-
Notifications
You must be signed in to change notification settings - Fork 1
/
data_reader.py
201 lines (162 loc) · 8.6 KB
/
data_reader.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
import os
from tqdm import tqdm
import numpy as np
import torch
from Utils import config
class Lang:
def __init__(self):
self.word2count = {}
def add_funs(self, init_index2word):
self.init_index2word = init_index2word
self.word2count = {str(v): 1 for k, v in init_index2word.items()}
self.n_words = len(init_index2word) # Count default tokens
self.word2index = {str(v): int(k) for k, v in init_index2word.items()}
self.index2word = init_index2word
def index_words(self, sentence):
for word in sentence:
self.index_word(word)
def index_word(self, word):
if word not in self.word2count:
self.word2count[word] = 1
else:
self.word2count[word] += 1
def solve_behavior_data(data_dir, User_POI_dict):
# Behavior (Input data)
Click_dict = np.load(os.path.join(data_dir, 'Source_Data/Click_UP.npy'), allow_pickle=True).item()
Favor_dict = np.load(os.path.join(data_dir, 'Source_Data/Favor_UP.npy'), allow_pickle=True).item()
Consume_dict = np.load(os.path.join(data_dir, 'Source_Data/Buy_UP.npy'), allow_pickle=True).item()
behavior_graph = {}
for uid in User_POI_dict:
if uid not in Click_dict:
print("ERROR! click can not cover total user.")
Favor_dict[uid] = [] if uid not in Favor_dict else Favor_dict[uid]
Consume_dict[uid] = [] if uid not in Consume_dict else Consume_dict[uid]
behavior_graph[uid] = {'click': Click_dict[uid], 'favor': Favor_dict[uid], 'consume': Consume_dict[uid],
'review': User_POI_dict[uid]}
return behavior_graph
def solve_cl_data(data_dir, pos_neg_sample, User_POI_dict):
# CL (Input data)
POI_User_dict = np.load(os.path.join(data_dir, 'Source_Data/POI_User.npy'), allow_pickle=True).item()
positive_pu_dict, negative_pu_dict = {}, {}
for pid in POI_User_dict:
positive_pu = POI_User_dict[pid]
if len(positive_pu) > pos_neg_sample:
positive_pu_dict[pid] = np.random.choice(positive_pu, pos_neg_sample, replace=False)
else:
positive_pu_dict[pid] = positive_pu
negative_pu = []
for uid in User_POI_dict:
if uid not in positive_pu:
negative_pu.append(uid)
if len(negative_pu) > pos_neg_sample:
negative_pu_dict[pid] = np.random.choice(negative_pu, pos_neg_sample, replace=False)
else:
negative_pu_dict[pid] = negative_pu
return positive_pu_dict, negative_pu_dict
def read_source_data(para):
# input data
User_POI_dict = np.load(os.path.join(para.data_dir, 'Source_Data/User_POI_rank.npy'), allow_pickle=True).item()
UGC_Aspect_Word_dict = np.load(os.path.join(para.data_dir, 'Source_Data/UGC_Aspect_Word.npy'), allow_pickle=True).item()
User_PR_dict = np.load(os.path.join(para.data_dir, 'Source_Data/User_PR.npy'), allow_pickle=True).item()
# POI_Info_dict = np.load(os.path.join(para.data_dir, 'POI_Info.npy'), allow_pickle=True).item()
# output data
UGC_RankedTags_dict = np.load(os.path.join(para.data_dir, 'Source_Data/UGC_RankedTags.npy'), allow_pickle=True).item()
UGC_RankedAspects_dict = np.load(os.path.join(para.data_dir, 'Source_Data/UGC_RankedAspects.npy'), allow_pickle=True).item()
UGC_UP_dict = np.load(os.path.join(para.data_dir, 'Source_Data/UGC_UP.npy'), allow_pickle=True).item()
# Construct WAP data of WAPU-meta path and WAU data of WAU-meta path in HHRG
user_history_aw_dict, poi_history_aw_dict = {}, {}
for uid in User_PR_dict:
user_history_aw_dict[uid] = {}
# user_total_words_number = {}
for pr_tuple in User_PR_dict[uid]:
rid = pr_tuple[1]
pid = pr_tuple[0]
if pid not in poi_history_aw_dict:
poi_history_aw_dict[pid] = {}
for a in UGC_Aspect_Word_dict[rid]:
if a not in user_history_aw_dict[uid]:
user_history_aw_dict[uid][a] = []
user_history_aw_dict[uid][a].extend(UGC_Aspect_Word_dict[rid][a])
if a not in poi_history_aw_dict[pid]:
poi_history_aw_dict[pid][a] = []
poi_history_aw_dict[pid][a].extend(UGC_Aspect_Word_dict[rid][a])
positive_pu_dict, negative_pu_dict = solve_cl_data(para.data_dir, para.pos_neg_sample, User_POI_dict)
behavior_graph = solve_behavior_data(para.data_dir, User_POI_dict)
src_dict = (user_history_aw_dict, poi_history_aw_dict, positive_pu_dict, negative_pu_dict, behavior_graph)
# Construct data pair
data_pair, input_words_appear_number = [], 0
vocab = Lang()
for rid in tqdm(UGC_UP_dict):
uid = UGC_UP_dict[rid][0]
pid = UGC_UP_dict[rid][1]
single_data = {}
single_data['src_Uid'] = uid
single_data['src_Pid'] = pid
single_data['src_Rid'] = rid
single_data['src_Up'] = User_POI_dict[uid] # user reviewed poi
# Statistics word frequency
for user_pid in User_POI_dict[uid]: # input_user poi_list
for user_pid_aspect in poi_history_aw_dict[user_pid]:
vocab.index_words(poi_history_aw_dict[user_pid][user_pid_aspect])
input_words_appear_number += len(poi_history_aw_dict[user_pid][user_pid_aspect])
for pid_aspect in poi_history_aw_dict[pid]: # input_poi
vocab.index_words(poi_history_aw_dict[pid][pid_aspect])
input_words_appear_number += len(poi_history_aw_dict[pid][pid_aspect])
label_tags_dict = {} # {0: [ASP, w1, w2, EOS], 1: [ASP, w1, w2, EOS]} key: rank，value: asp+word
label_rank = torch.zeros([len(config.aspect_token)], dtype=torch.long)
for aw_index in range(len(UGC_RankedTags_dict[rid])):
aspect_tags = []
word_list = UGC_RankedTags_dict[rid][aw_index]
aspect_id = UGC_RankedAspects_dict[rid][aw_index]
aspect_tags.append(config.aspect_token[aspect_id + config.begin]) # ASP
aspect_tags.extend(word_list) # word
aspect_tags.append("EOS")
label_tags_dict[aw_index] = aspect_tags
label_rank[aspect_id] = aw_index + 1 # rank
vocab.index_words(word_list)
input_words_appear_number += len(word_list)
single_data['src_label_tags'] = label_tags_dict
label_score = max(label_rank) + 1 - label_rank
final_label_score = (label_score < max(label_rank) + 1) * label_score
single_data['src_label_rank'] = final_label_score.tolist()
poi_sequence_words = []
for aspect in poi_history_aw_dict[pid]:
poi_sequence_words.append(config.aspect_token[aspect + config.begin])
poi_sequence_words.extend(poi_history_aw_dict[pid][aspect])
poi_sequence_words.append("EOS")
# Transformer input
single_data['src_Pw_seq'] = poi_sequence_words # [t1_word1, t1_word2, 'ASP', t2_word, 'ASP', t3_word, 'EOS']
data_pair.append(single_data)
# filter vocab words
w2c = dict(sorted(vocab.word2count.items(), key=lambda kv: (kv[1], kv[0]), reverse=True))
# default tokens
tokens = {config.PAD_idx: "PAD", config.UNK_idx: "UNK", config.BOS_idx: "BOS", config.EOS_idx: "EOS",
config.SEP_idx: "SEP", config.CLSU_idx: "CLSU", config.CLSP_idx: "CLSP",
config.CLSUA_idx: "CLSUA", config.CLSPA_idx: "CLSPA"}
tokens.update(config.aspect_token)
vocab.add_funs(tokens)
dict_words_appear_number = 0
for w in w2c:
if w2c[w] <= 150:
break
vocab.word2index[w] = vocab.n_words
vocab.index2word[vocab.n_words] = w
vocab.n_words += 1
dict_words_appear_number += w2c[w]
print("vocab_size:", vocab.n_words)
print("ratio: vocab_size | total_words:", vocab.n_words / len(w2c))
print("ratio: vocab_words_appear_number | total_words_appear_number:", dict_words_appear_number / input_words_appear_number)
# Construct word2vector
# Download from https://ai.tencent.com/ailab/nlp/en/embedding.html
# word_vector_Tencent = open(os.path.join(para.data_dir, 'Tencent_AILab_ChineseEmbedding.txt'), 'r', encoding='utf-8')
word_vector_Tencent = open('/home/hadoop-aipnlp/cephfs/data/zhaomengxue/2022SIGIR/Data/Tencent_AILab_ChineseEmbedding.txt', 'r', encoding='utf-8')
vector_dict = {}
for i, line in enumerate(word_vector_Tencent.readlines()):
wv = line.rstrip('\n').split(" ")
w = wv[0]
if w in vocab.word2index:
v = wv[1:]
v = [float(value) for value in v]
vector_dict[w] = v
print('pretrained vector size: ', len(vector_dict))
return data_pair, vocab, vector_dict, src_dict