Lyle H. Ungar

Sharath Chandra Guntuku, Thomas Talhelm, Garrick Sherman et al.

We used natural language processing to analyze a billion words to study cultural differences on Weibo, one of China's largest social media platforms. We compared predictions from two common explanations about cultural differences in China (economic development and urban-rural differences) against the less-obvious legacy of rice versus wheat farming. Rice farmers had to coordinate shared irrigation networks and exchange labor to cope with higher labor requirements. In contrast, wheat relied on rainfall and required half as much labor. We test whether this legacy made southern China more interdependent. Across all word categories, rice explained twice as much variance as economic development and urbanization. Rice areas used more words reflecting tight social ties, holistic thought, and a cautious, prevention orientation. We then used Twitter data comparing prefectures in Japan, which largely replicated the results from China. This provides crucial evidence of the rice theory in a different nation, language, and platform.

Ilias Triantafyllopoulos, Renyi Qu, Salvatore Giorgi et al.

Ensuring safety and in-domain responses for Retrieval-Augmented Generation (RAG) systems is paramount in safety-critical applications, yet remains a significant challenge. To address this, we evaluate four methodologies for Out-Of-Domain (OOD) query detection: GPT-4o, regression-based, Principal Component Analysis (PCA)-based, and Neural Collapse (NC), to ensure the RAG system only responds to queries confined to the system's knowledge base. Specifically, our evaluation explores two novel dimensionality reduction and feature separation strategies: \textit{PCA}, where top components are selected using explained variance or OOD separability, and an adaptation of \textit{Neural Collapse Feature Separation}. We validate our approach on standard datasets (StackExchange and MSMARCO) and real-world applications (Substance Use and COVID-19), including tests against LLM-simulated and actual attacks on a COVID-19 vaccine chatbot. Through human and LLM-based evaluations of response correctness and relevance, we confirm that an external OOD detector is crucial for maintaining response relevance.

Salvatore Giorgi, Tingting Liu, Ankit Aich et al.

Large language models (LLMs) are increasingly being used in human-centered social scientific tasks, such as data annotation, synthetic data creation, and engaging in dialog. However, these tasks are highly subjective and dependent on human factors, such as one's environment, attitudes, beliefs, and lived experiences. Thus, it may be the case that employing LLMs (which do not have such human factors) in these tasks results in a lack of variation in data, failing to reflect the diversity of human experiences. In this paper, we examine the role of prompting LLMs with human-like personas and asking the models to answer as if they were a specific human. This is done explicitly, with exact demographics, political beliefs, and lived experiences, or implicitly via names prevalent in specific populations. The LLM personas are then evaluated via (1) subjective annotation task (e.g., detecting toxicity) and (2) a belief generation task, where both tasks are known to vary across human factors. We examine the impact of explicit vs. implicit personas and investigate which human factors LLMs recognize and respond to. Results show that explicit LLM personas show mixed results when reproducing known human biases, but generally fail to demonstrate implicit biases. We conclude that LLMs may capture the statistical patterns of how people speak, but are generally unable to model the complex interactions and subtleties of human perceptions, potentially limiting their effectiveness in social science applications.

Aadesh Salecha, Molly E. Ireland, Shashanka Subrahmanya et al.

As Large Language Models (LLMs) become widely used to model and simulate human behavior, understanding their biases becomes critical. We developed an experimental framework using Big Five personality surveys and uncovered a previously undetected social desirability bias in a wide range of LLMs. By systematically varying the number of questions LLMs were exposed to, we demonstrate their ability to infer when they are being evaluated. When personality evaluation is inferred, LLMs skew their scores towards the desirable ends of trait dimensions (i.e., increased extraversion, decreased neuroticism, etc). This bias exists in all tested models, including GPT-4/3.5, Claude 3, Llama 3, and PaLM-2. Bias levels appear to increase in more recent models, with GPT-4's survey responses changing by 1.20 (human) standard deviations and Llama 3's by 0.98 standard deviations-very large effects. This bias is robust to randomization of question order and paraphrasing. Reverse-coding all the questions decreases bias levels but does not eliminate them, suggesting that this effect cannot be attributed to acquiescence bias. Our findings reveal an emergent social desirability bias and suggest constraints on profiling LLMs with psychometric tests and on using LLMs as proxies for human participants.

Salvatore Giorgi, Shreya Havaldar, Farhan Ahmed et al.

We present metrics for evaluating dialog systems through a psychologically-grounded "human" lens in which conversational agents express a diversity of both states (e.g., emotion) and traits (e.g., personality), just as people do. We present five interpretable metrics from established psychology that are fundamental to human communication and relationships: emotional entropy, linguistic style and emotion matching, agreeableness, and empathy. These metrics can be applied (1) across dialogs and (2) on turns within dialogs. The psychological metrics are compared against seven state-of-the-art traditional metrics (e.g., BARTScore and BLEURT) on seven standard dialog system data sets. We also introduce a novel data set, the Three Bot Dialog Evaluation Corpus, which consists of annotated conversations from ChatGPT, GPT-3, and BlenderBot. We demonstrate that our proposed metrics offer novel information; they are uncorrelated with traditional metrics, can be used to meaningfully compare dialog systems, and lead to increased accuracy (beyond existing traditional metrics) in predicting crowd-sourced dialog judgements. The interpretability and unique signal of our psychological metrics make them a valuable tool for evaluating and improving dialog systems.

Roshan Santosh, H. Andrew Schwartz, Johannes C. Eichstaedt et al.

In this paper, we present an iterative graph-based approach for the detection of symptoms of COVID-19, the pathology of which seems to be evolving. More generally, the method can be applied to finding context-specific words and texts (e.g. symptom mentions) in large imbalanced corpora (e.g. all tweets mentioning #COVID-19). Given the novelty of COVID-19, we also test if the proposed approach generalizes to the problem of detecting Adverse Drug Reaction (ADR). We find that the approach applied to Twitter data can detect symptom mentions substantially before being reported by the Centers for Disease Control (CDC).

Sharath Chandra Guntuku, Mingyang Li, Louis Tay et al.

Global acceptance of Emojis suggests a cross-cultural, normative use of Emojis. Meanwhile, nuances in Emoji use across cultures may also exist due to linguistic differences in expressing emotions and diversity in conceptualizing topics. Indeed, literature in cross-cultural psychology has found both normative and culture-specific ways in which emotions are expressed. In this paper, using social media, we compare the Emoji usage based on frequency, context, and topic associations across countries in the East (China and Japan) and the West (United States, United Kingdom, and Canada). Across the East and the West, our study examines a) similarities and differences on the usage of different categories of Emojis such as People, Food \& Drink, Travel \& Places etc., b) potential mapping of Emoji use differences with previously identified cultural differences in users' expression about diverse concepts such as death, money emotions and family, and c) relative correspondence of validated psycho-linguistic categories with Ekman's emotions. The analysis of Emoji use in the East and the West reveals recognizable normative and culture specific patterns. This research reveals the ways in which Emojis can be used for cross-cultural communication.

Sharath Chandra Guntuku, Daniel Preotiuc-Pietro, Johannes C. Eichstaedt et al.

Previous work has found strong links between the choice of social media images and users' emotions, demographics and personality traits. In this study, we examine which attributes of profile and posted images are associated with depression and anxiety of Twitter users. We used a sample of 28,749 Facebook users to build a language prediction model of survey-reported depression and anxiety, and validated it on Twitter on a sample of 887 users who had taken anxiety and depression surveys. We then applied it to a different set of 4,132 Twitter users to impute language-based depression and anxiety labels, and extracted interpretable features of posted and profile pictures to uncover the associations with users' depression and anxiety, controlling for demographics. For depression, we find that profile pictures suppress positive emotions rather than display more negative emotions, likely because of social media self-presentation biases. They also tend to show the single face of the user (rather than show her in groups of friends), marking increased focus on the self, emblematic for depression. Posted images are dominated by grayscale and low aesthetic cohesion across a variety of image features. Profile images of anxious users are similarly marked by grayscale and low aesthetic cohesion, but less so than those of depressed users. Finally, we show that image features can be used to predict depression and anxiety, and that multitask learning that includes a joint modeling of demographics improves prediction performance. Overall, we find that the image attributes that mark depression and anxiety offer a rich lens into these conditions largely congruent with the psychological literature, and that images on Twitter allow inferences about the mental health status of users.

José Marcio Luna, Eric Eaton, Lyle H. Ungar et al.

Additive models, such as produced by gradient boosting, and full interaction models, such as classification and regression trees (CART), are widely used algorithms that have been investigated largely in isolation. We show that these models exist along a spectrum, revealing never-before-known connections between these two approaches. This paper introduces a novel technique called tree-structured boosting for creating a single decision tree, and shows that this method can produce models equivalent to CART or gradient boosted stumps at the extremes by varying a single parameter. Although tree-structured boosting is designed primarily to provide both the model interpretability and predictive performance needed for high-stake applications like medicine, it also can produce decision trees represented by hybrid models between CART and boosted stumps that can outperform either of these approaches.

Alexandrin Popescul, Lyle H. Ungar, David M Pennock et al.

Recommender systems leverage product and community information to target products to consumers. Researchers have developed collaborative recommenders, content-based recommenders, and (largely ad-hoc) hybrid systems. We propose a unified probabilistic framework for merging collaborative and content-based recommendations. We extend Hofmann's [1999] aspect model to incorporate three-way co-occurrence data among users, items, and item content. The relative influence of collaboration data versus content data is not imposed as an exogenous parameter, but rather emerges naturally from the given data sources. Global probabilistic models coupled with standard Expectation Maximization (EM) learning algorithms tend to drastically overfit in sparse-data situations, as is typical in recommendation applications. We show that secondary content information can often be used to overcome sparsity. Experiments on data from the ResearchIndex library of Computer Science publications show that appropriate mixture models incorporating secondary data produce significantly better quality recommenders than k-nearest neighbors (k-NN). Global probabilistic models also allow more general inferences than local methods like k-NN.

Dean P. Foster, Jordan Rodu, Lyle H. Ungar

Hidden Markov Models (HMMs) can be accurately approximated using co-occurrence frequencies of pairs and triples of observations by using a fast spectral method in contrast to the usual slow methods like EM or Gibbs sampling. We provide a new spectral method which significantly reduces the number of model parameters that need to be estimated, and generates a sample complexity that does not depend on the size of the observation vocabulary. We present an elementary proof giving bounds on the relative accuracy of probability estimates from our model. (Correlaries show our bounds can be weakened to provide either L1 bounds or KL bounds which provide easier direct comparisons to previous work.) Our theorem uses conditions that are checkable from the data, instead of putting conditions on the unobservable Markov transition matrix.