Cassandra based Mahout user friend recommendations

问题:

I want to recommend a user , a list of users which the current user can add as friends.

I am using Cassandra and mahout. there is already a implementation of CassandraDataModel in mahout integration package. I want to use this class.

So my recommend-er class looks like follows

public class UserFriendsRecommender {

@Inject
private CassandraDataModel dataModel;

public List<RecommendedItem> recommend(Long userId, int number) throws TasteException{
    UserSimilarity userSimilarity = new PearsonCorrelationSimilarity(dataModel);
    // Optional:
    userSimilarity.setPreferenceInferrer(new AveragingPreferenceInferrer(dataModel));

    UserNeighborhood neighborhood =
              new NearestNUserNeighborhood(3, userSimilarity, dataModel);
    Recommender recommender = new GenericUserBasedRecommender(dataModel, neighborhood, userSimilarity); 
    Recommender cachingRecommender = new CachingRecommender(recommender);
    List<RecommendedItem> recommendations = cachingRecommender.recommend(userId, number);
    return recommendations;
}

}

CassandraDataModel has 4 column familys

static final String USERS_CF = "users";
  static final String ITEMS_CF = "items";
  static final String USER_IDS_CF = "userIDs";
  static final String ITEM_IDS_CF = "itemIDs";

i have a hard time understanding this class especially the column family's. is there any example where i can look for or if someone can explain will be great with a small example.?

javadoc says this

* <p>
 * First, it uses a column family called "users". This is keyed by the user ID
 * as an 8-byte long. It contains a column for every preference the user
 * expresses. The column name is item ID, again as an 8-byte long, and value is
 * a floating point value represnted as an IEEE 32-bit floating poitn value.
 * </p>
 * 
 * <p>
 * It uses an analogous column family called "items" for the same data, but
 * keyed by item ID rather than user ID. In this column family, column names are
 * user IDs instead.
 * </p>
 * 
 * <p>
 * It uses a column family called "userIDs" as well, with an identical schema.
 * It has one row under key 0. It contains a column for every user ID in the
 * model. It has no values.
 * </p>
 * 
 * <p>
 * Finally it also uses an analogous column family "itemIDs" containing item
 * IDs.
 * </p>

回答1:

All the following instructions about required column families by CassandraDataMdoel should be performed in cassandra-cli under the keyspace you created (recommender or other name).

1: Table users

userID is the row key, each itemID has a separate column name, and value is the preference:

CREATE COLUMN FAMILY users
WITH comparator = LongType
AND key_validation_class=LongType
AND default_validation_class=FloatType;

Insert values:

set users[0][0]='1.0';
set users[1][0]='3.0';
set users[2][2]='1.0';

2: Table items

itemID is the row key, each userID has a separate column name, and value is the preference:

CREATE COLUMN FAMILY items
WITH comparator = LongType
AND key_validation_class=LongType
AND default_validation_class=FloatType;

Insert Values:

set items[0][0]='1.0';
set items[0][1]='3.0';
set items[2][2]='1.0';

3: Table userIDs

This table just has one row, but many columns, i.e. each userID has a separate column:

CREATE COLUMN FAMILY userIDs
WITH comparator = LongType
AND key_validation_class=LongType;

Insert Values:

set userIDs[0][0]='';
set userIDs[0][1]='';
set userIDs[0][2]='';

4: Table itemIDs:

This table just has one row, but many columns, i.e. each itemID has a separate column:

CREATE COLUMN FAMILY itemIDs
WITH comparator = LongType
AND key_validation_class=LongType;

Insert Values:

set itemIDs[0][0]='';
set itemIDs[0][1]='';
set itemIDs[0][2]='';

回答2:

Complementing the answer above, for Cassandra 2.0 the new syntax is the following, according that cli is deprecated.

Table users:

CREATE TABLE users (userID bigint, itemID bigint, value float, PRIMARY KEY (userID, itemID));

Table items:

CREATE TABLE items (itemID bigint, userID bigint, value float, PRIMARY KEY (itemID, userID));

Table userIDs:

CREATE TABLE userIDs (id bigint, userID bigint PRIMARY KEY(id, userID));

Table itemIDs:

CREATE TABLE itemIDs (id bigint, itemID bigint PRIMARY KEY(id, itemID));