高聖倫

Sheng-LunKao

碩士論文（2018）

以圖文特徵為基之消費者購買決策型態辨識方法研發

Development of Method for Graphic and Text Features based Consumer Decision-Making Style Identification

關鍵字 Keywords

消費者購買決策型態、臉書、影像分析、機器學習、群眾外包

Consumer decision-making styles、Facebook、Image analysis、Machine learning、Crowdsourcing

摘要

當今企業競爭激烈的環境中，行銷策略擬定已成為企業成敗的重要關鍵因素，而企業是否瞭解消費者之購買決策型態是行銷策略所不可忽略。以往企業為了解消費者購買決策型態在大量問卷施測以及統計分析上費時費力。隨著網際網路發達與社群媒體普及，越來越多使用者會在社群媒體上留下文字、相片等資料，並且從中揭露了自身的資訊。目前最主要的社群網站臉書，每天有數以億計的內容被分享，這也提供了企業可以客觀地暸解消費者購買決策型態。

本研究目的在於使用青年臉書使用者之「文章」與「相片」作為資料來源，以主題與情感特徵為線索，發展一消費者購買決策型態辨識方法，以供企業分析消費者購買決策型態，進而訂定有效的行銷策略、提高競爭力。針對上述目的，本研究主要研究項目包括：(i)圖文特徵為基之購買決策型態辨識方法設計，(ii)圖文特徵擷取方法發展，(iii)以消費者購買決策型態分類方法發展，(iv)系統實作與驗證。

本研究使用Sproles and Kendall提出之消費者購買決策模型，來辨識臉書使用者的消費者購買決策型態，透過群眾外包之資料標註，運用影像分析技術分析相片內容、文字探勘技術分析文字內容、機器學習技術將使用者之消費者購買決策型態進行分類。

本研究提出以圖文特徵為基之消費者購買決策型態辨識方法之實驗結果，經5-fold cross validation，驗證本研究提出之辨識方法，顯示8種購買決策型態，分別有40%~63%之準確度，其中最高之準確度為第四種購買決策型態：娛樂-快樂主義導向，經本研究提出之文章、相片辨識方法，皆為近63%之準確度。而圖文整合之準確度相較僅用文章或相片的辨識方法略有提升。

Abstract

In today's competitive environment, the formulation of marketing strategy has become an important key factor in the success or failure of enterprises, and whether the enterprise is aware of consumer decision-making styles is that a marketing strategy should not be ignored.It is time-consuming and laborious that enterprises spent a lot on a large number of questionnaire surveys and statistical analysis to understand consumer decision-making styles in the past.With the development of Internet and the popularity of community media, more and more users will leave text, graphics and other data in the community media, and from which to expose their own information.At present, in the most important community website - Facebook (Wikipedia. 2016), there are hundreds of millions of content to be shared every day, which also provides an enterprise can objectively understand consumer decision-making styles.

The aim of this study is to use the text and graphics of young Facebook users as as the data source, with themes and emotional characteristics for clues, to develop a method of consumer decision-making style identificationa for the enterprise’s analysis of consumer decision-making style, and then the development of an effective marketing strategy to improve competitiveness.For the above purposes, the main research projects of this study include: (i) the design of method for graphic and text features based consumer decision-making style identification, (ii) the development of method for graphic and text feature extraction, (iii) the development of method for the classification of consumer decision-making style, (iv) the system’s implementation and validation.

In this study, we use the model of consumer decision-making styles proposed by Sproles and Kendall to identify the consumers' decision-making styles of Facebook users. In addition, through crowdsourcing’s data label, we use the image analysis technique to analyze the content of the graphics, the text mining technique to analyze the text content, and the machine learning technique to classify the consumer decision-making styles.

In this study, the experimental results of the method for graphic and text features based consumer decision-making style identification are presented. Through five-fold cross validation, the identification method proposed in this study have shown that eight kinds of decision-making styles had 40%~63% accuracy respectively. The highest accuracy of decision-making styles is the fourth style: recreational and hedonistic whose text and graphics’s identifications proposed by this study are nearly 63% of the accuracy. And the accuracy of the integration of graphic and text compared with only text or graphics identification method is slightly improved.