使用神經網路實作文本分類

近年來，在深度學習熱潮興起的同時，機器學習與人工智慧的話題也再次被提起，這些技術在生活中的應用也越來越廣泛，文本分類也只是其中的一部分。然而使用機器學習或是深度學習實作文本分類的差異有哪些？深度學習的方法一定會優於機器學習嗎？然而在深度學習的方法中，在相同系列但不同的模型所實作的分類結果會有所差異嗎？因此藉由實作機器學習模型與深度學習模型去比較他們之間的結果來探討這些問題。在本文的第二章和第三章會先介紹基本的概念，內容包括神經網路的構造、訓練神經網路模型所會使用到的函數種類與作用，還有深度學習的RNN、LSTM和GRU這三種模型原理的介紹。接著在第四章和第五章則是實作的內容，包含資料的前處理，模型的訓練過程與測試結果紀錄，最後會針對實作的內容進行比較與分析。在最後一章會對這次實作的結果進行檢討，再針對可以深入探討的部分提出研究方向。

關鍵字

深度學習；機器學習；文本分類；神經網路

並列摘要

In recent years, as deep learning is becoming more popular, various topics on machine learning and deep learning are brought up quite frequently. Applications of these technologies can be found in many aspect of our daily life. One of such applications is text classification. What differences can we find when applying machine learning and deep learning to text classification problems? Is the result generated by deep learning always better than that of machine learning? When we apply deep learning algorithms, are different deep learning models going to give us different results? Thus, in this thesis, we will try to apply various models of machine learning and deep learning to address above questions. In this thesis, we will begin by introducing the basic concepts in chapter 2 and 3.The concepts covered include the basic structures of neural networks, how to train network models, and various functions related to the models. In particular, we will introduce concepts and the designs of the RNN, LSTM and GRU models. Then, in chapter 4 and 5, we will cover how these models are implemented. We will cover the preprocessing of data, the training processes and also the testing processes. Finally, we will compare the results of all the tests done. In the final chapter of the thesis, we will give a summary of all our test results, and discuss topics that we should consider in the future.