Verbosity $ eq$ Veracity: Demystify Verbosity Compensation Behavior of Large Language Models

When unsure about an answer, humans often respond with more words than necessary, hoping that part of the response will be correct. We observe a similar behavior in large language models (LLMs), which we term "Verbosity Compensation" (VC). VC is harmful because it confuses the user understanding, leading to low efficiency, and influences the LLM services by increasing the latency and cost of generating useless tokens. In this paper, we present the first work that defines and analyzes Verbosity Compensation, explores its causes, and proposes a simple mitigating approach. We define Verbosity Compensation as the behavior of generating responses that can be compressed without information loss when prompted to write concisely. Our experiments, conducted on five datasets of knowledge and reasoning-based QA tasks with 14 newly developed LLMs, reveal three conclusions. 1) We reveal a pervasive presence of verbosity compensation across all models and all datasets. Notably, GPT-4 exhibits a VC frequency of 50.40%. 2) We reveal the large performance gap between verbose and concise responses, with a notable difference of 27.61% on the Qasper dataset. We also demonstrate that this difference does not naturally diminish as LLM capability increases. Both 1) and 2) highlight the urgent need to mitigate the frequency of VC behavior and disentangle verbosity with veracity. We propose a simple yet effective cascade algorithm that replaces the verbose responses with the other model-generated responses. The results show that our approach effectively alleviates the VC of the Mistral model from 63.81% to 16.16% on the Qasper dataset. 3) We also find that verbose responses exhibit higher uncertainty across all five datasets, suggesting a strong connection between verbosity and model uncertainty. Our dataset and code are available at https://github.com/psunlpgroup/VerbosityLLM.

Evaluating LLMs at Detecting Errors in LLM Responses

With Large Language Models (LLMs) being widely used across various tasks, detecting errors in their responses is increasingly crucial. However, little research has been conducted on error detection of LLM responses. Collecting error annotations on LLM responses is challenging due to the subjective nature of many NLP tasks, and thus previous research focuses on tasks of little practical value (e.g., word sorting) or limited error types (e.g., faithfulness in summarization). This work introduces ReaLMistake, the first error detection benchmark consisting of objective, realistic, and diverse errors made by LLMs. ReaLMistake contains three challenging and meaningful tasks that introduce objectively assessable errors in four categories (reasoning correctness, instruction-following, context-faithfulness, and parameterized knowledge), eliciting naturally observed and diverse errors in responses of GPT-4 and Llama 2 70B annotated by experts. We use ReaLMistake to evaluate error detectors based on 12 LLMs. Our findings show: 1) Top LLMs like GPT-4 and Claude 3 detect errors made by LLMs at very low recall, and all LLM-based error detectors perform much worse than humans. 2) Explanations by LLM-based error detectors lack reliability. 3) LLMs-based error detection is sensitive to small changes in prompts but remains challenging to improve. 4) Popular approaches to improving LLMs, including self-consistency and majority vote, do not improve the error detection performance. Our benchmark and code are provided at https://github.com/psunlpgroup/ReaLMistake.

Unified Low-Resource Sequence Labeling by Sample-Aware Dynamic Sparse Finetuning

Unified Sequence Labeling that articulates different sequence labeling problems such as Named Entity Recognition, Relation Extraction, Semantic Role Labeling, etc. in a generalized sequence-to-sequence format opens up the opportunity to make the maximum utilization of large language model knowledge toward structured prediction. Unfortunately, this requires formatting them into specialized augmented format unknown to the base pretrained language model (PLMs) necessitating finetuning to the target format. This significantly bounds its usefulness in data-limited settings where finetuning large models cannot properly generalize to the target format. To address this challenge and leverage PLM knowledge effectively, we propose FISH-DIP, a sample-aware dynamic sparse finetuning strategy that selectively focuses on a fraction of parameters, informed by feedback from highly regressing examples, during the fine-tuning process. By leveraging the dynamism of sparsity, our approach mitigates the impact of well-learned samples and prioritizes underperforming instances for improvement in generalization. Across five tasks of sequence labeling, we demonstrate that FISH-DIP can smoothly optimize the model in low resource settings offering upto 40% performance improvements over full fine-tuning depending on target evaluation settings. Also, compared to in-context learning and other parameter-efficient fine-tuning approaches, FISH-DIP performs comparably or better, notably in extreme low-resource settings.

Hermes: Unlocking Security Analysis of Cellular Network Protocols by Synthesizing Finite State Machines from Natural Language Specifications

In this paper, we present Hermes, an end-to-end framework to automatically generate formal representations from natural language cellular specifications. We first develop a neural constituency parser, NEUTREX, to process transition-relevant texts and extract transition components (i.e., states, conditions, and actions). We also design a domain-specific language to translate these transition components to logical formulas by leveraging dependency parse trees. Finally, we compile these logical formulas to generate transitions and create the formal model as finite state machines. To demonstrate the effectiveness of Hermes, we evaluate it on 4G NAS, 5G NAS, and 5G RRC specifications and obtain an overall accuracy of 81-87%, which is a substantial improvement over the state-of-the-art. Our security analysis of the extracted models uncovers 3 new vulnerabilities and identifies 19 previous attacks in 4G and 5G specifications, and 7 deviations in commercial 4G basebands.

S3-DST: Structured Open-Domain Dialogue Segmentation and State Tracking in the Era of LLMs

The traditional Dialogue State Tracking (DST) problem aims to track user preferences and intents in user-agent conversations. While sufficient for task-oriented dialogue systems supporting narrow domain applications, the advent of Large Language Model (LLM)-based chat systems has introduced many real-world intricacies in open-domain dialogues. These intricacies manifest in the form of increased complexity in contextual interactions, extended dialogue sessions encompassing a diverse array of topics, and more frequent contextual shifts. To handle these intricacies arising from evolving LLM-based chat systems, we propose joint dialogue segmentation and state tracking per segment in open-domain dialogue systems. Assuming a zero-shot setting appropriate to a true open-domain dialogue system, we propose S3-DST, a structured prompting technique that harnesses Pre-Analytical Recollection, a novel grounding mechanism we designed for improving long context tracking. To demonstrate the efficacy of our proposed approach in joint segmentation and state tracking, we evaluate S3-DST on a proprietary anonymized open-domain dialogue dataset, as well as publicly available DST and segmentation datasets. Across all datasets and settings, S3-DST consistently outperforms the state-of-the-art, demonstrating its potency and robustness the next generation of LLM-based chat systems.

Using Large Language Models to Generate, Validate, and Apply User Intent Taxonomies

Log data can reveal valuable information about how users interact with web search services, what they want, and how satisfied they are. However, analyzing user intents in log data is not easy, especially for new forms of web search such as AI-driven chat. To understand user intents from log data, we need a way to label them with meaningful categories that capture their diversity and dynamics. Existing methods rely on manual or ML-based labeling, which are either expensive or inflexible for large and changing datasets. We propose a novel solution using large language models (LLMs), which can generate rich and relevant concepts, descriptions, and examples for user intents. However, using LLMs to generate a user intent taxonomy and apply it to do log analysis can be problematic for two main reasons: such a taxonomy is not externally validated, and there may be an undesirable feedback loop. To overcome these issues, we propose a new methodology with human experts and assessors to verify the quality of the LLM-generated taxonomy. We also present an end-to-end pipeline that uses an LLM with human-in-the-loop to produce, refine, and use labels for user intent analysis in log data. Our method offers a scalable and adaptable way to analyze user intents in web-scale log data with minimal human effort. We demonstrate its effectiveness by uncovering new insights into user intents from search and chat logs from Bing.

CONTaiNER: Few-Shot Named Entity Recognition via Contrastive Learning

Named Entity Recognition (NER) in Few-Shot setting is imperative for entity tagging in low resource domains. Existing approaches only learn class-specific semantic features and intermediate representations from source domains. This affects generalizability to unseen target domains, resulting in suboptimal performances. To this end, we present CONTaiNER, a novel contrastive learning technique that optimizes the inter-token distribution distance for Few-Shot NER. Instead of optimizing class-specific attributes, CONTaiNER optimizes a generalized objective of differentiating between token categories based on their Gaussian-distributed embeddings. This effectively alleviates overfitting issues originating from training domains. Our experiments in several traditional test domains (OntoNotes, CoNLL'03, WNUT '17, GUM) and a new large scale Few-Shot NER dataset (Few-NERD) demonstrate that on average, CONTaiNER outperforms previous methods by 3%-13% absolute F1 points while showing consistent performance trends, even in challenging scenarios where previous approaches could not achieve appreciable performance.

A Survey on Deep Learning Based Point-Of-Interest Recommendations

Location-based Social Networks (LBSNs) enable users to socialize with friends and acquaintances by sharing their check-ins, opinions, photos, and reviews. Huge volume of data generated from LBSNs opens up a new avenue of research that gives birth to a new sub-field of recommendation systems, known as Point-of-Interest (POI) recommendation. A POI recommendation technique essentially exploits users' historical check-ins and other multi-modal information such as POI attributes and friendship network, to recommend the next set of POIs suitable for a user. A plethora of earlier works focused on traditional machine learning techniques by using hand-crafted features from the dataset. With the recent surge of deep learning research, we have witnessed a large variety of POI recommendation works utilizing different deep learning paradigms. These techniques largely vary in problem formulations, proposed techniques, used datasets, and features, etc. To the best of our knowledge, this work is the first comprehensive survey of all major deep learning-based POI recommendation works. Our work categorizes and critically analyzes the recent POI recommendation works based on different deep learning paradigms and other relevant features. This review can be considered a cookbook for researchers or practitioners working in the area of POI recommendation.

BayesBeat: A Bayesian Deep Learning Approach for Atrial Fibrillation Detection from Noisy Photoplethysmography Data

The increasing popularity of smartwatches as affordable and longitudinal monitoring devices enables us to capture photoplethysmography (PPG) sensor data for detecting Atrial Fibrillation (AF) in real-time. A significant challenge in AF detection from PPG signals comes from the inherent noise in the smartwatch PPG signals. In this paper, we propose a novel deep learning based approach, BayesBeat that leverages the power of Bayesian deep learning to accurately infer AF risks from noisy PPG signals, and at the same time provide the uncertainty estimate of the prediction. Bayesbeat is efficient, robust, flexible, and highly scalable which makes it particularly suitable for deployment in commercially available wearable devices. Extensive experiments on a recently published large dataset reveal that our proposed method BayesBeat substantially outperforms the existing state-of-the-art methods.

Boosting House Price Predictions using Geo-Spatial Network Embedding

Real estate contributes significantly to all major economies around the world. In particular, house prices have a direct impact on stakeholders, ranging from house buyers to financing companies. Thus, a plethora of techniques have been developed for real estate price prediction. Most of the existing techniques rely on different house features to build a variety of prediction models to predict house prices. Perceiving the effect of spatial dependence on house prices, some later works focused on introducing spatial regression models for improving prediction performance. However, they fail to take into account the geo-spatial context of the neighborhood amenities such as how close a house is to a train station, or a highly-ranked school, or a shopping center. Such contextual information may play a vital role in users' interests in a house and thereby has a direct influence on its price. In this paper, we propose to leverage the concept of graph neural networks to capture the geo-spatial context of the neighborhood of a house. In particular, we present a novel method, the Geo-Spatial Network Embedding (GSNE), that learns the embeddings of houses and various types of Points of Interest (POIs) in the form of multipartite networks, where the houses and the POIs are represented as attributed nodes and the relationships between them as edges. Extensive experiments with a large number of regression techniques show that the embeddings produced by our proposed GSNE technique consistently and significantly improve the performance of the house price prediction task regardless of the downstream regression model.