結合語意關鍵詞與卷積神經網路之文本分類研究

A Study of Text Classification Based on Keyword Semantics and Convolutional Neural Network

黃臆榤(Yi-Jie Huang)

指導教授 ： 許聞廉

國立清華大學/電機資訊學院/資訊工程學系所/碩士(2018年)

摘要

近年來，卷積神經網路不只於影像處理領域中取得重大成果，於自然語言處理相關課題中也有一定成效。本論文提出一個結合語意關鍵詞的卷積神經網路模型，並應用於自然語言處理中的文本分類問題。語意關鍵詞從廣義知網中萃取知識用以建立詞彙意義，將其結果應用於卷積神經網路，獲得具有語意概念的深度學習模型。
本研究於新聞主題偵測領域與讀者情緒分析領域進行實驗，分別使用兩筆資料集各含有將近140,000篇與50,000篇的雅虎（Yahoo）中文新聞。實驗結果顯示本系統於新聞主題偵測資料集優於其他常見的主題模型，如LDA-SVM，而且Macro平均的F1-measure較FBA提升7.5\%；於讀者情緒分析資料集中Macro平均之F1-measure達到90%以上的優異成果，並且比TBA提高將近6%。
根據實驗結果顯示，本論文之模型能夠結合語意關鍵詞與卷積神經網路之優點，使模型能學習人類的知識且準確預測結果。

關鍵字

文本分類 ； 深度學習 ； 卷積神經網路 ； 語意關鍵詞 ； 主題偵測 ； 讀者情緒分析

並列摘要

In recent years, convolutional neural network (CNN) achieved a remarkable success not only in image processing, but also in natural language processing. In this thesis, we proposed a deep learning model combining keyword semantics with convolutional neural network and applied it to text classification in natural language processing. The keyword semantics extracted knowledge from E-HowNet and generated the sense of words. Moreover, these structures are regarded as input of CNNs to acquire a deep learning model with keyword semantics.
We conducted experiments with our model on text classification in two domains, topic detection and reader emotion analysis. There are two corpora for topic detection and reader emotion respectively. Moreover, the experiment results show that our model improves the macro F1-measure by 7.5% compared to several well-known topic models including LDA-SVM and FBA. According to the experiment results, our model combines both benefits of keyword semantics and convolutional nerual network. Thus, allowing the model to accurately predict the result by learning human knowledge.

並列關鍵字

Text Classification ； Deep Learning ； Convolutional Neural Network ； Keyword Semantics ； Topic Detection ； Reader Emotion Analysis

參考文獻

[1] I. G. Israel, “Information explosion and university libraries : Current trends and strategies for intervention,” vol. 30, 12 2010.

[2] IBM Marketing Cloud, “10 key marketing trends for 2017,” 2017. Available at https://www-01.ibm.com/common/ssi/cgi-bin/ssialias?htmlfid=WRL12345USEN.

[3] F. Sebastiani, “Machine learning in automated text categorization,” ACM Comput. Surv., vol. 34, no. 1, pp. 1–47, 2002.

[4] J. P. Turian, L. Ratinov, and Y. Bengio, “Word representations: A simple and general method for semi-supervised learning,” in ACL 2010, Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, July 11-16, 2010, Uppsala, Sweden, pp. 384–394, 2010.

[5] R. E. Bellman, Dynamic programming. Princeton University Press, 1957. ISBN 978-0-691-07951-6.