社群媒體廣告成效預測模型

Abstract

廣告是商業活動中不可或缺的一環，商品行銷、品牌形象建立都需要由廣告來宣傳。而隨著網路的發達，社群媒體也蓬勃發展，社群媒體上的廣告具有即時性、不受時間地點限制、且可針對特定對象或特定目的做行銷，與傳統的廣告模式相比大大提升了投遞效益，因此透過社群媒體廣告來行銷儼然成為趨勢。然而在執行廣告前並無法預知廣告的成效，廣告主無法評估需要多少預算才能達到期望的目標，如：目標的曝光量或是應用程式下載量，預算過高或過低都可能造成不必要的浪費或是目標達成困難，無法讓預算達到最大的成效。在本篇論文中，我們針對Facebook上推廣粉絲專頁的廣告來建立預測成效模型，以利廣告主評估廣告預算，獲得最大的效益，其共分為四個階段。在第一個階段中，我們對粉絲專頁的數據做前處理，透過此階段我們可以除去極端值並取得標準化粉絲專頁的特徵數據，如：粉絲數量、貼文數量等。在第二階段中，我們利用K-Means分群法將相似的預測資料根據特徵數據分成集群。接著在第三階段，我們預測粉絲專頁粉絲價格(Cost per Fan, CPF)，針對每個集群建立了決策樹。根據預測結果，我們可以得到粉絲專頁CPF落在各個價格區間的機率。接著在第四階段，我們對決策樹得到的結果作拓展，得到最終結果，以提高命中率。根據模擬結果，驗證了此模型的預測結果有61%與實際結果吻合。透過此模型可預測粉絲專頁的CPF，提供給廣告主做預算分配的參考，以最少的預算達到最大的效益。

Advertising is a very important part in business activities and is needed for marketing and building brand image. By the developing of the internet, social media become popular. The advertisements on social media can be shown to target audience at any time and any place for certain objective. It is surly more effective than traditional advertisements and becomes a trend. However, the performance cannot be known before the campaign runs. Thus, it is hard for campaign managers to decide the budget. No matter the budget is too high or too low can cause waste of money. Therefore, we proposed a model to predict social media campaign performance. The model can predict the performance of fan page promoting campaigns on Facebook. It includes four phases. First, we preprocess the data by removing outliers and normalizing. Second, we group the data into several clusters according the characteristics with K-Means clustering. Third, we build decision trees for each cluster in order to predict the cost per fan (CPF). Finally, we expand the result we get from the decision trees and decide the final result. According to the experiments, the hit rate is 61%. With this model, we can provide the result to campaign owners and help them allocating the budget.