItemEvents()),
OptionHelper.some(OptionHelper.some("item")),
OptionHelper.<Option<String>>none(),
OptionHelper.<DateTime>none(),
OptionHelper.<DateTime>none(),
OptionHelper.some(10),
true,
Duration.apply(10, TimeUnit.SECONDS));
for (final Event event : events) {
if (event.targetEntityId().isDefined()) {
JavaPairRDD<String, Integer> filtered = model.getItemIndex().filter(new Function<Tuple2<String, Integer>, Boolean>() {
@Override
public Boolean call(Tuple2<String, Integer> element) throws Exception {
return element._1().equals(event.targetEntityId().get());
}
});
final String itemIndex = filtered.first()._1();
Item item = model.getItems().get(itemIndex);
if(item.getCategories()!=null && item.getCategories().size()>0){
result.add(item.getCategories());
}
}
}
return result;
} catch (Exception e) {
logger.error("Error reading recent events for user " + query.getUserEntityId());
throw new RuntimeException(e.getMessage(), e);
}
}
//相似性比较
private List<ItemScore> similarItemsByCategory(final List<Set<String>> category, Model model, Query query) {
Map<String, Item> items =model.getItems();
if(items==null || items.size()==0){
return null;
}
if(category==null || category.size()==0){
return null;
}
JavaRDD<ItemScore> itemScores = model.getItemIndex().map(new Function<Tuple2<String, Integer>, ItemScore>() {
@Override
public ItemScore call(Tuple2<String, Integer> idItem) throws Exception {
String itemid= idItem._1();
Item item = items.get(itemid);
double similarity = 0.0;
for(int i=0 ; i<category.size(); i++){
similarity+=getDistance(category.get(i),item.getCategories());
}
logger.info(itemid+"->"+similarity);
return (new ItemScore(itemid, similarity));
}
});
itemScores = validScores(itemScores, query.getWhitelist(), query.getBlacklist(), query.getCategories(), model.getItems(), query.getUserEntityId());
return sortAndTake(itemScores, query.getNumber());
/* List<ItemScore> itemScores=new ArrayList<ItemScore>();
for (Map.Entry<String, Item> entry : items.entrySet()) {
Item it = entry.getValue();
double similarity = 0.0;
for(int i=0 ; i<category.size(); i++){
similarity+=getDistance(category.get(i),it.getCategories());
}
itemScores.add(new ItemScore(it.getEntityId(), similarity));
}
itemScores = validScoresForList(itemScores, query.getWhitelist(), query.getBlacklist(), query.getCategories(), model.getItems(), query.getUserEntityId());
return sortAndTake(itemScores, query.getNumber());*/
}
//相似算法，比较简单
public static int getDistance(Set<String> t, Set<String> s) {
if (t==null || t.size()==0 || s==null || s.size()==0 || t.size() != s.size()) {
return 0;
}
HashSet<String> t_temp=new HashSet<String>(t);//必须转一下
HashSet<String> s_temp=new HashSet<String>(s);
t_temp.retainAll(s_temp);
return t_temp.size();
}
最后按照你的业务逻辑，加入相似的物品。predictionIO E-Commerce Recommendation 买了又买-物相似
标签：predictionio e-commerce recommendation 买了又买-物相似
原文地址：http://12597095.blog.51cto.com/12587095/1983709
