JCJing Cao

Jing Cao

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Jing Cao has a total of 32 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Digital Transformation of Wine Opinion: From Critics to Peer Reviews Analysis

    by Magalie Dubois, Michał Folwarczny, Jing Cao, Þóranna Bender


  • Open Access

    Interpretable Word-Level Context-Based Sentiment Analysis

    by Chenyu Yang, Eric Larson, Jing Cao

    We propose a fine-grained attention-based multiple instance classification (FAMIC) model for interpretable word-level sentiment analysis (SA) using only document-level sentiment labels. By operating at the word level, FAMIC enhances interpretability while maintaining competitive performance in document-level classification. The model generates interpretable outputs such as contextual weighting, word neutrality, and negation cues, offering insights into how context shapes sentiment and how the model arrives at its predictions. FAMIC is built on a straightforward yet effective architecture that combines a multiple instance classification framework with self-attention and positionally encoded self-attention blocks. This design enables the model to capture both local and global contextual dependencies, supporting nuanced sentiment interpretation. We evaluate FAMIC on two sentiment classification datasets and provide an extensive analysis of its interpretability and performance.


  • Open Access

    Comparison of language models for wine sentiment analysis

    by Chenyu Yang, Jing Cao

    This study presents a comparative evaluation of sentiment analysis models applied to a large corpus of expert wine reviews from Wine Spectator, with the goal of classifying reviews into binary sentiment categories based on expert ratings. We assess six models: logistic regression, XGBoost, LSTM, BERT, the interpretable Attention-based Multiple Instance Classification (AMIC) model, and the generative language model LLAMA 3.1, highlighting their differences in accuracy, interpretability, and computational efficiency. While LLAMA 3.1 achieves the highest accuracy, its marginal improvement over AMIC and BERT comes at a significantly higher computational cost. Notably, AMIC matches the performance of pretrained large language models while offering superior interpretability, making it particularly effective for domain-specific tasks such as wine sentiment analysis. Through qualitative analysis of sentiment-bearing words, we demonstrate AMIC’s ability to uncover nuanced, context-dependent language patterns unique to wine reviews. These findings challenge the assumption of generative models’ universal superiority and underscore the importance of aligning model selection with domain-specific requirements, especially in applications where transparency and linguistic nuance are critical.


  • Open Access

    A study on why foreign-born East Asians but not US-born East Asians are underrepresented in leadership attainment in the U.S.

    by Jing Cao, Song Zhang

    A recent study investigated the impact of culture of Asian groups on leadership attainment in the U.S. It revealed that East Asians (EAs) are less likely than South Asians (SAs) and white people (WP) to attain leadership positions, and SAs may even surpass WP in leadership attainment. The study explained that the underrepresentation of EAs in leadership positions in the U.S. (the so called bamboo ceiling) is partly because EAs communicate less assertively. Specifically, EA cultures value collectivism (e.g., humility and harmony), which are at odds with western cultures that value individualism (e.g., extraversion and assertiveness), whereas SA cultures are congruent with western cultures. However, the study did not distinguish the different impact of home culture (i.e., EA cultures) and host culture (i.e., western cultures) on US-born EAs versus foreign-born EAs. We argue that for US-born EAs (i.e., second generation EAs), host culture plays a more important role than home culture in their growth and they may not be underrepresented in leadership attainment compared to WP. The bamboo ceiling effect is mostly demonstrated among foreign-born EAs (i.e., first generation EAs) who are shaped mainly under the home culture. We support the argument by conducting analysis on one of the datasets in the original study and a new dataset from Fortune’s 40-under-40 list. Our study suggests that when studying the underrepresentation of leadership attainment for EAs, US-born EAs and foreign-born EAs should not be aggregated in one category. Considering the ethnic EA group, the bamboo ceiling phenomenon may exist mainly among foreign-born EAs because US-born EAs, with a median age of 21.3, are much younger than the general American population, who may not be experienced enough to be considered for leadership positions in established large companies.


  • A Bayesian Latent Variable Model for Analysis of Empathic Accuracy

    by Linh H. Nghiem, Benjamin A. Tabak, Zachary Wallmark, Talha Alvi, Jing Cao

  • Open Access

    Using Neural Network Models for Wine Review Classification

    by Duwani Katumullage, Chenyu Yang, Jackson Barth, Jing Cao

    Wines are usually evaluated by wine experts and enthusiasts who give numeric ratings as well as text reviews. While most wine classification studies have been based on conventional statistical models using numeric variables, there has been very limited work on implementing neural network models using wine reviews. In this paper, we apply neural network models (CNN, BiLSTM, and BERT) to extract useful information from wine reviews and classify wines according to different rating classes. Using a large collection of wine reviews from Wine Spectator, the study shows that BERT, a neural network framework recently developed by Google, has the best performance. In the two-class classification (90–100 and 80–89), BERT achieves an accuracy of 89.12%, followed by BiLSTM (88.69%) and CNN (88.02%). In the four-class classification (95–100, 90–94, 85–89, and 80–84), BERT yields an 81.57% accuracy, while the other two produce an 80% accuracy. The neural network models in the paper are independent of domain knowledge and thus can be easily extended to other kinds of text analysis. Expanding the limited work on wine text review classification studies, these models are up-to-date and provide valuable additions to wine data analysis. (JEL Classifications: C45, C88, D83)


  • Open Access

    Wine Review Descriptors as Quality Predictors: Evidence from Language Processing Techniques

    by Chenyu Yang, Jackson Barth, Duwani Katumullage, Jing Cao

    There is an ongoing debate on whether wine reviews provide meaningful information on wine properties and quality. However, few studies have been conducted aiming directly at comparing the utility of wine reviews and numeric measurements in wine data analysis. Based on data from close to 300,000 wines reviewed by Wine Spectator, we use logistic regression models to investigate whether wine reviews are useful in predicting a wine's quality classification. We group our sample into one of two binary quality brackets, wines with a critical rating of 90 or above and the other group with ratings of 89 or below. This binary outcome constitutes our dependent variable. The explanatory variables include different combinations of numerical covariates such as the price and age of wines and numerical representations of text reviews. By comparing the explanatory accuracy of the models, our results suggest that wine review descriptors are more accurate in predicting binary wine quality classifications than are various numerical covariates—including the wine's price. In the study, we include three different feature extraction methods in text analysis: latent Dirichlet allocation, term frequency-inverse document frequency, and Doc2Vec text embedding. We find that Doc2Vec is the best performing feature extraction method that produces the highest classification accuracy due to its capability of using contextual information from text documents. (JEL Classifications: C45, C88, D83)


  • Classification of Wines Using Principal Component Analysis

    by Jackson Barth, Duwani Katumullage, Chenyu Yang, Jing Cao

    Classification of wines with a large number of correlated covariates may lead to classification results that are difficult to interpret. In this study, we use a publicly available dataset on wines from three known cultivars, where there are 13 highly correlated variables measuring chemical compounds of wines. The goal is to produce an efficient classifier with straightforward interpretation to shed light on the important features of wines in the classification. To achieve the goal, we incorporate principal component analysis (PCA) in the k-nearest neighbor (kNN) classification to deal with the serious multicollinearity among the explanatory variables. PCA can identify the underlying dominant features and provide a more succinct and straightforward summary over the correlated covariates. The study shows that kNN combined with PCA yields a much simpler and interpretable classifier that has comparable performance with kNN based on all the 13 variables. The appropriate number of principal components is chosen to strike a balance between predictive accuracy and simplicity of interpretation. Our final classifier is based on only two principal components, which can be interpreted as the strength of taste and level of alcohol and fermentation in wines, respectively. (JEL Classifications: C10, Cl4, D83)


  • Bayesian functional enrichment analysis for the Reactome database

    by Jing Cao

    The first step in the analysis of high-throughput experiment results is often to identify genes or proteins with certain characteristics, such as genes being differentially expressed (DE). To gain more insights into the underlying biology, functional enrichment analysis is then conducted to provide functional interpretation for the identified genes or proteins. The hypergeometric P value has been widely used to investigate whether genes from predefined functional terms, e.g., Reactome, are enriched in the DE genes. The hypergeometric P value has several limitations: (1) computed independently for each term, thus neglecting biological dependence; (2) subject to a size constraint that leads to the tendency of selecting less-specific terms. In this paper, a Bayesian approach is proposed to overcome these limitations by incorporating the interconnected dependence structure of biological functions in the Reactome database through a CAR prior in a Bayesian hierarchical logistic model. The inference on functional enrichment is then based on posterior probabilities that are immune to the size constraint. This method can detect moderate but consistent enrichment signals and identify sets of closely related and biologically meaningful functional terms rather than isolated terms. The performance of the Bayesian method is demonstrated via a simulation study and a real data application.


  • Analyses of Wine-Tasting Data: A Tutorial

    by Ingram Olkin, Ying Lou, Lynne Stokes, Jing Cao

    The purpose of this paper is to provide a tutorial of data analysis methods for answering questions that arise in analyzing data from wine-tasting events: (i) measuring agreement of two judges and its extension to m judges; (ii) making comparisons of judges across years; (iii) comparing two wines; (iv) designing tasting procedures to reduce burden of multiple tastings; (v) ranking of judges; and (vi) assessing causes of disagreement. In each case we describe one or more analyses and make recommendations on the conditions of use for each. (JEL Classifications: C10, C12, C13, C59, C90)