| 摘要: | 預測運動賽事結果主要是使用球隊比賽數據，較少考量其他向度。本研究預測NBA賽事結果，除了使用球隊數據，也加入該球季當場比賽被登錄球員的薪資，作為個別球 員表現反映其職業運動的專業程度與貢獻，以及社群媒體情緒分數，作為大眾對於整 體球隊評價、表現讚賞與認同的程度。社群媒體情緒是透過Python撰寫爬蟲程式抓取 使用者在Twitter上的推文，使用自然語言處理技術將其轉換成情緒分數。預測方法採用機器學習中的邏輯斯迴歸、支持向量機、決策樹、隨機森林、以及梯度上升決策樹。結果發現若只用球隊數據，最佳為邏輯斯迴歸，有67.6%的準確率。在加入球員薪資與社群媒體情緒分數後，最佳為梯度上升決策樹模型，可達 69.4%的準確率。依研究結果提出討論、管理意涵、研究限制與未來發展方向可供後續研究參考。;To predict an athletic game outcome, most are based on game statistics, and seldom consider other dimensions. Besides game data, player salary that reflects the individual’s contribution to the professional sport, as well as social media sentiment scores that reflects the general public evaluation, approval, and identification with the team, are proposed to predict an NBA game outcome. Postings called tweets on Twitter were first scraped using a Python crawler, followed by natural language processing to convert data into sentiment scores. The prediction methods used in machine learning includes logistic regression, support vector machine, decision tree, random forest, and gradient boosting decision tree. Results show that to use game data only, the highest is logistic regression, with 67.6% accuracy. After adding player salary and social media sentiment scores, the highest becomes gradient boosting decision tree, with accuracy increased to 69.4%. Results based discussion, management implications, limitations and directions for future research are included for future research reference. |
