A Bi-Phase LSTM Architecture on Dynamic Social Network Prediction

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/86576

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/86576

Title:	A Bi-Phase LSTM Architecture on Dynamic Social Network Prediction
Authors:	蘇恆毅;Su, Heng-Yi
Contributors:	資訊管理學系
Keywords:	特徵擷取;長短期記憶網路;動態社交網路;Feature Extraction;Long Short-Term Memory;Dynamic Social Network
Date:	2021-07-20
Issue Date:	2021-12-07 12:59:26 (UTC+8)
Publisher:	國立中央大學
Abstract:	近年來社群網路非常流行，大部分的人都有參與社群網路，社群網路中藏有非常豐富的資訊，像是價值觀、興趣等。我們可以透過社群網路的互動關係來了解哪些使用者比較親近，進而推薦使用者未來可能的好友。然而由於時間的演變，以前可能是好友的友誼關係可能改變了，我們認為社交網路也是會演變的。因此我們利用 BP-LSTM auto-encoder 和 BP-LSTM predictor，將動態網路圖的特徵保留起來並用來預測下一個時間點的社交網路圖。我們在三個資料集中評估模型，並且與不同的模型比較預測結果，實驗結果顯示 BP-LSTM 在三個資料集中都有相當優秀的表現。最後我們對我們的模型進行參數調整，以達到最佳的預測結果。 ;Social networks have become very popular in recent years. Most people participate in social networks. The social network contains a wealth of information, such as values and interests. We can learn about those users who are close to each other through the interaction of social networks, and then recommend users who may be friends in the future. However, due to the evolution of time, the friendship that may have been friends may have changed, and we believe that social networks will also evolve. Therefore, we use BP-LSTM auto-encoder and BP-LSTM predictor to retain the features of the dynamic network graphs and use it to predict the social network graph at the next point in time. We evaluated the model in three datasets and compared the prediction results with different models. The experimental results showed that BP-LSTM performed quite well in the three data sets. Finally, we adjust the parameters of our model to achieve the best prediction results.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	53	View/Open

社群 sharing

Loading...