技術標籤：DWpython

一、背景目的
arXiv 重要的學術公開⽹站，也是搜尋、瀏覽和下載學術論⽂的重要⼯具。arXiv論⽂涵蓋的範圍⾮常 ⼴，涉及物理學的龐⼤分⽀和電腦科學的眾多⼦學科，如數學、統計學、電⽓⼯程、定量⽣物學和經濟學等等。

目的：將使⽤arXiv在公開的17萬篇論⽂資料集，通過資料分析能夠挖掘出最近學術的發展趨勢和學術關鍵詞。

使用工具：python

主題：論文分類（資料建模任務），利用已有資料建模，對新論文進行類別分類；

二、資料處理
2.1 匯入包

#匯入所需的package並讀取原始資料
import seaborn as sns
import re

import json
import pandas as pd  # 資料處理和分析
import matplotlib.pyplot as plt  # 畫圖工具
 

2.2 欄位讀取

data = []
with open('arxiv-metadata-oai-2019.json','r') as f:
    for idx, line in enumerate(f):
        d = json.loads(line)
        d = {'title':d['title'],'categories':d['categories'],'abstract':d['abstract']}
        data.append(d)
        
        # 只選取部分資料
        if idx > 200000:
            break

data = pd.DataFrame(data)
data.head()
 

2.3 將標題和摘要拼接一起完成分類

# 合併title和abstract
data['text'] = data['title'] + data['abstract']

# 將換行符替換為空格
data['text'] = data['text'].apply(lambda x: x.replace('\n',''))

# 將所有大寫字母替換為小寫字母
data['text'] = data['text'].apply(lambda x: x.lower())

# 刪除多餘的列
data = data.drop(['abstract','title'], axis=1)
data.head()
 

2.4 處理類別（原始論文有可能有多個類別）

# 多個類別，包含子分類
data['categories'] = data['categories'].apply(lambda x: x.split(' '))

# 單個類別，不包含子分類
data['categories_big'] = data['categories'].apply(lambda x: [xx.split('.')[0] for xx in x])
data.head()

2.5 將類別進行編碼，這裡類別是多個，所以需要多編碼：

from sklearn.preprocessing import MultiLabelBinarizer
mlb = MultiLabelBinarizer()
data_label = mlb.fit_transform(data['categories_big'].iloc[:])
data_label

三、論文分類
3.1 方法一、（1）使用TFIDF提取特徵（限制最多4000個單詞）

from sklearn.feature_extraction.text import TfidfVectorizer
vectorizer = TfidfVectorizer(max_features=4000)           # 限制最多詞4000
data_tfidf = vectorizer.fit_transform(data['text'].iloc[:])
data_tfidf

3.2 方法一、（2）使用sklearn的多標籤分類進行封裝（多標籤分類）

# 劃分訓練集和測試集
from sklearn.model_selection import train_test_split
x_train,x_test,y_train,y_test = train_test_split(data_tfidf,  data_label, test_size=0.2, random_state=1)

# 構建多標籤分類模型
from sklearn.multioutput import MultiOutputClassifier
from sklearn.naive_bayes import MultinomialNB
clf =MultiOutputClassifier(MultinomialNB()).fit(x_train,y_train)

# 精度評價
from sklearn.metrics import accuracy_score
accuracy_score(y_test,clf.predict(x_test))

3.3 方法二、（1）使用深度學習模型，單詞進行詞嵌入然後訓練

#思路2使用深度學習模型，單詞進行詞嵌入然後訓練。首先按照文字劃分資料集


from sklearn.model_selection import train_test_split
x_train, x_test, y_train, y_test = train_test_split(data['text'].iloc[:], data_label,
                                                 test_size = 0.2,random_state = 1)

3.3 方法二、（2）將資料集處理進行編碼，並進行截斷

# parameter
from sklearn.model_selection import train_test_split
x_train, x_test, y_train, y_test = train_test_split(data['text'].iloc[:100000], 
                                                    data_label[:100000],
                                                 test_size = 0.95,random_state = 1)
# parameter
max_features= 500
max_len= 150
embed_size=100
batch_size = 128
epochs = 5

from keras.preprocessing.text import Tokenizer
from keras.preprocessing import sequence

tokens = Tokenizer(num_words = max_features)
tokens.fit_on_texts(list(data['text'].iloc[:100000]))

y_train = data_label[:100000]
x_sub_train = tokens.texts_to_sequences(data['text'].iloc[:100000])
x_sub_train = sequence.pad_sequences(x_sub_train, maxlen=max_len)

定義模型並完成訓練：

# LSTM model
# Keras Layers:
from keras.layers import Dense,Input,LSTM,Bidirectional,Activation,Conv1D,GRU
from keras.layers import Dropout,Embedding,GlobalMaxPooling1D, MaxPooling1D, Add, Flatten
from keras.layers import GlobalAveragePooling1D, GlobalMaxPooling1D, concatenate, SpatialDropout1D# Keras Callback Functions:
from keras.callbacks import Callback
from keras.callbacks import EarlyStopping,ModelCheckpoint
from keras import initializers, regularizers, constraints, optimizers, layers, callbacks
from keras.models import Model
from keras.optimizers import Adam

sequence_input = Input(shape=(max_len, ))
x = Embedding(max_features, embed_size,trainable = False)(sequence_input)
x = SpatialDropout1D(0.2)(x)
x = Bidirectional(GRU(128, return_sequences=True,dropout=0.1,recurrent_dropout=0.1))(x)
x = Conv1D(64, kernel_size = 3, padding = "valid", kernel_initializer = "glorot_uniform")(x)
avg_pool = GlobalAveragePooling1D()(x)
max_pool = GlobalMaxPooling1D()(x)
x = concatenate([avg_pool, max_pool]) 
preds = Dense(20, activation="sigmoid")(x)

model = Model(sequence_input, preds)
model.compile(loss='binary_crossentropy',optimizer=Adam(lr=1e-3),metrics=['accuracy'])
model.fit(x_sub_train, y_train, batch_size=batch_size, epochs=epochs)