Time: 10:50 AM - 12:20 AM

Place: Building 2, 4th floor, Meeting Room 1, Tama Campus, Chuo University
  中央大学・多摩キャンパス・2号館4階・研究所会議室1

Chair
Hiroshi Kumakura (Chuo University)

Kosei Shioura (Chuo University)

Title

A Good Evaluation Causes a Bad Evaluation: A Mechanism of Dynamic Change in Word-of-Mouth

Abstract
The primary objective of this study is to model and verify the dynamic change mechanisms of word-of-mouth (WOM) evaluations, with a particular focus on acquisition bias as a key factor. Acquisition bias refers to the phenomenon whereby average evaluations are systematically distorted due to the tendency of consumers with favorable expectations to voluntarily select products and write reviews. Among various temporal dynamics, this study also focuses on the distinctive phenomenon known as undershooting dynamics. This intriguing phenomenon is characterized by an initially high average rating that gradually declines and eventually rises again.

The study proposes a model that categorizes consumers' purchase decisions into search attributes (e.g., performers, prior knowledge before purchase) and experience attributes (e.g., story, post-purchase knowledge). The core element of the model is mismatch cost, which quantifies the loss of utility when a product does not match a consumer's specific preferences. This cost is influenced by whether the product is a niche product or a mainstream product. 

Purchase decisions depend on an important trade-off: expectations of quality and anxiety about mismatch. This dynamic is captured by a parameter, which is formulated based on the correlation between preferences for two attributes and the correlation between exploration attributes and consumers' sensitivity to mismatch. The sign of the parameter determines the purchase threshold, influencing whether consumers with strong preferences (fans) are more or less likely to purchase. This non-random selection of purchasers is the fundamental cause of acquisition bias.

Furthermore, the model establishes a dynamic feedback loop. The cumulative average rating and its variation from the previous period directly influence consumer beliefs in the next period. Specifically, past average ratings update prior quality expectations, while rating variations adjust the sensitivity of the market to mismatches. This continuous update process drives the evolution of WOM ratings over time.

Keywords

Acquisition Bias, Undershooting Dynamics, Search Attributes, Experience Attributes, Mismatch Cost,  Undershooting Dynamics

Discussant
Desmond Lo (Santa Clara University)

Minjung Kwon (Syracuse University)

Title

A Structural Life Course Model of Dynamic Role Selection on an Open Q&A Platform

Abstract
This study expands the scope of two-sided market literature by exploring indirect network effects that arise from users’ transitions from one side of the networks to the other as opposed to arising from the interaction between distinct user groups. Utilizing unstructured text data from an open Q&A platform, we employ topic modeling to identify latent themes within a collection of posts. On this platform, users seamlessly shift roles as question askers and answer givers while engaging with the themes. We examine the dynamics of distinct topics discussed on the platform and report stylized data patterns that suggest the platform internalizes individual users’ knowledge acquired on the platform by deploying them as answer givers. We develop a structural model of two-sided markets where users endogenously choose to participate as answerers based on their prior learning experiences as askers. Counterfactual experiments shed light on users’ dynamic role selections that promote the internalization of knowledge shared on open Q&A platforms and uncover the platforms’ potential for growth through enhanced user interaction. 

Keywords

Q&A Platforms, Two-Sided Markets, Latent Dirichlet Allocation, User-Generated Contents 

発表は日本語です！

Time: 12:30 - 13:20

Place: Building 2, 4th floor, Meeting Room 1, Tama Campus, Chuo University
  中央大学・多摩キャンパス・2号館4階・研究所会議室1

Chair
Kyung Tae Lee (Chuo University)

Fumiko Kano Glückstad (Copenhagen Business School)

Title

大規模言語モデルを用いた消費者研究

Abstract
近年の人工知能技術の発展に伴い、ソーシャルメディア上の大規模な質的データを使って、消費者を心理的に基づいて分類し、各消費者セグメントの態度や行動を探ることが比較的用意に行えるようになった。このトークでは、２つの事例を紹介する。

１つ目は、自然言語処理技術に焦点をあててTourism Management誌に掲載された我々の論文(Hardt & Glückstad, 2024)での事例を紹介する．この研究ではReddit 英語版の旅行に関するサブレディットにて3つの期間（1.コロナ危機前１、2.コロナ危機前２、3.コロナ危機中）に投稿された約100万件の投稿から3期間全てに投稿した著者を抽出し、これらの著者が第1期間中に投稿した文章を使ってWord Embeddingという手法を用いて心理的属性に基づく分類を行った。この心理的属性に基づく分類ついては、シュヴァルツの基本価値観理論に基づき保守的及び開放的な属性を記述する単語ベクトルを定義して、各著者の投稿と各単語ベクトル間の意味的類似度をもとに著者を分類した。次に、各著者が第2・第3期間中に投稿した内容をもとに、各セグメントの旅行に関する感情と危機感意識など、コロナ危機前とコロナ危機中で比較して、様々な考察を行った。この事例において、大規模データの著者を心理学的属性に基き分類することで将来に著者が発する態度や行動を予測できること、また、心理学等の理論を応用することで大規模データから得られる知見が深掘りできることが明らかになった。

２つ目は、前述の事例のように心理学的属性の基づき大規模データの著者を分類することで得られる知見が役に立つということに焦点をあて、最新の大規模言語モデルGPT4oを使っても前例と同じような知見が得られるかどうかについての実験を行った。２つ目の事例では、テーマとして人間の美容価値観理論（Glückstad et al., 2025）で定義された３つのカテゴリー（他者優越性、社会調和性、個人尊重性）のキーワードと定義文、更にはFew-Shot-Learningで人間による分類例と分類根拠で指示を与えることで、Reddit上の美に関するサブレディットの著者１００名を分類し、人間の分類との精度の比較を行った。人間の分類に最も近い指示モデルを定義した上で、２万人全ての著者を分類し、各セグメントごとに、最も重要な美容に関する心配事、美容に関するアプローチ、及び、好ましいブランドが何なのかをGPT4oに抽出させた。GPT4oに指示を出す際に定義した心理学理論定義とその応用の重要性と結果の考察について議論する。

Related Work(s)

Daniel Hardt; Fumiko Kano Glückstad (2024) A Social Media Analysis of Travel Preferences and Attitudes, Before and During Covid-19 In: Tourism Management, Vol. 100, 2.2024.

Fumiko Kano Glückstad; Hiromi Kobayashi; Daniel Seddig; Eldad Davidov; Rie Nakamura (2025) Personal Beauty Values : Development and Validation of a Multidimensional Measurement Scale. In: Journal of Consumer Behaviour, Vol. 24, No. 1, 1.2025, p. 282-303.

Time: After the seminar (around 13:20)

Place: Building 2, 4th floor, Meeting Room 2, Tama Campus, Chuo University
  中央大学・多摩キャンパス・2号館4階・研究所会議室2

Registration

Please kindly register to join the small social until June 23 (Mon).

Around 500 JPY will be charged for participation (excluding speakers).

Link (Please come to join the social, even if you do not attend the seminar!)