基於社群媒體深度學習與情感分析之個人化推薦方法之研究

The Study of Social Media, Deep Learning and Sentiment Analysis for the Methods of Personalized Recommendation

洪進國(HENDRY)

指導教授 ： 陳榮靜

朝陽科技大學/資訊學院/資訊管理系/博士(2018年)

摘要

社群媒體和web 2.0的發展使得使用者可以在社群網路上更有互動。在社群網網路中，使用者間的關係通常會透過使用者的發文和互動來驗證。而使用這些資料可以讓系統推薦特定的資訊給不同的使用者。然而，若使用者是首次使用社群網路，因為沒有歷史資料所以推薦系統就沒有辦法推廣資訊。在本研究我們設計了本體論結合社群網路。我們導入了以使用者和朋友間的資料為主的本體論。利用使用者的興趣和社群的影響力來解決推薦系統慢熱的情況。系統會先計算使用者間的相似度。接著會用使用者的喜好和演算法來推論出規則。本體論會隨著個人的本體論更新而更新。最新的本體論會被保留來增強下一次推薦系統的準確率。
利用類神經網路，深度學習在很多領域取得了突破。使用者的評論對於推案是系統是重要的，因為它包含著多種情緒的資訊，並且會影響到推薦的準確率。如何增加推薦系統內使用者評分預測的準確率是重要的。在本論文我們提出了使用深度學習來處理使用者的評論，並且預測使用者會給出的評分來給推薦系統。首先系統會使用情緒分析來產生作為輸入的特徵。接著會使用去雜訊來提高對使用者評分的分類。最後會用深度學習與情緒分析來產生訓練模型。研究結果顯示，比起傳統的方式，此系統有更高的準確率。

關鍵字

社群網路分析 ； 推薦系統 ； 社群影響 ； 全域影響推薦 ； 深度學習 ； 分類 ； 受限玻爾茲曼機

並列摘要

Social media and the development of web 2.0 encourage the user to participate more interactively in social networks. In the social network, the relationships may be identified by the user posts and interactions. Using this data, the system can make recommendations tailored to specific users. However, when the user is on the social network for the first time, the recommendation system cannot make recommendations, since the user has no history. In this paper, we design an ontology combined with social networks. We develop the ontology based on data from users and their friends. Using the user interest and community influences, we propose a system to solve the cold start problem in recommendation systems. The system calculates the similarity among users. Then, user preferences and a rule generating algorithm create the dynamic inference rule. The ontology is updated each time the content of the personal ontology is updated. The newest ontology will be retained to increase the accuracy the next time the recommendation system is executed.
Deep learning is one of the methodologies which are applied to neural networks and learning that have been applied in many fields and have achieved many breakthrough successes for many applications. User comments are important for recommender systems because they include various types of emotional information that may influence the correctness or precision of the recommendation. How to improve the accuracy of user rating from obtained feasible recommendations is important. In this paper, we propose a deep learning model to process user comments and to generate a feasible user rating for the recommendation. First, the system uses sentiment analysis to create a feature vector as the input nodes. Next, the system implemented a noise reduction in the dataset to improve the classification of user ratings. Finally, generate training model by a deep belief network and sentiment analysis (DBNSA). The experimental results indicated the system has better accuracy than traditional methods.

並列關鍵字

Social Media Analysis ； Recommendation System ； Social Influence ； Global Influence Recommendation ； Deep Learning ； Classification ； Restricted Boltzmann Machine

參考文獻

[1] J. Lee, "No. 1 Position in Google Gets 33% of Search Traffic [Study]," Chikita Research, 20 June 2013. [Online]. Available: https://searchenginewatch.com/sew/study/2276184/no-1-position-in-google-gets-33-of-search-traffic-study. [Accessed 30 September 2016].

[2] J. L. Herlocker, J. A. Konstan, L. G. Terveen and J. T. Riedl, "Evaluating collaborative filtering recommender systems," ACM Transactions on Information Systems (TOIS), vol. 22, no. 1, pp. 5-53, 2004.

[3] X. Qian, H. Feng, G. Zhao and T. Mei, "Personalized Recommendation Combining User Interest and Social Circle," IEEE Transactions on Knowledge and Data Engineering, vol. 26, no. 7, pp. 1763-1777, 2014.

[4] X. Qian, X. Liu, C. Zheng, Y. Du and X. Hou, "Tagging photos using users' vocabularies," Neurocomputing, vol. 111, pp. 144-153, 2013.

[5] J. Shen, H. Pang, M. Wang and S. Yan, "Modeling concept dynamics for large scale music search," in SIGIR '12 Proceedings of the 35th international ACM SIGIR conference on Research and development in information retrieval, Portland, Oregon, USA, 2012.