Large language models (LLMs) are increasingly used as raters for evaluation tasks. However, their reliability is often limited for subjective tasks, when human judgments involve subtle reasoning beyond annotation labels. Thinking traces, the reasoning behind a judgment, are highly informative but challenging to collect and curate. We present a human-LLM collaborative framework to infer thinking traces from label-only annotations. The proposed framework uses a simple and effective rejection sampling method to reconstruct these traces at scale. These inferred thinking traces are applied to two complementary tasks: (1) fine-tuning open LLM raters; and (2) synthesizing clearer annotation guidelines for proprietary LLM raters. Across multiple datasets, our methods lead to significantly improved LLM-human agreement. Additionally, the refined annotation guidelines increase agreement among different LLM models. These results suggest that LLMs can serve as practical proxies for otherwise unrevealed human thinking traces, enabling label-only corpora to be extended into thinking-trace-augmented resources that enhance the reliability of LLM raters.

Investment Analysis is a cornerstone of the Financial Services industry. The rapid integration of advanced machine learning techniques, particularly Large Language Models (LLMs), offers opportunities to enhance the equity rating process. This paper explores the application of LLMs to generate multi-horizon stock ratings by ingesting diverse datasets. Traditional stock rating methods rely heavily on the expertise of financial analysts, and face several challenges such as data overload, inconsistencies in filings, and delayed reactions to market events. Our study addresses these issues by leveraging LLMs to improve the accuracy and consistency of stock ratings. Additionally, we assess the efficacy of using different data modalities with LLMs for the financial domain. We utilize varied datasets comprising fundamental financial, market, and news data from January 2022 to June 2024, along with GPT-4-32k (v0613) (with a training cutoff in Sep. 2021 to prevent information leakage). Our results show that our benchmark method outperforms traditional stock rating methods when assessed by forward returns, specially when incorporating financial fundamentals. While integrating news data improves short-term performance, substituting detailed news summaries with sentiment scores reduces token use without loss of performance. In many cases, omitting news data entirely enhances performance by reducing bias. Our research shows that LLMs can be leveraged to effectively utilize large amounts of multimodal financial data, as showcased by their effectiveness at the stock rating prediction task. Our work provides a reproducible and efficient framework for generating accurate stock ratings, serving as a cost-effective alternative to traditional methods. Future work will extend to longer timeframes, incorporate diverse data, and utilize newer models for enhanced insights.

In an era where societal narratives are increasingly shaped by algorithmic curation, investigating the political neutrality of LLMs is an important research question. This study presents a fresh perspective on quantifying the political neutrality of LLMs through the lens of abstractive text summarization of polarizing news articles. We consider five pressing issues in current US politics: abortion, gun control/rights, healthcare, immigration, and LGBTQ+ rights. Via a substantial corpus of 20,344 news articles, our study reveals a consistent trend towards pro-Democratic biases in several well-known LLMs, with gun control and healthcare exhibiting the most pronounced biases (max polarization differences of -9.49% and -6.14%, respectively). Further analysis uncovers a strong convergence in the vocabulary of the LLM outputs for these divisive topics (55% overlap for Democrat-leaning representations, 52% for Republican). Being months away from a US election of consequence, we consider our findings important.

Abstract: With the continuous advancement of artificial intelligence, natural language processing technology has become widely utilized in various fields. At the same time, there are many challenges in creating Chinese news summaries. First of all, the semantics of Chinese news is complex, and the amount of information is enormous. Extracting critical information from Chinese news presents a significant challenge. Second, the news summary should be concise and clear, focusing on the main content and avoiding redundancy. In addition, the particularity of the Chinese language, such as polysemy, word segmentation, etc., makes it challenging to generate Chinese news summaries. Based on the above, this paper studies the information extraction method of the LCSTS dataset based on an improved BERTSum-LSTM model. We improve the BERTSum-LSTM model to make it perform better in generating Chinese news summaries. The experimental results show that the proposed method has a good effect on creating news summaries, which is of great importance to the construction of news summaries.

News summary generation is an important task in the field of intelligence analysis, which can provide accurate and comprehensive information to help people better understand and respond to complex real-world events. However, traditional news summary generation methods face some challenges, which are limited by the model itself and the amount of training data, as well as the influence of text noise, making it difficult to generate reliable information accurately. In this paper, we propose a new paradigm for news summary generation using LLM with powerful natural language understanding and generative capabilities. We use LLM to extract multiple structured event patterns from the events contained in news paragraphs, evolve the event pattern population with genetic algorithm, and select the most adaptive event pattern to input into the LLM to generate news summaries. A News Summary Generator (NSG) is designed to select and evolve the event pattern populations and generate news summaries. The experimental results show that the news summary generator is able to generate accurate and reliable news summaries with some generalization ability.

The rapid advancement of Large Language Models (LLMs) has led to extensive discourse regarding their potential to boost the return of quantitative stock trading strategies. This discourse primarily revolves around harnessing the remarkable comprehension capabilities of LLMs to extract sentiment factors which facilitate informed and high-frequency investment portfolio adjustments. To ensure successful implementations of these LLMs into the analysis of Chinese financial texts and the subsequent trading strategy development within the Chinese stock market, we provide a rigorous and encompassing benchmark as well as a standardized back-testing framework aiming at objectively assessing the efficacy of various types of LLMs in the specialized domain of sentiment factor extraction from Chinese news text data. To illustrate how our benchmark works, we reference three distinctive models: 1) the generative LLM (ChatGPT), 2) the Chinese language-specific pre-trained LLM (Erlangshen-RoBERTa), and 3) the financial domain-specific fine-tuned LLM classifier(Chinese FinBERT). We apply them directly to the task of sentiment factor extraction from large volumes of Chinese news summary texts. We then proceed to building quantitative trading strategies and running back-tests under realistic trading scenarios based on the derived sentiment factors and evaluate their performances with our benchmark. By constructing such a comparative analysis, we invoke the question of what constitutes the most important element for improving a LLM's performance on extracting sentiment factors. And by ensuring that the LLMs are evaluated on the same benchmark, following the same standardized experimental procedures that are designed with sufficient expertise in quantitative trading, we make the first stride toward answering such a question.

Researchers have been investigating automated solutions for fact-checking in a variety of fronts. However, current approaches often overlook the fact that the amount of information released every day is escalating, and a large amount of them overlap. Intending to accelerate fact-checking, we bridge this gap by grouping similar messages and summarizing them into aggregated claims. Specifically, we first clean a set of social media posts (e.g., tweets) and build a graph of all posts based on their semantics; Then, we perform two clustering methods to group the messages for further claim summarization. We evaluate the summaries both quantitatively with ROUGE scores and qualitatively with human evaluation. We also generate a graph of summaries to verify that there is no significant overlap among them. The results reduced 28,818 original messages to 700 summary claims, showing the potential to speed up the fact-checking process by organizing and selecting representative claims from massive disorganized and redundant messages.