探勘顧客對於產品新功能之情感變化

Abstract

廠商通常會在新產品中加入較優越、較新穎的功能去吸引消費者；但是產品新功能卻不一定能滿足消費者的需求。因此，我們提出一個架構去捕捉消費者對於產品新功能的評價，以及這些評價隨時間的變化。我們的架構包含四個階段：第一階段，我們找出產品新功能並產生對應的種子字詞。第二階段，我們修改隱含狄利克雷分布模型，以蒐集出現在評論中且和新功能相關的字詞。第三階段，我們用蒐集到的特徵詞做情感分析，並且針對每一個產品新功能，依照時間畫出情感趨勢圖。最後，我們透過特徵詞以及情感分析的協助，針對每個產品新功能，產生簡單扼要的摘要。實驗結果顯示，我們提出的方法能夠產生具有辨別力的主題，以及可以找到真正和特徵相關的字詞，且情感趨勢圖和新功能的摘要，可提供消費者及廠商豐富的資訊，以及管理上的見解及應用。

Manufacturers often introduce some highlighted features of their new products to attract consumers. However, highlighted features may or may not fulfill consumers’ needs. Therefore, we propose a framework to capture consumers’ attitude toward highlighted features and how their feedback changes with time. The proposed framework contains four phases. First, we preprocess online consumer reviews, find the highlighted features and generate seed words for each highlighted feature. Second, we modify the Latent Dirichlet Allocation Model (LDA) to iteratively collect the feature words relevant to each highlighted feature from consumer reviews. Third, for each highlighted feature, we perform sentiment analysis of the feature in each time period by using the feature words collected, and then visualize its sentiment tendency graph. Finally, we produce summaries for each highlighted feature and analyze the results obtained. The experiment results show our proposed framework can generate discriminative topics and capture the majority of feature words for each highlighted feature. Also, it can generate useful sentiment tendency graph and summary, which can provide quick references for consumers and valuable managerial insights for manufacturers.