arXiv subjects

Aldo Lipani

Publications and source records attributed to Aldo Lipani.

At least 19 recordsLinked to original sources

arXiv2026

Mitigating Context Interference for Reliable and Efficient Search Agents

Recent research empowers Large Language Models (LLMs) as multi-turn search agents to iteratively retrieve and generate outputs until complex tasks are solved. However, the contexts of multi-turn search agents are lengthy and complex. For example, the retrieved set of documents in each turn would inevitably introduce irrelevant information that distracts LLMs, referring to \textit{context interference}, potentially hindering the reliability and efficiency of search agents. Therefore, we conduct a systematic study on context interference in multi-turn search agents, focusing on investigating i) which parts of the context of search agents will contribute to the context interference, ii) how to refine the contexts of search agents to mitigate the interference, and iii) can incorporating context refinement into search agent training yield further improvements. We reveal that interference primarily arises from the latest retrieved documents. Based on the explored findings, we then introduce a distill-based context refiner to dynamically mitigate context interference for multi-turn search agents. Finally, we validate that incorporating context refinement into RL training pipelines of search agents can significantly enhance both reliability and efficiency. This study highlights the importance of mitigating context interference of search agents, inspiring a novel paradigm of ``refine context and then generate'' for AI agents.

cs.CL↗

arXiv2026

Multilingual Refusal Alignment for Safer Large Language Models

As Large Language Models (LLMs) are deployed globally, ensuring their safety and alignment across multiple languages becomes paramount. However, safety behaviors often vary unpredictably between languages, posing significant challenges for consistent and ethical AI. In this work, we systematically investigate the dynamics of multilingual alignment, exploring whether single-language alignment transfers cross-lingually, how language consistency is preserved during training, and the resulting trade-offs with general knowledge capabilities. We introduce RefusEU, a novel refusal alignment dataset covering 12 European languages, including a dedicated test set for evaluating current state-of-the-art models. Our controlled Direct Preference Optimization (DPO) experiments provide two key insights: aligning models exclusively in English is insufficient to ensure cross-lingual safety, even for the same harm categories, whereas training on multilingual datasets can improve safety without degrading general performance, as measured by the Global MMLU benchmark.

cs.CL↗

arXiv2026

Interplay: Training Independent Simulators for Reference-Free Conversational Recommendation

Training conversational recommender systems (CRS) requires extensive dialogue data, which is challenging to collect at scale. To address this, researchers have used simulated user-recommender conversations. Traditional simulation approaches often utilize a single large language model (LLM) that generates entire conversations with prior knowledge of the target items, leading to scripted and artificial dialogues. We propose a reference-free simulation framework that trains two independent LLMs, one as the user and one as the conversational recommender. These models interact in real-time without access to predetermined target items, but preference summaries and target attributes, enabling the recommender to genuinely infer user preferences through dialogue. This approach produces more realistic and diverse conversations that closely mirror authentic human-AI interactions. Our reference-free simulators match or exceed existing methods in quality, while offering a scalable solution for generating high-quality conversational recommendation data without constraining conversations to pre-defined target items. We conduct both quantitative and human evaluations to confirm the effectiveness of our reference-free approach.

cs.AI↗

arXiv2025

Encode, Think, Decode: Scaling test-time reasoning with recursive latent thoughts

Most efforts to improve the reasoning capabilities of large language models (LLMs) involve either scaling the number of parameters and the size of training data, or scaling inference computation by letting models generate complex chains of thought. Motivated by interpretability studies showing that the crucial computation required for reasoning tasks is concentrated in a limited range of layers, we introduce Encode-Think-Decode (ETD), a method that enhances the reasoning capabilities of a base model by training it to iterate over a small subset of reasoning-relevant layers during the mid-training stage. ETD amplifies latent reasoning while preserving the original architecture, parameter count, hyperparameters, and training data composition. When iterating on the selected layers at inference time, ETD models yield substantial gains on 17 reasoning benchmarks, including +28.4% relative accuracy improvement on GSM8K and +36% on MATH with the OLMo-2 1B Base model. We also explore an adaptive depth strategy that adjusts the computation per input token. Our results show that recursive latent reasoning offers a simple and effective path to stronger LLM reasoning.

cs.LG↗

arXiv2025

PREF: Reference-Free Evaluation of Personalised Text Generation in LLMs

Personalised text generation is essential for user-centric information systems, yet most evaluation methods overlook the individuality of users. We introduce \textbf{PREF}, a \textbf{P}ersonalised \textbf{R}eference-free \textbf{E}valuation \textbf{F}ramework that jointly measures general output quality and user-specific alignment without requiring gold personalised references. PREF operates in a three-step pipeline: (1) a coverage stage uses a large language model (LLM) to generate a comprehensive, query-specific guideline covering universal criteria such as factuality, coherence, and completeness; (2) a preference stage re-ranks and selectively augments these factors using the target user's profile, stated or inferred preferences, and context, producing a personalised evaluation rubric; and (3) a scoring stage applies an LLM judge to rate candidate answers against this rubric, ensuring baseline adequacy while capturing subjective priorities. This separation of coverage from preference improves robustness, transparency, and reusability, and allows smaller models to approximate the personalised quality of larger ones. Experiments on the PrefEval benchmark, including implicit preference-following tasks, show that PREF achieves higher accuracy, better calibration, and closer alignment with human judgments than strong baselines. By enabling scalable, interpretable, and user-aligned evaluation, PREF lays the groundwork for more reliable assessment and development of personalised language generation systems.

cs.CL↗

arXiv2025

Dagstuhl Perspectives Workshop 24352 -- Conversational Agents: A Framework for Evaluation (CAFE): Manifesto

During the workshop, we deeply discussed what CONversational Information ACcess (CONIAC) is and its unique features, proposing a world model abstracting it, and defined the Conversational Agents Framework for Evaluation (CAFE) for the evaluation of CONIAC systems, consisting of six major components: 1) goals of the system's stakeholders, 2) user tasks to be studied in the evaluation, 3) aspects of the users carrying out the tasks, 4) evaluation criteria to be considered, 5) evaluation methodology to be applied, and 6) measures for the quantitative criteria chosen.

cs.CL↗

arXiv2024

A Reproducibility and Generalizability Study of Large Language Models for Query Generation

Systematic literature reviews (SLRs) are a cornerstone of academic research, yet they are often labour-intensive and time-consuming due to the detailed literature curation process. The advent of generative AI and large language models (LLMs) promises to revolutionize this process by assisting researchers in several tedious tasks, one of them being the generation of effective Boolean queries that will select the publications to consider including in a review. This paper presents an extensive study of Boolean query generation using LLMs for systematic reviews, reproducing and extending the work of Wang et al. and Alaniz et al. Our study investigates the replicability and reliability of results achieved using ChatGPT and compares its performance with open-source alternatives like Mistral and Zephyr to provide a more comprehensive analysis of LLMs for query generation. Therefore, we implemented a pipeline, which automatically creates a Boolean query for a given review topic by using a previously defined LLM, retrieves all documents for this query from the PubMed database and then evaluates the results. With this pipeline we first assess whether the results obtained using ChatGPT for query generation are reproducible and consistent. We then generalize our results by analyzing and evaluating open-source models and evaluating their efficacy in generating Boolean queries. Finally, we conduct a failure analysis to identify and discuss the limitations and shortcomings of using LLMs for Boolean query generation. This examination helps to understand the gaps and potential areas for improvement in the application of LLMs to information retrieval tasks. Our findings highlight the strengths, limitations, and potential of LLMs in the domain of information retrieval and literature review automation.

cs.IR↗

arXiv2024

Instruction Tuning With Loss Over Instructions

Instruction tuning plays a crucial role in shaping the outputs of language models (LMs) to desired styles. In this work, we propose a simple yet effective method, Instruction Modelling (IM), which trains LMs by applying a loss function to the instruction and prompt part rather than solely to the output part. Through experiments across 21 diverse benchmarks, we show that, in many scenarios, IM can effectively improve the LM performance on both NLP tasks (e.g., MMLU, TruthfulQA, and HumanEval) and open-ended generation benchmarks (e.g., MT-Bench and AlpacaEval). Remarkably, in the most advantageous case, IM boosts model performance on AlpacaEval 1.0 by over 100%. We identify two key factors influencing the effectiveness of IM: (1) The ratio between instruction length and output length in the training data; and (2) The number of training examples. We observe that IM is especially beneficial when trained on datasets with lengthy instructions paired with brief outputs, or under the Superficial Alignment Hypothesis (SAH) where a small amount of training examples are used for instruction tuning. Further analysis substantiates our hypothesis that our improvement can be attributed to reduced overfitting to instruction tuning datasets. It is worth noting that we are not proposing \ours as a replacement for current fine-tuning processes. Instead, our work aims to provide practical guidance for instruction tuning LMs, especially in low-resource scenarios.

cs.CL↗

arXiv2024

Transparent and Scrutable Recommendations Using Natural Language User Profiles

Recent state-of-the-art recommender systems predominantly rely on either implicit or explicit feedback from users to suggest new items. While effective in recommending novel options, many recommender systems often use uninterpretable embeddings to represent user preferences. This lack of transparency not only limits user understanding of why certain items are suggested but also reduces the user's ability to scrutinize and modify their preferences, thereby affecting their ability to receive a list of preferred recommendations. Given the recent advances in Large Language Models (LLMs), we investigate how a properly crafted prompt can be used to summarize a user's preferences from past reviews and recommend items based only on language-based preferences. In particular, we study how LLMs can be prompted to generate a natural language (NL) user profile that holistically describe a user's preferences. These NL profiles can then be leveraged to fine-tune a LLM using only NL profiles to make transparent and scrutable recommendations. Furthermore, we validate the scrutability of our user profile-based recommender by investigating the impact on recommendation changes after editing NL user profiles. According to our evaluations of the model's rating prediction performance on two benchmarking rating prediction datasets, we observe that this novel approach maintains a performance level on par with established recommender systems in a warm-start setting. With a systematic analysis into the effect of updating user profiles and system prompts, we show the advantage of our approach in easier adjustment of user preferences and a greater autonomy over users' received recommendations.

cs.IR↗

arXiv2024

Collaborative User Prompt for Personalized Generative Recommendation

Large Language Models (LLMs) have become powerful foundations for generative recommender systems, framing recommendation tasks as text generation tasks. However, existing generative recommendation methods often rely on discrete ID-based prompts or task-specific soft prompts, which overlook the valuable collaborative signals shared among users with similar interests. To address this limitation, this paper presents a compositional framework that integrates a user's individual preferences with collective preferences from similar users to build personalized soft prompts. Specifically, an attention-based mechanism fuses embeddings from users with similar interests, creating a richer representation that captures multiple facets of user preferences. This design dynamically emphasizes shared interests while preserving individual user preferences. Experiments on three real-world datasets demonstrate the effectiveness of the proposed approach across sequential recommendation, top-n recommendation, and explanation generation tasks, underscoring the advantages of incorporating collaborative signals through an attention-based compositional strategy.

cs.IR↗

arXiv2024

Do Sentence Transformers Learn Quasi-Geospatial Concepts from General Text?

Sentence transformers are language models designed to perform semantic search. This study investigates the capacity of sentence transformers, fine-tuned on general question-answering datasets for asymmetric semantic search, to associate descriptions of human-generated routes across Great Britain with queries often used to describe hiking experiences. We find that sentence transformers have some zero-shot capabilities to understand quasi-geospatial concepts, such as route types and difficulty, suggesting their potential utility for routing recommendation systems.

cs.CL↗

arXiv2024

Application of Zone Method based Physics-Informed Neural Networks in Reheating Furnaces

Foundation Industries (FIs) constitute glass, metals, cement, ceramics, bulk chemicals, paper, steel, etc. and provide crucial, foundational materials for a diverse set of economically relevant industries: automobiles, machinery, construction, household appliances, chemicals, etc. Reheating furnaces within the manufacturing chain of FIs are energy-intensive. Accurate and real-time prediction of underlying temperatures in reheating furnaces has the potential to reduce the overall heating time, thereby controlling the energy consumption for achieving the Net-Zero goals in FIs. In this paper, we cast this prediction as a regression task and explore neural networks due to their inherent capability of being effective and efficient, given adequate data. However, due to the infeasibility of achieving good-quality real data in scenarios like reheating furnaces, classical Hottel's zone method based computational model has been used to generate data for model training. To further enhance the Out-Of-Distribution generalization capability of the trained model, we propose a Physics-Informed Neural Network (PINN) by incorporating prior physical knowledge using a set of novel Energy-Balance regularizers.

cs.LG↗

arXiv2024

DePT: Decomposed Prompt Tuning for Parameter-Efficient Fine-tuning

Prompt tuning (PT), where a small amount of trainable soft (continuous) prompt vectors is affixed to the input of language models (LM), has shown promising results across various tasks and models for parameter-efficient fine-tuning (PEFT). PT stands out from other PEFT approaches because it maintains competitive performance with fewer trainable parameters and does not drastically scale up its parameters as the model size expands. However, PT introduces additional soft prompt tokens, leading to longer input sequences, which significantly impacts training and inference time and memory usage due to the Transformer's quadratic complexity. Particularly concerning for Large Language Models (LLMs) that face heavy daily querying. To address this issue, we propose Decomposed Prompt Tuning (DePT), which decomposes the soft prompt into a shorter soft prompt and a pair of low-rank matrices that are then optimised with two different learning rates. This allows DePT to achieve better performance while saving substantial memory and time costs compared to vanilla PT and its variants, without changing trainable parameter sizes. Through extensive experiments on 23 natural language processing (NLP) and vision-language (VL) tasks, we demonstrate that DePT outperforms state-of-the-art PEFT approaches, including the full fine-tuning baseline, in some scenarios. Additionally, we empirically show that DEPT grows more efficient as the model size increases. Our further study reveals that DePT integrates seamlessly with parameter-efficient transfer learning in the few-shot learning setting and highlights its adaptability to various model architectures and sizes.

cs.CL↗

arXiv2023

Self Contrastive Learning for Session-based Recommendation

Session-based recommendation, which aims to predict the next item of users' interest as per an existing sequence interaction of items, has attracted growing applications of Contrastive Learning (CL) with improved user and item representations. However, these contrastive objectives: (1) serve a similar role as the cross-entropy loss while ignoring the item representation space optimisation; and (2) commonly require complicated modelling, including complex positive/negative sample constructions and extra data augmentation. In this work, we introduce Self-Contrastive Learning (SCL), which simplifies the application of CL and enhances the performance of state-of-the-art CL-based recommendation techniques. Specifically, SCL is formulated as an objective function that directly promotes a uniform distribution among item representations and efficiently replaces all the existing contrastive objective components of state-of-the-art models. Unlike previous works, SCL eliminates the need for any positive/negative sample construction or data augmentation, leading to enhanced interpretability of the item representation space and facilitating its extensibility to existing recommender systems. Through experiments on three benchmark datasets, we demonstrate that SCL consistently improves the performance of state-of-the-art models with statistical significance. Notably, our experiments show that SCL improves the performance of two best-performing models by 8.2% and 9.5% in P@10 (Precision) and 9.9% and 11.2% in MRR@10 (Mean Reciprocal Rank) on average across different benchmarks. Additionally, our analysis elucidates the improvement in terms of alignment and uniformity of representations, as well as the effectiveness of SCL with a low computational cost.

cs.IR↗

arXiv2023

Unlocking the Potential of User Feedback: Leveraging Large Language Model as User Simulator to Enhance Dialogue System

Dialogue systems and large language models (LLMs) have gained considerable attention. However, the direct utilization of LLMs as task-oriented dialogue (TOD) models has been found to underperform compared to smaller task-specific models. Nonetheless, it is crucial to acknowledge the significant potential of LLMs and explore improved approaches for leveraging their impressive abilities. Motivated by the goal of leveraging LLMs, we propose an alternative approach called User-Guided Response Optimization (UGRO) to combine it with a smaller TOD model. This approach uses LLM as annotation-free user simulator to assess dialogue responses, combining them with smaller fine-tuned end-to-end TOD models. By utilizing the satisfaction feedback generated by LLMs, UGRO further optimizes the supervised fine-tuned TOD model. Specifically, the TOD model takes the dialogue history as input and, with the assistance of the user simulator's feedback, generates high-satisfaction responses that meet the user's requirements. Through empirical experiments on two TOD benchmarks, we validate the effectiveness of our method. The results demonstrate that our approach outperforms previous state-of-the-art (SOTA) results.

cs.CL↗

arXiv2023

Lexical Entrainment for Conversational Systems

Conversational agents have become ubiquitous in assisting with daily tasks, and are expected to possess human-like features. One such feature is lexical entrainment (LE), a phenomenon in which speakers in human-human conversations tend to naturally and subconsciously align their lexical choices with those of their interlocutors, leading to more successful and engaging conversations. As an example, if a digital assistant replies 'Your appointment for Jinling Noodle Pub is at 7 pm' to the question 'When is my reservation for Jinling Noodle Bar today?', it may feel as though the assistant is trying to correct the speaker, whereas a response of 'Your reservation for Jinling Noodle Bar is at 7 pm' would likely be perceived as more positive. This highlights the importance of LE in establishing a shared terminology for maximum clarity and reducing ambiguity in conversations. However, we demonstrate in this work that current response generation models do not adequately address this crucial humanlike phenomenon. To address this, we propose a new dataset, named MULTIWOZ-ENTR, and a measure for LE for conversational systems. Additionally, we suggest a way to explicitly integrate LE into conversational systems with two new tasks, a LE extraction task and a LE generation task. We also present two baseline approaches for the LE extraction task, which aim to detect LE expressions from dialogue contexts.

cs.CL↗

arXiv2023

Don't Stop Pretraining? Make Prompt-based Fine-tuning Powerful Learner

Language models (LMs) trained on vast quantities of unlabelled data have greatly advanced the field of natural language processing (NLP). In this study, we re-visit the widely accepted notion in NLP that continued pre-training LMs on task-related texts improves the performance of fine-tuning (FT) in downstream tasks. Through experiments on eight single-sentence tasks and eight sentence-pair tasks in both semi-supervised and fully-supervised settings, we find that conventional continued pre-training does not consistently provide benefits and can even be detrimental for sentence-pair tasks or when prompt-based FT is used. To tackle these issues, we propose Prompt-based Continued Pre-training (PCP), which combines the idea of instruction tuning with conventional continued pre-training. Our approach aims to improve the performance of prompt-based FT by presenting both task-related texts and prompt templates to LMs through unsupervised pre-training objectives before fine-tuning for the target task. Our empirical evaluations on 21 benchmarks demonstrate that the PCP consistently improves the performance of state-of-the-art prompt-based FT approaches (up to 20.1% absolute) in both semi-supervised and fully-supervised settings, even with only hundreds of unlabelled examples. Additionally, prompt-based FT with the PCP outperforms state-of-the-art semi-supervised approaches with greater simplicity, eliminating the need for an iterative process and extra data augmentation. Our further analysis explores the performance lower bound of the PCP and reveals that the advantages of PCP persist across different sizes of models and datasets.

cs.CL↗

arXiv2023

Rethink the Effectiveness of Text Data Augmentation: An Empirical Analysis

In recent years, language models (LMs) have made remarkable progress in advancing the field of natural language processing (NLP). However, the impact of data augmentation (DA) techniques on the fine-tuning (FT) performance of these LMs has been a topic of ongoing debate. In this study, we evaluate the effectiveness of three different FT methods in conjugation with back-translation across an array of 7 diverse NLP tasks, including classification and regression types, covering single-sentence and sentence-pair tasks. Contrary to prior assumptions that DA does not contribute to the enhancement of LMs' FT performance, our findings reveal that continued pre-training on augmented data can effectively improve the FT performance of the downstream tasks. In the most favourable case, continued pre-training improves the performance of FT by more than 10% in the few-shot learning setting. Our finding highlights the potential of DA as a powerful tool for bolstering LMs' performance.

cs.CL↗