def _build_low_layers(self, args):
self.user_emb_matrix = tf.get_variable('user_emb_matrix',
[self.n_user, args.dim])
self.item_emb_matrix = tf.get_variable('item_emb_matrix',
[self.n_item, args.dim])
self.entity_emb_matrix = tf.get_variable('entity_emb_matrix',
[self.n_entity, args.dim])
self.relation_emb_matrix = tf.get_variable('relation_emb_matrix',
[self.n_relation, args.dim])
# [batch_size, dim]
self.user_embeddings = tf.nn.embedding_lookup(self.user_emb_matrix,
self.user_indices)
self.item_embeddings = tf.nn.embedding_lookup(self.item_emb_matrix,
self.item_indices)
self.head_embeddings = tf.nn.embedding_lookup(self.entity_emb_matrix,
self.head_indices)
self.relation_embeddings = tf.nn.embedding_lookup(
self.relation_emb_matrix, self.relation_indices)
self.tail_embeddings = tf.nn.embedding_lookup(self.entity_emb_matrix,
self.tail_indices)
for _ in range(args.L):
user_mlp = Dense(input_dim=args.dim, output_dim=args.dim)
tail_mlp = Dense(input_dim=args.dim, output_dim=args.dim)
cc_unit = CrossCompressUnit(args.dim)
self.user_embeddings = user_mlp(self.user_embeddings)
self.item_embeddings, self.head_embeddings = cc_unit(
[self.item_embeddings, self.head_embeddings])
self.tail_embeddings = tail_mlp(self.tail_embeddings)
self.vars_rs.extend(user_mlp.vars)
self.vars_rs.extend(cc_unit.vars)
self.vars_kge.extend(tail_mlp.vars)
self.vars_kge.extend(cc_unit.vars)
def __init__(self, args, n_users, n_items, n_entities, n_relations):
super(MKR, self).__init__()
self.n_user = n_users
self.n_item = n_items
self.n_entity = n_entities
self.n_relation = n_relations
self.L = args.L
self.H = args.H
self.dim = args.dim
# 定义embedding矩阵
self.user_emb_matrix = nn.Embedding(n_users, args.dim)
self.item_emb_matrix = nn.Embedding(n_items, args.dim)
self.entity_emb_matrix = nn.Embedding(n_entities, args.dim)
self.relation_emb_matrix = nn.Embedding(n_relations, args.dim)
# 定义网络
self.user_mlps, self.tail_mlps, self.cc_units = [], [], []
self.kge_mlps = []
for _ in range(args.L):
self.user_mlps.append(Dense(args.dim, args.dim))
self.tail_mlps.append(Dense(args.dim, args.dim))
self.cc_units.append(CrossCompressUnit(args.dim))
for _ in range(args.H):
self.kge_mlps.append(Dense(args.dim * 2, args.dim * 2))
self.kge_pred_mlp = Dense(args.dim * 2, args.dim)
self.sigmoid = nn.Sigmoid()
def _build_low_layers(self, args):
if self.restore_path is None:
self.user_emb_matrix = tf.get_variable('user_emb_matrix', [self.n_user, args.dim])
self.item_emb_matrix = tf.get_variable('item_emb_matrix', [self.n_item, args.dim])
self.entity_emb_matrix = tf.get_variable('entity_emb_matrix', [self.n_entity, args.dim])
self.relation_emb_matrix = tf.get_variable('relation_emb_matrix', [self.n_relation, args.dim])
else:
self.user_emb = tf.placeholder(tf.float32, [None, args.dim], 'user_emb')
self.item_emb = tf.placeholder(tf.float32, [None, args.dim], 'item_emb')
self.entity_emb = tf.placeholder(tf.float32, [None, args.dim], 'entity_emb')
self.relation_emb = tf.placeholder(tf.float32, [None, args.dim], 'relation_emb')
self.user_emb_matrix = tf.Variable(tf.truncated_normal([self.n_user, args.dim]), name='user_emb_matrix', trainable=True)
self.item_emb_matrix = tf.Variable(tf.truncated_normal([self.n_item, args.dim]), name='item_emb_matrix', trainable=True)
self.entity_emb_matrix = tf.Variable(tf.truncated_normal([self.n_entity, args.dim]), name='entity_emb_matrix', trainable=True)
self.relation_emb_matrix = tf.Variable(tf.truncated_normal([self.n_relation, args.dim]), name='relation_emb_matrix', trainable=True)
self.user_emb_init = self.user_emb_matrix.assign(self.user_emb)
self.item_emb_init = self.item_emb_matrix.assign(self.item_emb)
self.entity_emb_init = self.entity_emb_matrix.assign(self.entity_emb)
self.relation_emb_init = self.relation_emb_matrix.assign(self.relation_emb)
# [batch_size, dim]
self.user_embeddings = tf.nn.embedding_lookup(self.user_emb_matrix, self.user_indices)
self.item_embeddings = tf.nn.embedding_lookup(self.item_emb_matrix, self.item_indices)
self.head_embeddings = tf.nn.embedding_lookup(self.entity_emb_matrix, self.head_indices)
self.relation_embeddings = tf.nn.embedding_lookup(self.relation_emb_matrix, self.relation_indices)
self.tail_embeddings = tf.nn.embedding_lookup(self.entity_emb_matrix, self.tail_indices)
for _ in range(args.L):
user_mlp = Dense(input_dim=args.dim, output_dim=args.dim, dropout=self.dropout_param)
tail_mlp = Dense(input_dim=args.dim, output_dim=args.dim, dropout=self.dropout_param)
cc_unit = CrossCompressUnit(args.dim)
self.user_embeddings = user_mlp(self.user_embeddings)
self.item_embeddings, self.head_embeddings = cc_unit([self.item_embeddings, self.head_embeddings])
self.tail_embeddings = tail_mlp(self.tail_embeddings)
self.vars_rs.extend(user_mlp.vars)
self.vars_rs.extend(cc_unit.vars)
self.vars_kge.extend(tail_mlp.vars)
self.vars_kge.extend(cc_unit.vars)
def _build_low_layers(self, args):
# 创建用户嵌入矩阵的tf变量，其shape为用户数*嵌入维度
self.user_emb_matrix = tf.get_variable('user_emb_matrix',
[self.n_user, args.dim])
self.item_emb_matrix = tf.get_variable('item_emb_matrix',
[self.n_item, args.dim])
self.entity_emb_matrix = tf.get_variable('entity_emb_matrix',
[self.n_entity, args.dim])
self.relation_emb_matrix = tf.get_variable('relation_emb_matrix',
[self.n_relation, args.dim])
# [batch_size, dim]
# 选取user_emb_matrix张量中，user_indices为索引时对应的元素
self.user_embeddings = tf.nn.embedding_lookup(self.user_emb_matrix,
self.user_indices)
self.item_embeddings = tf.nn.embedding_lookup(self.item_emb_matrix,
self.item_indices)
self.head_embeddings = tf.nn.embedding_lookup(self.entity_emb_matrix,
self.head_indices)
self.relation_embeddings = tf.nn.embedding_lookup(
self.relation_emb_matrix, self.relation_indices)
self.tail_embeddings = tf.nn.embedding_lookup(self.entity_emb_matrix,
self.tail_indices)
for _ in range(args.L):
# Dense 一个线性回归过程
user_mlp = Dense(input_dim=args.dim, output_dim=args.dim)
tail_mlp = Dense(input_dim=args.dim, output_dim=args.dim)
# 交叉压缩单元
cc_unit = CrossCompressUnit(args.dim)
self.user_embeddings = user_mlp(self.user_embeddings)
self.item_embeddings, self.head_embeddings = cc_unit(
[self.item_embeddings, self.head_embeddings])
self.tail_embeddings = tail_mlp(self.tail_embeddings)
self.vars_rs.extend(user_mlp.vars)
self.vars_rs.extend(cc_unit.vars)
self.vars_kge.extend(tail_mlp.vars)
self.vars_kge.extend(cc_unit.vars)
def __init__(self, args, n_user, n_item, n_entity, n_relation, use_inner_product=True):
super(MKR_model, self).__init__()
# <Lower Model>
self.args = args
self.n_user = n_user
self.n_item = n_item
self.n_entity = n_entity
self.n_relation = n_relation
self.use_inner_product = use_inner_product
# Init embeddings
self.user_embeddings_lookup = nn.Embedding(self.n_user, self.args.dim)
self.item_embeddings_lookup = nn.Embedding(self.n_item, self.args.dim)
self.entity_embeddings_lookup = nn.Embedding(self.n_entity, self.args.dim)
self.relation_embeddings_lookup = nn.Embedding(self.n_relation, self.args.dim)
self.user_mlp = nn.Sequential()
self.tail_mlp = nn.Sequential()
self.cc_unit = nn.Sequential()
for i_cnt in range(self.args.L):
self.user_mlp.add_module('user_mlp{}'.format(i_cnt),
Dense(self.args.dim, self.args.dim))
self.tail_mlp.add_module('tail_mlp{}'.format(i_cnt),
Dense(self.args.dim, self.args.dim))
self.cc_unit.add_module('cc_unit{}'.format(i_cnt),
CrossCompressUnit(self.args.dim))
# <Higher Model>
self.kge_pred_mlp = Dense(self.args.dim * 2, self.args.dim)
self.kge_mlp = nn.Sequential()
for i_cnt in range(self.args.H - 1):
self.kge_mlp.add_module('kge_mlp{}'.format(i_cnt),
Dense(self.args.dim * 2, self.args.dim * 2))
if self.use_inner_product==False:
self.rs_pred_mlp = Dense(self.args.dim * 2, 1)
self.rs_mlp = nn.Sequential()
for i_cnt in range(self.args.H - 1):
self.rs_mlp.add_module('rs_mlp{}'.format(i_cnt),
Dense(self.args.dim * 2, self.args.dim * 2))
def __init__(self, args, n_user, n_item, n_entity, n_relation):
super(Test_model, self).__init__()
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
self.args = args
self.n_user = n_user
self.n_item = n_item
self.n_entity = n_entity
self.n_relation = n_relation
self.user_embeddings_lookup = nn.Embedding(self.n_user, self.args.dim).to(self.device)
self.item_embeddings_lookup = nn.Embedding(self.n_item, self.args.dim).to(self.device)
self.entity_embeddings_lookup = nn.Embedding(self.n_entity, self.args.dim).to(self.device)
self.relation_embeddings_lookup = nn.Embedding(self.n_relation, self.args.dim).to(self.device)
L_ = self.args.L_hgn
d_ = self.args.dim
self.feature_gate_item = nn.Linear(d_, d_).to(self.device)
self.feature_gate_user = nn.Linear(d_, d_).to(self.device)
self.instance_gate_item = Variable(torch.zeros(d_, 1).type(torch.FloatTensor),requires_grad=True).to(self.device)
self.instance_gate_user = Variable(torch.zeros(d_, L_).type(torch.FloatTensor),requires_grad=True).to(self.device)
self.instance_gate_item = torch.nn.init.xavier_uniform_(self.instance_gate_item)
self.instance_gate_user = torch.nn.init.xavier_uniform_(self.instance_gate_user)
self.user_embeddings_lookup.weight.data.normal_(0, 1.0 / self.user_embeddings_lookup.embedding_dim)
self.item_embeddings_lookup.weight.data.normal_(0, 1.0 / self.item_embeddings_lookup.embedding_dim)
self.entity_embeddings_lookup.weight.data.normal_(0, 1.0 / self.entity_embeddings_lookup.embedding_dim)
self.relation_embeddings_lookup.weight.data.normal_(0, 1.0 / self.relation_embeddings_lookup.embedding_dim)
self.W2 = nn.Embedding(self.n_item, d_, padding_idx=0).to(self.device)
self.b2 = nn.Embedding(self.n_item, 1, padding_idx=0).to(self.device)
self.W2.weight.data.normal_(0, 1.0 / self.W2.embedding_dim)
self.b2.weight.data.zero_()
# CNN Layer
n_filters, filter_height = self.args.n_filters, self.args.filter_height
self.horizontal_cnn_layer = nn.Conv2d(in_channels=1,out_channels=self.args.n_filters,kernel_size=(filter_height,d_),stride=1).to(self.device)
self.vertical_cnn_layer = nn.Conv2d(in_channels=1,out_channels=self.args.n_filters,kernel_size=(L_,1),stride=1).to(self.device)
# Maxpooling Layer
self.maxpooling_layer = nn.MaxPool2d((1,L_-1), stride=1)
# Fully Connected Layer
self.W_FC_layer = Variable(torch.zeros(n_filters+(n_filters*d_), d_).type(torch.FloatTensor),requires_grad=True).to(self.device)
self.W_FC_layer = torch.nn.init.xavier_uniform_(self.W_FC_layer)
self.W_plun_FC_layer = Variable(torch.zeros(2*d_, d_).type(torch.FloatTensor),requires_grad=True).to(self.device)
self.W_plun_FC_layer = torch.nn.init.xavier_uniform_(self.W_plun_FC_layer)
self.tail_mlp = nn.Sequential()
self.cc_unit = nn.Sequential()
for i_cnt in range(self.args.L_mkr):
self.tail_mlp.add_module('tail_mlp{}'.format(i_cnt),
Dense(self.args.dim, self.args.dim,self.device))
self.cc_unit.add_module('cc_unit{}'.format(i_cnt),
CrossCompressUnit(self.args.dim,self.device))
self.kge_pred_mlp = Dense(self.args.dim *2, self.args.dim,self.device)
self.kge_mlp = nn.Sequential()
for i_cnt in range(self.args.H_mkr-1):
self.kge_mlp.add_module('kge_mlp{}'.format(i_cnt),
Dense(self.args.dim *2, self.args.dim *2,self.device))
self.W_r = Variable(torch.zeros(d_, d_).type(torch.FloatTensor),requires_grad=True).to(self.device)
self.W_r = torch.nn.init.xavier_uniform_(self.W_r)