基於知識圖譜之異質性網路財金新聞推薦

隨著網路的蓬勃發展，近年來推薦系統在各領域的應用已相當廣泛，人們收集各式資訊的方式不斷轉換、電子媒體的傳播速度迅速，投資人獲得股市消息也變得快速且容易。但要如何在龐大的資訊中篩選自身所需的資料，以作為投資決策的參考，仍是一個具有挑戰的議題。本研究的目的為改善推薦系統在財金新聞的推薦效果。本研究所提出的方法首先藉由個股、關鍵字詞，以及上下游資訊建立社群網絡，再以圖嵌入方法取得網絡中每個節點的向量，並且計算節點間的相似度，最後建立推薦模型進行新聞推薦。本研究藉由實際推薦新聞給受訪者，測試本研究所提出的推薦系統。透過受訪者的回饋，可發現本研究所提出的系統可成功推薦使用者感興趣的新聞。即使無法取得使用者過去的歷史紀錄，仍可藉由使用者自行挑選的股票，以及社群網絡中的關聯，推薦使用者感興趣的相關新聞。然而，本研究目前所納入的資料來源有限，也尚未將新聞的時效性納入考量，未來研究仍有許多可改善的空間。

關鍵字

知識圖譜；社群媒體挖掘；圖嵌入；推薦系統

並列摘要

With the vigorous development of the Internet, in recent years, the applications of recommendation systems in various fields have been very widespread. The way people collect all kinds of information is constantly changing, and the spread of electronic media is rapid. It has become fast and easy for investors to obtain stock market news. However, how to extract useful data from a huge amount of information is still a challenging issue. This research aims to improve the recommendation effect of financial news recommending. The proposed method first builds a network with individual stocks, keywords, and supply chain information. The graph embedding method is used to obtain a vector for each node in the network, and the similarities between nodes are then calculated. Finally, a model is constructed for recommending financial news. The proposed recommendation system was tested by recommending news to interviewees. According to the feedback of interviewees, the proposed system can generate recommendations that meet the needs of users. Even there is no historical record, the system can produce successful recommendations with user-chosen stocks and relationships in the network. However, the data sources included in this study are limited, and the timeliness of news has not been considered. The proposed method can still be improved by some future works.