No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
The Enrollment Effect: A Study of Amazon's Vine Program
Abstract
Do rewards from retailers such as free products and recognition in the form of status badges 1 influence the recipient’s behavior? We present a novel application of natural language processing to detect differences in consumer behavior due to such rewards. Specifically, we investigate the “Enrollment” effect, i.e. whether receiving products for free affect how consumer reviews are written. Using data from Amazon’s Vine program, we conduct a detailed analysis to detect stylistic differences in product reviews written by reviewers before and after enrollment in the Vine program. Our analysis suggests that the “Enrollment” effect exists. Further, we are able to characterize the effect on syntactic and semantic dimensions. This work has implications for researchers, firms and consumer advocates studying the influence of user-generated content as these changes in style could potentially influence consumer decisions.
Online reviews increase consumer visits, increase the time spent on the website, and create a sense of community among the frequent shoppers. Because of the importance of online reviews, online retailers such as Amazon.com and eOpinions provide detailed guidelines for writing reviews. However, though these guidelines provide instructions on how to write reviews, reviewers are not provided instructions for writing product-specific reviews. As a result, poorly-written reviews are abound and a customer may need to scroll through a large number of reviews, which could be up to 6000 pixels down from the top of the page, in order to find helpful information about a product (Porter, 2010). Thus, there is a need to train reviewers to write better reviews, which could in turn better serve customers, vendors, and online e-stores. In this Thesis, we propose a review recommendation framework to train reviewers to better write about their experiences with a product by leveraging the behaviors of expert reviewers who are good at writing helpful reviews. First, we use clustering to model reviewers into different classes that reflect different skill levels to write a quality review such as expert, novice, etc. Through temporal analysis of reviewer behavior, we have found that reviewers evolve over time, with their reviews becoming better or worse in quality and more or less in quantity. We also investigate how reviews are valued differently across different product categories. Through machine learning-based classification techniques, we have found that, for products associated with prevention consumption goal, longer reviews are perceived to be more helpful; and, for products associated with promotion consumption goal, positive reviews are more helpful than negative ones. In this Thesis, our proposed review recommendation framework is aimed to help a novice or conscientious reviewer become an expert reviewer. Our assumption is that a reviewer will reach the highest level of expertise by learning from the experiences of his or her closest experts who have a similar evolutionary pattern to that of the reviewer who is being trained. In order to provide assistance with intermediate steps for the reviewer to grow from his or her current state to the highest level of expertise, we want to recommend the positive actions—that are not too far out of reach of the reviewer—and discourage the negative actions—that are within reach of the reviewer—of the reviewer’s closest experts. Recommendations are personalized to fit the expertise level of reviewers, their evolution trend, and product category. Using the proposed review recommendation system framework we have found that for a random reviewer, at least 80% of the reviews posted by closest experts were of higher quality than that of the novice reviewer. This is verified in a dataset of 2.3 million reviewers, whose reviews cover products from nine different product categories such as Books, Electronics, Cellphones and accessories, Grocery and gourmet food, Office product, Health and personal care, Baby, Beauty, and Pet supplies. Advisor: Leen-Kiat Soh
We study the relative importance of online word of mouth and advertising on firm performance over time since product introduction. The current research separates the volume of consumer-generated online word of mouth (OWOM) from its valence, which has three dimensions-attribute, emotion, and recommendation oriented. Firm-initiated advertising content is also classified as attribute or emotion advertising. We also shed light on the role played by advertising content on generating the different types of OWOM conversations. We use a dynamic hierarchical linear model (DHLM) for our analysis. The proposed model is compared with a dynamic linear model, vector autoregressive/system of equations model, and a generalized Bass model. Our estimation accounts for potential endogeneity in the key measures. Among the different OWOM measures, only the valence of recommendation OWOM is found to have a direct impact on sales; i.e., not all OWOM is the same. This impact increases over time. In contrast, the impact of attribute advertising and emotion advertising decreases over time. Also, consistent with prior research, we observe that rational messages (i.e., attributeoriented advertising) wears out a bit faster than emotion-oriented advertising. Moreover, the volume of OWOM does not have a significant impact on sales. This suggests that, in our data, "what people say" is more important than "how much people say." Next, we find that recommendation OWOM valence is driven primarily by the valence of attribute OWOM when the product is new and driven by the valence of emotion OWOM when the product is more mature. Our brand-level results help us classify brands as consumer driven or firm driven, depending on the relative importance of the OWOM and advertising measures, respectively.
We present an approach to automatically recover hidden attributes of scientific articles, such as whether the author is a native English speaker, whether the author is a male or a female, and whether the paper was published in a conference or workshop proceedings. We train classifiers to predict these attributes in computational linguistics papers. The classifiers perform well in this challenging domain, identifying non-native writing with 95% accuracy (over a baseline of 67%). We show the benefits of using syntactic features in stylometry; syntax leads to significant improvements over bag-of-words models on all three tasks, achieving 10% to 25% relative error reduction. We give a detailed analysis of which words and syntax most predict a particular attribute, and we show a strong correlation between our predictions and a paper's number of citations.
Many consumer price promotions (including new product launches) offer a product for free or for a low, discounted price along with a required purchase. This research demonstrates that consumers’ willingness to pay for the product after the promotion is retracted is higher when it was offered for free than when it was offered at a low, discounted price. The underlying reasoning is that the price of the product on promotion is used as a natural anchor for value estimation. However, when the product is offered for free (i.e., zero price), consumers are less likely to consider the value of the product and are influenced by anchors such as the price of the focal purchase. In contrast to some prior findings, a free offer does not devalue the product at all and, at a minimum, devalues the product less than if it were offered for a low, discounted price.
Much of the writing styles recognized in rhetorical and composition theories involve deep syntactic elements. However, most previous research for computational stylometric analysis has relied on shallow lexico-syntactic patterns. Some very recent work has shown that PCFG models can detect distributional difference in syntactic styles, but without offering much insights into exactly what constitute salient stylistic elements in sentence structure characterizing each authorship. In this paper, we present a comprehensive exploration of syntactic elements in writing styles, with particular emphasis on interpretable characterization of stylistic elements. We present analytic insights with respect to the authorship attribution task in two different domains.
Existing computer programs that measure readability are based largely upon subroutines which estimate number of syllables, usually by counting vowels. The shortcoming in estimating syllables is that it necessitates keypunching the prose into the computer. There is no need to estimate syllables since word length in letters is a better predictor of readability than word length in syllables. Therefore, a new readability formula was computed that has for its predictors letters per 100 words and sentences per 100 words. Both predictors can be counted by an optical scanning device, and thus the formula makes it economically feasible for an organization (e.g., the US Office of Education) to calibrate the readability of all textbooks for the public school system. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
We introduce the author-topic model, a generative model for documents that
extends Latent Dirichlet Allocation (LDA; Blei, Ng, & Jordan, 2003) to include
authorship information. Each author is associated with a multinomial
distribution over topics and each topic is associated with a multinomial
distribution over words. A document with multiple authors is modeled as a
distribution over topics that is a mixture of the distributions associated with
the authors. We apply the model to a collection of 1,700 NIPS conference papers
and 160,000 CiteSeer abstracts. Exact inference is intractable for these
datasets and we use Gibbs sampling to estimate the topic and author
distributions. We compare the performance with two other generative models for
documents, which are special cases of the author-topic model: LDA (a topic
model) and a simple author model in which each author is associated with a
distribution over words rather than a distribution over topics. We show topics
recovered by the author-topic model, and demonstrate applications to computing
similarity between authors and entropy of author output.
We are in the midst of a technological revolution whereby, for the first time, researchers can link daily word use to a broad array of real-world behaviors. This article reviews several computerized text analysis methods and describes how Linguistic Inquiry and Word Count (LIWC) was created and validated. LIWC is a transparent text analysis program that counts words in psychologically meaningful categories. Empirical results using LIWC demonstrate its ability to detect meaning in a wide variety of experimental settings, including to show attentional focus, emotionality, social relationships, thinking styles, and individual differences.
The authors present a model of free sample effects and evidence from two field experiments on free samples. The model incorporates three potential effects of free samples on sales: (1) an acceleration effect, whereby consumers begin repeat purchasing of the sampled brand earlier than they otherwise would; (2) a cannibalization effect, which reduces the number of paid trial purchases of the brand; and (3) an expansion effect, which induces purchasing by consumers who would not consider buying the brand without a free sample. The empirical findings suggest that, unlike other consumer promotions such as coupons, free samples can produce measurable long-term effects on sales that can be observed as much as 12 months after the promotion. The data also show that the effectiveness of free sample promotions can vary widely, even between brands in the same product category.
Application of the model to the data from the two experiments reveals that the magnitude of acceleration, cannibalization, and expansion effects varies substantially across the two free sample promotions. These and other findings suggest that the model can be a useful tool for obtaining insights into the nature of free sample promotions.
LIBLINEAR is an open source library for large-scale linear classification. It supports logistic regres- sion and linear support vector machines. We provide easy-to-use command-line tools and library calls for users and developers. Comprehensive documents are available for both beginners and advanced users. Experiments demonstrate that LIBLINEAR is very efficient on large sparse data sets.
Understanding social interaction within groups is key to analyzing online communities. Most current work focuses on structural properties: who talks to whom, and how such interactions form larger network structures. The interactions themselves, however, generally take place in the form of natural language --- either spoken or written --- and one could reasonably suppose that signals manifested in language might also provide information about roles, status, and other aspects of the group's dynamics. To date, however, finding domain-independent language-based signals has been a challenge.
Here, we show that in group discussions, power differentials between participants are subtly revealed by how much one individual immediately echoes the linguistic style of the person they are responding to. Starting from this observation, we propose an analysis framework based on linguistic coordination that can be used to shed light on power relationships and that works consistently across multiple types of power --- including a more "static" form of power based on status differences, and a more "situational" form of power in which one individual experiences a type of dependence on another. Using this framework, we study how conversational behavior can reveal power relationships in two very different settings: discussions among Wikipedians and arguments before the U. S. Supreme Court.
Consumers increasingly rate, review and research products online.
Consequently, websites containing consumer reviews are becoming targets of
opinion spam. While recent work has focused primarily on manually identifiable
instances of opinion spam, in this work we study deceptive opinion
spam---fictitious opinions that have been deliberately written to sound
authentic. Integrating work from psychology and computational linguistics, we
develop and compare three approaches to detecting deceptive opinion spam, and
ultimately develop a classifier that is nearly 90% accurate on our
gold-standard opinion spam dataset. Based on feature analysis of our learned
models, we additionally make several theoretical contributions, including
revealing a relationship between deceptive opinions and imaginative writing.
We demonstrate that an unlexicalized PCFG can parse much more accurately than previously shown, by making use of simple, linguistically motivated state splits, which break down false independence assumptions latent in a vanilla treebank grammar.
Predicting the success of literary works is a curious question among publishers and aspiring writers alike. We examine the quantitative connection, if any, between writing style and successful literature. Based on novels over several different genres, we probe the predictive power of statistical stylometry in discriminating successful literary works, and identify characteristic stylistic elements that are more prominent in successful writings. Our study reports for the first time that statistical stylom-etry can be surprisingly effective in discriminating highly successful literature from less successful counterpart, achieving accuracy up to 84%. Closer analyses lead to several new insights into characteristics of the writing style in successful literature, including findings that are contrary to the conventional wisdom with respect to good writing style and readability.
In this research, the authors show that sampling a consumption cue (e.g., a flavored beverage) that is high in incentive value (i.e., tastes good) not only enhances subsequent consumption of other similar consumption cues (e.g., Pepsi) but also prompts people to seek anything rewarding (e.g., chocolate, a massage). The authors propose that this phenomenon, which they term "reverse alliesthesia," occurs as a result of the activation of a general motivational state by cues that are high in incentive value. They test this proposition by showing that the effect is (1) attenuated if the activated motivational state is satiated before the subsequent consumption task and (2) stronger for people who are high rather than low on the behavioral activation system scale. Finally, the authors extend their core findings to nongustatory cues by showing that (un)pleasant odors can also lead to an increase (decrease) in subsequent reward-seeking behaviors.
When faced with a choice of selecting one of several available products (or possibly buying nothing), according to standard theoretical perspectives, people will choose the option with the highest cost–benefit difference. However, we propose that decisions about free (zero price) products differ, in that people do not simply subtract costs from benefits but instead they perceive the benefits associated with free products as higher.
We test this proposal by contrasting demand for two products across conditions that maintain the price difference between the goods, but vary the prices such that the cheaper good in the set is priced at either a low positive or zero price. In contrast with a standard cost–benefit perspective, in the zero-price condition, dramatically more participants choose the cheaper option, whereas dramatically fewer participants choose the more expensive option. Thus, people appear to act as if zero pricing of a good not only decreases its cost, but also adds to its benefits. After documenting this basic effect, we propose and test several psychological antecedents of the effect, including social norms, mapping difficulty, and affect. Affect emerges as the most likely account for the effect.
This paper describes a corpus annotation project to study issues in
the manual annotation of opinions, emotions, sentiments, speculations,
evaluations and other private states in language. The resulting
corpus annotation scheme is described, as well as examples of its
use. In addition, the manual annotation process and the results
of an inter-annotator agreement study on a 10,000-sentence corpus
of articles drawn from the world press are presented.
Three readability formulas were recalculated to be more suitable for Navy use. The three formulas are the Automated Readability Index (ARI), Fog Count, and Flesch Reading Ease Formula. They were derived from test results of 531 Navy enlisted personnel enrolled in four technical training schools at two Navy bases: Naval Air Station at Memphis, and Great Lakes Navy Training Center. Personnel were tested for their reading comprehension level according to the comprehension section of the Gates-McGinitie reading test. At the same time, they were tested for their comprehension of 18 passages taken from Rate Training Manuals. Acknowledgments -- Summary -- Introduction -- Method -- Results -- Discussion and Conclusions -- References -- Appendix A (Eighteen Test Passages Taken From Rate Training Manuals) -- Appendix B (Instructions for Recalculated Formulas)
The creation of online consumer communities to provide product reviews and advice has been touted as an important, albeit somewhat expensive component of Internet retail strategies. In this paper, we characterize reviewer behavior at two popular Internet sites and examine the effect of consumer reviews on firms' sales. We use publicly available data from the two leading online booksellers, Amazon.com and BarnesandNoble.com, to construct measures of each firm's sales of individual books. We also gather extensive consumer review data at the two sites. First, we characterize the reviewer behavior on the two sites such as the distribution of the number of ratings and the valence and length of ratings, as well as ratings across different subject categories. Second, we measure the effect of individual reviews on the relative shares of books across the two sites. We argue that our methodology of comparing the sales and reviews of a given book across Internet retailers allows us to improve on the existing literature by better capturing a causal relationship between word of mouth (reviews) and sales since we are able to difference out factors that affect the sales and word of mouth of both retailers, such as the book's quality. We examine the incremental sales effects of having reviews for a particular book versus not having reviews and also the differential sales effects of positive and negative reviews. Our large database of books also allows us to control for other important confounding factors such as differences across the sites in prices and shipping times.
This paper develops a theoretical learning model of text classification for Support Vector Machines (SVMs). It connects the statistical properties of text-classification tasks with the generalization performance of a SVM in a quantitative way. Unlike conventional approaches to learning text classifiers, which rely primarily on empirical evidence, this model explains why and when SVMs perform well for text classification. In particular, it addresses the following questions: Why can support vector machines handle the large feature spaces in text classification effectively? How is this related to the statistical properties of text? What are sufficient conditions for applying SVMs to text-classification problems successfully?
- M David
- Blei
- Y Andrew
- Ng
- I Michael
- Jordan
David M Blei, Andrew Y Ng, and Michael I Jordan. Latent dirichlet allocation. the Journal of
machine Learning research, 3:993–1022, 2003.
- M David
- Blei
- Y Andrew
- Michael I Jordan Ng
David M Blei, Andrew Y Ng, and Michael I Jordan. Latent dirichlet allocation. the Journal of
machine Learning research, 3:993-1022, 2003.
A statistical learning learning model of text classification for support vector machines
- Thorsten Joachims
Thorsten Joachims. A statistical learning learning model of text classification for support vector machines. In Proceedings of the 24th annual international ACM SIGIR conference on
Research and development in information retrieval, pages 128-136. ACM, 2001.