Date of Award
Doctor of Philosophy (PhD)
This thesis studies the use of firm and user-generated unstructured data (e.g., text and videos) for improving market research combining advances in text, audio and video processing with traditional economic modeling. The first chapter is joint work with K. Sudhir and Minkyung Kim. It addresses two significant challenges in using online text reviews to obtain fine-grained attribute level sentiment ratings. First, we develop a deep learning convolutional-LSTM hybrid model to account for language structure, in contrast to methods that rely on word frequency. The convolutional layer accounts for the spatial structure (adjacent word groups or phrases) and LSTM accounts for the sequential structure of language (sentiment distributed and modified across non-adjacent phrases). Second, we address the problem of missing attributes in text in constructing attribute sentiment scores---as reviewers write only about a subset of attributes and remain silent on others. We develop a model-based imputation strategy using a structural model of heterogeneous rating behavior. Using Yelp restaurant review data, we show superior accuracy in converting text to numerical attribute sentiment scores with our model. The structural model finds three reviewer segments with different motivations: status seeking, altruism/want voice, and need to vent/praise. Interestingly, our results show that reviewers write to inform and vent/praise, but not based on attribute importance. Our heterogeneous model-based imputation performs better than other common imputations; and importantly leads to managerially significant corrections in restaurant attribute ratings. The second essay, which is joint work with Aniko Oery and Joyee Deb is an information-theoretic model to study what causes selection in valence in user-generated reviews. The propensity of consumers to engage in word-of-mouth (WOM) differs after good versus bad experiences, which can result in positive or negative selection of user-generated reviews. We show how the strength of brand image (dispersion of consumer beliefs about quality) and the informativeness of good and bad experiences impacts selection of WOM in equilibrium. WOM is costly: Early adopters talk only if they can affect the receiver’s purchase. If the brand image is strong (consumer beliefs are homogeneous), only negative WOM can arise. With a weak brand image or heterogeneous beliefs, positive WOM can occur if positive experiences are sufficiently informative. Using data from Yelp.com, we show how strong brands (chain restaurants) systematically receive lower evaluations controlling for several restaurant and reviewer characteristics. The third essay which is joint work with K.Sudhir and Khai Chiong studies success factors of persuasive sales pitches from a multi-modal video dataset of buyer-seller interactions. A successful sales pitch is an outcome of both the content of the message as well as style of delivery. Moreover, unlike one-way interactions like speeches, sales pitches are a two-way process and hence interactivity as well as matching the wavelength of the buyer are also critical to the success of the pitch. We extract four groups of features: content-related, style-related, interactivity and similarity in order to build a predictive model of sales pitch effectiveness.
Chakraborty, Ishita Sunity Kumar, "Three Essays on the Role of Unstructured Data in Marketing Research" (2021). Yale Graduate School of Arts and Sciences Dissertations. 24.