Sentence Semantic Similarity Learning Based On Deep Learning

Deep learning is a part of machine learning methods based on data representations learning,as opposed to task-specific algorithms.It can make full use of a variety of massive data,learn to get abstract knowledge representations automatically,and transform the original data into a specific form that the final task required.In recent years,deep learning has set off a new research boom in many areas.Sentence semantic similarity modeling is to calculate the similarity between two sentences at the semantic level.It is an important and fundamental technique in the field of natural language processing and it plays an important role in many tasks.Recently,with the development of deep learning,the deep learning method has become the most fiercely studied model in the field of sentence semantic similarity.In this paper,we improve the sentence semantic similarity model in two ways:1.One is focus on the neural network,and improves the dropout algorithm.The dropout probability of each neuron in the traditional dropout algorithm is selected randomly,which may discard some important information of the input.Our improved dropout algorithm determines the dropout probability based on the importance and dependencies of input elements so that such important elements will be retained and the loss of information will be less.2.Another is to combine the deep learning model and feature engineering together to capture features of sentences.The feature our model is comprehensive,it not only takes the global semantic information into account,but also further refines the semantic information at the word level.The model we implement can also apply in other tasks of Natural Language Processing.We can choose different neural networks and formulate different features according to the needs of the task and the characteristics of the input,we can also join the attention mechanism.From the experimental results,we can infer that both methods have improved the accuracy of the sentence semantic similarity modeling to a certain extent,and also have made a useful exploration for later research.