Corpus for Customer Purchase Behavior Prediction in Social Media

Corpus for Customer Purchase Behavior Prediction in Social Media Shigeyuki Sakaki*†, Francine Chen*, Mandy Korpusik*‡, and Yan-Ying Chen* *FX Palo Alto Laboratory, Inc., Palo Alto, California, USA †Fuji Xerox Co., Ltd., Yokohama-shi, Kanagawa, Japan ‡CSAIL Massachusetts Institute of Technology, Cambridge, Massachusetts, USA E-mail: sakaki.shigeyuki@fujixerox.co.jp, chen@fxpal.com, korpusik@mit.edu, yanying@fxpal.com Abstract Many people post about their daily life on social media. These posts may include information about the purchase activity of people, and insights useful to companies can be derived from them: e.g. profile information of a user who mentioned something about their product. As a further advanced analysis, we consider extracting users who are likely to buy a product from the set of users who mentioned that the product is attractive. In this paper, we report our methodology for building a corpus for Twitter user purchase behavior prediction. First, we collected Twitter users who posted a want phrase + product name: e.g. “want a Xperia” as candidate want users, and also candidate bought users in the same way. Then, we asked an annotator to judge whether a candidate user actually bought a product. We also annotated whether tweets randomly sampled from want/bought user timelines are relevant or not to purchase. In this annotation, 58% of want user tweets and 35% of bought user tweets were annotated as relevant. Our data indicate that information embedded in timeline tweets can be used to predict purchase behavior of tweeted products. Keywords: purchase behavior, corpus for machine learning, micro-blogging 1. Introduction Recently, many companies have tried obtaining insights into possible customers from Twitter. An example of such an application is inferring user profile information (Ikeda et al., 2013; Sakaki et al., 2014; Taniguchi et al., 2015), for use in marketing and in targeted advertising. Another application is to identify people who are likely to buy a company’s products. By identifying prospective buyers, companies can remove barriers to purchasing their products, such as informing customers about product features through advertisements, offering coupons, and introducing shop or sales people. In addition, companies can estimate future sales from the expected number of users. In Twitter, there is an abundance of tweets indicating that the owners of the tweets want something (e.g. “I want an iPhone”, “I plan to buy a Nexus”). Figure 1 shows an example of a portion of the tweet timeline of a Twitter user who wants an iPhone. This user posted two tweets (want tweets) indicating interest in an iPhone, and four tweets later, he posted a tweet announcing that he actually bought an iPhone. As in the example timeline, we expect that purchase behavior prediction is available by utilizing text information around want tweets. We created a tweet corpus for use in research on automatically predicting whether a Twitter user will buy a product that was mentioned in their past tweets. We automatically collected English tweets related to purchase behavior and then manually annotated whether a user purchased a product. In creating this corpus, there were a number of challenges, including defining methods for identifying …

Content truncated to 500 words