Genomic question answering often requires complex reasoning and integration across diverse biomedical sources. GeneGPT addressed this challenge by combining domain-specific APIs with OpenAI's code-davinci-002 large language model to enable natural language interaction with genomic databases. However, its reliance on a proprietary model limits scalability, increases operational costs, and raises concerns about data privacy and generalization. In this work, we revisit and reproduce GeneGPT in a pilot study using open source models, including Llama 3.1, Qwen2.5, and Qwen2.5 Coder, within a monolithic architecture; this allows us to identify the limitations of this approach. Building on this foundation, we then develop OpenBioLLM, a modular multi-agent framework that extends GeneGPT by introducing agent specialization for tool routing, query generation, and response validation. This enables coordinated reasoning and role-based task execution. OpenBioLLM matches or outperforms GeneGPT on over 90
Designing document identifiers (docids) that carry rich semantic information while maintaining tractable search spaces is a important challenge in generative retrieval (GR). Popular codebook methods address this by building a hierarchical semantic tree and constraining generation to its child nodes, yet their numeric identifiers cannot leverage the large language model's pretrained natural language understanding. Conversely, using text as docid provides more semantic expressivity but inflates the decoding space, making the system brittle to early-step errors. To resolve this trade-off, we propose C2T-ID: (i) first construct semantic numerical docid via hierarchical clustering; (ii) then extract high-frequency metadata keywords and iteratively replace each numeric label with its cluster's top-K keywords; and (iii) an optional two-level semantic smoothing step further enhances the fluency of C2T-ID. Experiments on Natural Questions and Taobao's product search demonstrate that C2T-ID significantly outperforms atomic, semantic codebook, and pure-text docid baselines, demonstrating its effectiveness in balancing semantic expressiveness with search space constraints.
Task-oriented conversational systems are essential for efficiently addressing diverse user needs, yet their development requires substantial amounts of high-quality conversational data that is challenging and costly to obtain. While large language models (LLMs) have demonstrated potential in generating synthetic conversations, the extent to which these agent-generated interactions can effectively substitute real human conversations remains unclear. This work presents the first systematic comparison between LLM-simulated users and human users in personalized task-oriented conversations. We propose a comprehensive analytical framework encompassing three key aspects (conversation strategy, interaction style, and conversation evaluation) and ten distinct dimensions for evaluating user behaviors, and collect parallel conversational datasets from both human users and LLM agent users across four representative scenarios under identical conditions. Our analysis reveals significant behavioral differences between the two user types in problem-solving approaches, question broadness, user engagement, context dependency, feedback polarity and promise, language style, and hallucination awareness. We found consistency in the agent users and human users across the depth-first or breadth-first dimensions, as well as the usefulness dimensions. These findings provide critical insights for advancing LLM-based user simulation. Our multi-dimensional taxonomy constructed a generalizable framework for analyzing user behavior patterns, offering insights from LLM agent users and human users. By this work, we provide perspectives on rethinking how to use user simulation in conversational systems in the future.
Large language models (LLMs) are increasingly deployed in information systems, including being used as second-stage rerankers in information retrieval pipelines, yet their susceptibility to recency bias has received little attention. We investigate whether LLMs implicitly favour newer documents by prepending artificial publication dates to passages in the TREC Deep Learning passage retrieval collections in 2021 (DL21) and 2022 (DL22). Across seven models, GPT-3.5-turbo, GPT-4o, GPT-4, LLaMA-3 8B/70B, and Qwen-2.5 7B/72B, "fresh" passages are consistently promoted, shifting the Top-10's mean publication year forward by up to 4.78 years and moving individual items by as many as 95 ranks in our listwise reranking experiments. Although larger models attenuate the effect, none eliminate it. We also observe that the preference of LLMs between two passages with an identical relevance level can be reversed by up to 25
Experiments often result in the need to compare an observation against a reference, where observation and reference are selections made from some specified domain. The goal is to determine how close the observation is to the ideal result represented by the reference, so that, all other things being equal, systems that achieve outputs closer to the ideal reference can be preferred for deployment. Both observation and reference might be sets of items, or might be ordered sequences (rankings) of items. There are thus four possible combinations between sets and rankings. Three of those possibilities are already familiar to IR researchers, and have received detailed exploration. Here we consider the fourth combination, that of comparing an observation set relative to a reference ranking. We introduce a new measurement that we call rank-biased recall to cover this scenario, and demonstrate its usefulness with a case study from multi-phase ranking. We also present a new top-weighted ''ranking compared to ranking'' measurement, and show that it represents a complementary assessment to the previous rank-biased overlap mechanism, and possesses distinctive characteristics
Considering query variance in information retrieval (IR) experiments is beneficial for retrieval effectiveness. Especially ranking ensembles based on different topically related queries retrieve better results than rankings based on a single query alone. Recently, generative instruction-tuned Large Language Models (LLMs) improved on a variety of different tasks in capturing human language. To this end, this work explores the feasibility of using synthetic query variants generated by instruction-tuned LLMs in data fusion experiments. More specifically, we introduce a lightweight, unsupervised, and cost-efficient approach that exploits principled prompting and data fusion techniques. In our experiments, LLMs produce more effective queries when provided with additional context information on the topic. Furthermore, our analysis based on four TREC newswire benchmarks shows that data fusion based on synthetic query variants is significantly better than baselines with single queries and also outperforms pseudo-relevance feedback methods. We publicly share the code and query datasets with the community as resources for follow-up studies.
Information-seeking dialogues span a wide range of questions, from simple factoid to complex queries that require exploring multiple facets and viewpoints. When performing exploratory searches in unfamiliar domains, users may lack background knowledge and struggle to verify the system-provided information, making them vulnerable to misinformation. We investigate the limitations of response generation in conversational information-seeking systems, highlighting potential inaccuracies, pitfalls, and biases in the responses. The study addresses the problem of query answerability and the challenge of response incompleteness. Our user studies explore how these issues impact user experience, focusing on users' ability to identify biased, incorrect, or incomplete responses. We design two crowdsourcing tasks to assess user experience with different system response variants, highlighting critical issues to be addressed in future conversational information-seeking research. Our analysis reveals that it is easier for users to detect response incompleteness than query answerability and user satisfaction is mostly associated with response diversity, not factual correctness.
Language Models (LMs) memorize a vast amount of factual knowledge, exhibiting strong performance across diverse tasks and domains. However, it has been observed that the performance of these models diminishes when dealing with less-popular or low-frequency concepts, for example, in domain-specific applications. The two prominent approaches to enhance the performance of LMs on less frequent topics are Retrieval Augmented Generation (RAG) and fine-tuning (FT) over synthetic data. This paper explores and evaluates the impact of RAG and FT on customizing LMs in handling low-frequency entities in question answering tasks. We conduct extensive experiments on twelve LMs of varying size and type and different FT methods, data augmentation, and retrieval models. Our findings indicate that while FT boosts performance across entities of varying popularity, RAG surpasses FT by a large margin, particularly for least popular factual knowledge. Additionally, the success of both RAG and FT approaches is amplified by improving retrieval and data augmentation techniques. Fine-tuning, while beneficial for small LMs, requires extensive resources. To address this issue, we propose the new Stimulus RAG approach that surpasses the effectiveness of fine-tuning based approaches, thereby eliminating the need for the costly data augmentation and fine-tuning steps for enriching LMs with less popular factual knowledge.
The emergence of Vec2Text --- a method for text embedding inversion --- has raised serious privacy concerns for dense retrieval systems which use text embeddings, such as those offered by OpenAI and Cohere. This threat comes from the ability for a malicious attacker with access to embeddings to reconstruct the original text. In this paper, we investigate various factors related to embedding models that may impact text recoverability via Vec2Text. We explore factors such as distance metrics, pooling functions, bottleneck pre-training, training with noise addition, embedding quantization, and embedding dimensions, which were not considered in the original Vec2Text paper. Through a comprehensive analysis of these factors, our objective is to gain a deeper understanding of the key elements that affect the trade-offs between the text recoverability and retrieval effectiveness of dense retrieval systems, offering insights for practitioners designing privacy-aware dense retrieval systems. We also propose a simple embedding transformation fix that guarantees equal ranking effectiveness while mitigating the recoverability risk. Overall, this study reveals that Vec2Text could pose a threat to current dense retrieval systems, but there are some effective methods to patch such systems.
Recommendation foundation model utilizes large language models (LLM) for recommendation by converting recommendation tasks into natural language tasks. It enables generative recommendation which directly generates the item(s) to recommend rather than calculating a ranking score for each and every candidate item as in traditional recommendation models, simplifying the recommendation pipeline from multi-stage filtering to single-stage filtering. To avoid generating excessively long text and hallucinated recommendations when deciding which item(s) to recommend, creating LLM-compatible item IDs to uniquely identify each item is essential for recommendation foundation models. In this study, we systematically examine the item ID creation and indexing problem for recommendation foundation models, using P5 as an example of the backbone LLM. To emphasize the importance of item indexing, we first discuss the issues of several trivial item indexing methods, such as random indexing, title indexing, and independent indexing. We then propose four simple yet effective solutions, including sequential indexing, collaborative indexing, semantic (content-based) indexing, and hybrid indexing. Our study highlights the significant influence of item indexing methods on the performance of LLM-based recommendation, and our results on real-world datasets validate the effectiveness of our proposed solutions. The research also demonstrates how recent advances on language modeling and traditional IR principles such as indexing can help each other for better learning and inference. Source code and data are available at https://github.com/Wenyueh/LLM-RecSys-ID.
