我对这个问题有点困惑。但我猜你在问使用什么模型以及如何表示单词。

这似乎是一个非常介绍性的示例，因此您可能想尝试简单的词袋表示。尝试将其作为第一遍推入朴素贝叶斯分类器。如果对结果不满意，可以实现 word2vec 和更复杂的方法。

一种方法是将模型构建为文本分类。给定一个文档语料库，学习预测离散类别的特征。

文本分类可以通过学习嵌入和训练模型来学习将文档分类来完成。

像这样的东西：

from tensorflow.keras.layers                   import Activation, Dense, Embedding, Convolution1D, MaxPooling1D
from tensorflow.keras.models                   import Sequential
from   tensorflow.keras.preprocessing.text     import Tokenizer
from   tensorflow.keras.preprocessing.sequence import pad_sequences
from   tensorflow.keras.utils                  import to_categorical

docs = ['year of education',              
'years education',                 
'years of educational',            
'two years of education',          
'years of education beyond',       
'education four year',             
'of proven sales experience',      
'knowledge of and',                
'experience or education high',    
'assigned knowledge skills',       
'accountable for driving',         
'administrative and leadership skills',    
'advanced negotiations skills',            
'must have keyboarding skills',            
'must have skills',                        
'activities preferred skills',             
'of clinical skill ',]      

# Tranform string documents into numeric sequences
tokenizer = Tokenizer(num_words=100) # Tokenizer lowercases and strips punctuation 
tokenizer.fit_on_texts(X)
X = tokenizer.texts_to_sequences(X) # Convert texts to token sequences

y = ['Education']*5
y += ['Ability']*5
y += ['Experience']*5

# Tranform string categories into numeric categories
mapping = {word: i for i,word in enumerate(y)}  # Hashing trick to encode string as category 
vec = [mapping[word] for word in y]  # Apply hashing trick
y = to_categorical(vec) # Convert numerical values into one-hot encoded values

# Create the model
model = Sequential()

# The Embedding layer is similar to word2vec
model.add(Embedding(input_dim=100, 
                    output_dim=3, 
                    input_length=4
                       ))

# Convolution layer
model.add(Convolution1D(filters=64,
                            kernel_size=2,
                            activation='relu',))
model.add(MaxPooling1D())

# The target specific output layer
model.add(Dense(units=3))
model.add(Activation('softmax'))

model.compile(loss='categorical_crossentropy',
              optimizer='sgd',
              metrics=['accuracy'])

# Train the model
model.fit(X, 
          y, 
          batch_size=256, 
          epochs=100);

# Predict the categories for new data
docs_new = ["There is a requirement of four-year professional degrees", 
         "Able to drive the teams to higher levels",
         "Must have programming experience in C, C++",]

for X_new in docs_new:
    X_new = tokenizer.texts_to_sequences(X_new) 
model.predict(X_new)