A. B. Siddique

M. H. Maqbool, Umar Farooq, Adib Mosharrof et al.

Recommender systems have become ubiquitous in our digital lives, from recommending products on e-commerce websites to suggesting movies and music on streaming platforms. Existing recommendation datasets, such as Amazon Product Reviews and MovieLens, greatly facilitated the research and development of recommender systems in their respective domains. While the number of mobile users and applications (aka apps) has increased exponentially over the past decade, research in mobile app recommender systems has been significantly constrained, primarily due to the lack of high-quality benchmark datasets, as opposed to recommendations for products, movies, and news. To facilitate research for app recommendation systems, we introduce a large-scale dataset, called MobileRec. We constructed MobileRec from users' activity on the Google play store. MobileRec contains 19.3 million user interactions (i.e., user reviews on apps) with over 10K unique apps across 48 categories. MobileRec records the sequential activity of a total of 0.7 million distinct users. Each of these users has interacted with no fewer than five distinct apps, which stands in contrast to previous datasets on mobile apps that recorded only a single interaction per user. Furthermore, MobileRec presents users' ratings as well as sentiments on installed apps, and each app contains rich metadata such as app name, category, description, and overall rating, among others. We demonstrate that MobileRec can serve as an excellent testbed for app recommendation through a comparative study of several state-of-the-art recommendation approaches. The quantitative results can act as a baseline for other researchers to compare their results against. The MobileRec dataset is available at https://huggingface.co/datasets/recmeapp/mobilerec.

Jainam Dhruva, Yousaf Raza, A. B. Siddique et al.

Accurate short-term forecasting of residential energy load and indoor temperature is essential for home energy management systems, grid-level demand response, and community energy efficiency efforts. Domain adaptation and transfer learning have shown promise for improving forecasting accuracy under data heterogeneity and scarcity commonly seen in residential settings. However, progress is limited by the lack of comprehensive residential datasets: existing benchmarks are narrow in target coverage and rarely support structured cross-domain evaluation. We introduce RESCAST-100K, a large-scale residential forecasting benchmark for studying cross-domain generalization. It provides a configuration-driven interface that instantiates source and target domains along interpretable axes, including geography, climate zone, wall construction, and heating equipment, enabling systematic evaluation of transfer learning, domain adaptation, and zero-shot generalization under controlled domain shifts. The benchmark covers approximately 100,000 EnergyPlus-simulated U.S. homes derived from ResStock, with 15-minute time series for three coupled targets per home: total load, HVAC load, and indoor temperature. These are paired with weather channels, HVAC setpoints, and over 40 static building covariates. RESCAST-100K also integrates five real-world residential datasets under a unified schema, supporting sim-to-real evaluation on the same tasks. We benchmark recurrent, attention-based, and MLP-mixer architectures for zero-shot performance across domains, missing-input conditions, and forecasting tasks. Cross-attention and MLP-mixer models consistently outperform recurrent and classical transformer baselines under domain shift. RESCAST-100K is intended to aid the machine learning and building analytics communities advance cross-domain residential forecasting at home, community, and grid scale.

Hammad Rizwan, Muhammad Umair Haider, Nishant Subramani et al.

Representations extracted from large language models (LLMs) play an important role in many downstream applications. However, the structure of these representations is often influenced by lexical overlap rather than semantic content. Our understanding of the relationship between this lexical influence and semantic content, and its implications for downstream tasks, remains limited. In this work, we investigate representations to quantify the effect of lexical overlap relative to semantic content. We consider several adversarial semantic stress tests and further connect our findings to the information theory perspective. We find that lexical influence extends across the depth of models, consistently across architectures, training regimes, and objective functions, including the models trained for semantic similarity. Moreover, we observe a mid-depth region in which both lexical and semantic signals degrade simultaneously, indicating a transitional regime where representations are poor for both surface form and meaning. We further demonstrate the effect of lexical influence on downstream uses of LLMs using summarization and model editing as a case study.

Ayesha S. Dina, A. B. Siddique, D. Manivannan

Attacks on computer networks have increased significantly in recent days, due in part to the availability of sophisticated tools for launching such attacks as well as thriving underground cyber-crime economy to support it. Over the past several years, researchers in academia and industry used machine learning (ML) techniques to design and implement Intrusion Detection Systems (IDSes) for computer networks. Many of these researchers used datasets collected by various organizations to train ML models for predicting intrusions. In many of the datasets used in such systems, data are imbalanced (i.e., not all classes have equal amount of samples). With unbalanced data, the predictive models developed using ML algorithms may produce unsatisfactory classifiers which would affect accuracy in predicting intrusions. Traditionally, researchers used over-sampling and under-sampling for balancing data in datasets to overcome this problem. In this work, in addition to over-sampling, we also use a synthetic data generation method, called Conditional Generative Adversarial Network (CTGAN), to balance data and study their effect on various ML classifiers. To the best of our knowledge, no one else has used CTGAN to generate synthetic samples to balance intrusion detection datasets. Based on extensive experiments using a widely used dataset NSL-KDD, we found that training ML models on dataset balanced with synthetic samples generated by CTGAN increased prediction accuracy by up to $8\%$, compared to training the same ML models over unbalanced data. Our experiments also show that the accuracy of some ML models trained over data balanced with random over-sampling decline compared to the same ML models trained over unbalanced data.

A. B. Siddique, M. H. Maqbool, Kshitija Taywade et al.

Task-oriented dialog systems enable users to accomplish tasks using natural language. State-of-the-art systems respond to users in the same way regardless of their personalities, although personalizing dialogues can lead to higher levels of adoption and better user experiences. Building personalized dialog systems is an important, yet challenging endeavor and only a handful of works took on the challenge. Most existing works rely on supervised learning approaches and require laborious and expensive labeled training data for each user profile. Additionally, collecting and labeling data for each user profile is virtually impossible. In this work, we propose a novel framework, P-ToD, to personalize task-oriented dialog systems capable of adapting to a wide range of user profiles in an unsupervised fashion using a zero-shot generalizable reward function. P-ToD uses a pre-trained GPT-2 as a backbone model and works in three phases. Phase one performs task-specific training. Phase two kicks off unsupervised personalization by leveraging the proximal policy optimization algorithm that performs policy gradients guided by the zero-shot generalizable reward function. Our novel reward function can quantify the quality of the generated responses even for unseen profiles. The optional final phase fine-tunes the personalized model using a few labeled training examples. We conduct extensive experimental analysis using the personalized bAbI dialogue benchmark for five tasks and up to 180 diverse user profiles. The experimental results demonstrate that P-ToD, even when it had access to zero labeled examples, outperforms state-of-the-art supervised personalization models and achieves competitive performance on BLEU and ROUGE metrics when compared to a strong fully-supervised GPT-2 baseline

Daniel Liu, Krishna Upadhyay, Vinaik Chhetri et al.

The rapid emergence of multi-agent AI systems (MAS), including LangChain, CrewAI, and AutoGen, has shaped how large language model (LLM) applications are developed and orchestrated. However, little is known about how these systems evolve and are maintained in practice. This paper presents the first large-scale empirical study of open-source MAS, analyzing over 42K unique commits and over 4.7K resolved issues across eight leading systems. Our analysis identifies three distinct development profiles: sustained, steady, and burst-driven. These profiles reflect substantial variation in ecosystem maturity. Perfective commits constitute 40.8% of all changes, suggesting that feature enhancement is prioritized over corrective maintenance (27.4%) and adaptive updates (24.3%). Data about issues shows that the most frequent concerns involve bugs (22%), infrastructure (14%), and agent coordination challenges (10%). Issue reporting also increased sharply across all frameworks starting in 2023. Median resolution times range from under one day to about two weeks, with distributions skewed toward fast responses but a minority of issues requiring extended attention. These results highlight both the momentum and the fragility of the current ecosystem, emphasizing the need for improved testing infrastructure, documentation quality, and maintenance practices to ensure long-term reliability and sustainability.

Sing Hieng Wong, Hassan Sajjad, A. B. Siddique

Large language models (LLMs) show strong multilingual capabilities, yet reliably controlling the language of their outputs remains difficult. Representation-level steering addresses this by adding language-specific vectors to model activations at inference time, but identifying language-specific directions in the residual stream often relies on multilingual or parallel data that can be expensive to obtain. Sparse autoencoders (SAEs) decompose residual activations into interpretable, sparse feature directions and offer a natural basis for this search, yet existing SAE-based approaches face the same data constraint. We introduce LangFIR (Language Feature Identification via Random-token Filtering), a method that discovers language-specific SAE features using only a small amount of monolingual data and random-token sequences. Many SAE features consistently activated by target-language inputs do not encode language identity. Random-token sequences surface these language-agnostic features, allowing LangFIR to filter them out and isolate a sparse set of language-specific features. We show that these features are extremely sparse, highly selective for their target language, and causally important: directional ablation increases cross-entropy loss only for the corresponding language. Using these features to construct steering vectors for multilingual generation control, LangFIR achieves the best average accuracy BLEU across three models (Gemma 3 1B, Gemma 3 4B, and Llama 3.1 8B), three datasets, and twelve target languages, outperforming the strongest monolingual baseline by up to and surpassing methods that rely on parallel data. Our results suggest that language identity in multilingual LLMs is localized in a sparse set of feature directions discoverable with monolingual data. Code is available at https://anonymous.4open.science/r/LangFIR-C0F5/.

Adib Mosharrof, M. H. Maqbool, A. B. Siddique

Task-oriented dialog systems empower users to accomplish their goals by facilitating intuitive and expressive natural language interactions. State-of-the-art approaches in task-oriented dialog systems formulate the problem as a conditional sequence generation task and fine-tune pre-trained causal language models in the supervised setting. This requires labeled training data for each new domain or task, and acquiring such data is prohibitively laborious and expensive, thus making it a bottleneck for scaling systems to a wide range of domains. To overcome this challenge, we introduce a novel Zero-Shot generalizable end-to-end Task-oriented Dialog system, ZS-ToD, that leverages domain schemas to allow for robust generalization to unseen domains and exploits effective summarization of the dialog history. We employ GPT-2 as a backbone model and introduce a two-step training process where the goal of the first step is to learn the general structure of the dialog data and the second step optimizes the response generation as well as intermediate outputs, such as dialog state and system actions. As opposed to state-of-the-art systems that are trained to fulfill certain intents in the given domains and memorize task-specific conversational patterns, ZS-ToD learns generic task-completion skills by comprehending domain semantics via domain schemas and generalizing to unseen domains seamlessly. We conduct an extensive experimental evaluation on SGD and SGD-X datasets that span up to 20 unique domains and ZS-ToD outperforms state-of-the-art systems on key metrics, with an improvement of +17% on joint goal accuracy and +5 on inform. Additionally, we present a detailed ablation study to demonstrate the effectiveness of the proposed components and training mechanism

Muhammad Umair Haider, Umar Farooq, A. B. Siddique et al.

Language Models (LMs) have shown their application for tasks pertinent to code and several code~LMs have been proposed recently. The majority of the studies in this direction only focus on the improvements in performance of the LMs on different benchmarks, whereas LMs are considered black boxes. Besides this, a handful of works attempt to understand the role of attention layers in the code~LMs. Nonetheless, feed-forward layers remain under-explored which consist of two-thirds of a typical transformer model's parameters. In this work, we attempt to gain insights into the inner workings of code language models by examining the feed-forward layers. To conduct our investigations, we use two state-of-the-art code~LMs, Codegen-Mono and Ploycoder, and three widely used programming languages, Java, Go, and Python. We focus on examining the organization of stored concepts, the editability of these concepts, and the roles of different layers and input context size variations for output generation. Our empirical findings demonstrate that lower layers capture syntactic patterns while higher layers encode abstract concepts and semantics. We show concepts of interest can be edited within feed-forward layers without compromising code~LM performance. Additionally, we observe initial layers serve as ``thinking'' layers, while later layers are crucial for predicting subsequent code tokens. Furthermore, we discover earlier layers can accurately predict smaller contexts, but larger contexts need critical later layers' contributions. We anticipate these findings will facilitate better understanding, debugging, and testing of code~LMs.

Adib Mosharrof, Moghis Fereidouni, A. B. Siddique

Slot filling is one of the critical tasks in modern conversational systems. The majority of existing literature employs supervised learning methods, which require labeled training data for each new domain. Zero-shot learning and weak supervision approaches, among others, have shown promise as alternatives to manual labeling. Nonetheless, these learning paradigms are significantly inferior to supervised learning approaches in terms of performance. To minimize this performance gap and demonstrate the possibility of open-domain slot filling, we propose a Self-supervised Co-training framework, called SCot, that requires zero in-domain manually labeled training examples and works in three phases. Phase one acquires two sets of complementary pseudo labels automatically. Phase two leverages the power of the pre-trained language model BERT, by adapting it for the slot filling task using these sets of pseudo labels. In phase three, we introduce a self-supervised cotraining mechanism, where both models automatically select highconfidence soft labels to further improve the performance of the other in an iterative fashion. Our thorough evaluations show that SCot outperforms state-of-the-art models by 45.57% and 37.56% on SGD and MultiWoZ datasets, respectively. Moreover, our proposed framework SCot achieves comparable performance when compared to state-of-the-art fully supervised models.

Moghis Fereidouni, Adib Mosharrof, A. B. Siddique

The development of agents powered by large language models (LLMs) to accomplish complex high-level user intents, has attracted significant attention recently. However, employing LLMs with billions of parameters (e.g., GPT-4) may incur substantial costs on top of handcrafting extensive prompts. To address this, we introduce a Grounded Language Agent for Intelligent Web Interactions, named GLAINTEL. GLAINTEL employs Flan-T5 as its backbone and is flexible in training in various settings: unsupervised learning, supervised learning, and unsupervised domain adaptation. Specifically, we tackle both the challenge of learning without human demonstrations and the opportunity to leverage human demonstrations effectively when those are available. Additionally, we explore unsupervised domain adaptation for cases where demonstrations are limited to a specific domain. Experimental evaluations across diverse setups demonstrate the effectiveness of GLAINTEL in unsupervised settings, outperforming in-context learning-based approaches that employ larger models with up to 540 billion parameters. Surprisingly, behavioral cloning-based methods that straightforwardly use human demonstrations do not outperform unsupervised variants of GLAINTEL. Additionally, we show that combining human demonstrations with reinforcement learning-based training yields results comparable to methods utilizing GPT-4. The code is available at: https://github.com/MultifacetedNLP/WebAgents-Unsupervised.

Moshood A. Fakorede, Krishna Upadhyay, A. B. Siddique et al.

Large language models (LLMs) have shown strong performance on automated software engineering tasks, yet existing benchmarks focus primarily on general-purpose libraries or web applications, leaving mobile application development largely unexplored despite its strict platform constraints, framework-driven lifecycles, and complex platform API interactions. We introduce MobileDev-Bench, a benchmark comprising 384 real-world issue-resolution tasks collected from 18 production mobile applications spanning Android Native (Java/Kotlin), React Native (TypeScript), and Flutter (Dart). Each task pairs an authentic developer-reported issue with executable test patches, enabling fully automated validation of model-generated fixes within mobile build environments. The benchmark exhibits substantial patch complexity: fixes modify 12.5 files and 324.9 lines on average, and 35.7% of instances require coordinated changes across multiple artifact types, such as source and manifest files. Evaluation of four state-of-the-art code-capable LLMs, GPT- 5.2, Claude Sonnet 4.5, Gemini Flash 2.5, and Qwen3-Coder, yields low end-to-end resolution rates of 3.39%-5.21%, revealing significant performance gaps compared to prior benchmarks. Further analysis reveals systematic failure modes, with fault localization across multi-file and multi-artifact changes emerging as the primary bottleneck.

Adib Mosharrof, A. B. Siddique

Task-oriented dialogue (TOD) systems enable users to achieve their goals through natural language interactions. Traditionally, these systems have relied on turn-level manually annotated metadata, such as dialogue states and policy annotations, which are expensive, time-consuming, and often inconsistent or error-prone. This dependence limits the potential to leverage vast amounts of readily available conversational data for training TOD systems. Additionally, a critical challenge in TOD system design is determining when and how to access and integrate information from external sources. Current approaches typically expect this information to be provided alongside the dialogue context, rather than learning to identify and retrieve it autonomously. While pre-trained large language models (LLMs) have been used to develop TOD systems, their potential to train such systems without laborious annotations remains largely unexplored. This work employs multi-task instruction fine-tuning to create more efficient and scalable TOD systems that can effectively leverage natural language conversational data without manual annotations, while autonomously managing external information retrieval. Our extensive experimental evaluations, using three diverse TOD datasets and three LLMs of varying sizes, demonstrate that our approach can generalize to new, unseen domains. Notably, our approach outperforms both state-of-the-art models trained on annotated data and billion-scale parameter off-the-shelf ChatGPT models.

M. H. Maqbool, Moghis Fereidouni, Umar Farooq et al.

The mobile app market has expanded exponentially, offering millions of apps with diverse functionalities, yet research in mobile app recommendation remains limited. Traditional sequential recommender systems utilize the order of items in users' historical interactions to predict the next item for the users. Position embeddings, well-established in transformer-based architectures for natural language processing tasks, effectively distinguish token positions in sequences. In sequential recommendation systems, position embeddings can capture the order of items in a user's historical interaction sequence. Nevertheless, this ordering does not consider the time elapsed between two interactions of the same user (e.g., 1 day, 1 week, 1 month), referred to as "user rhythm". In mobile app recommendation datasets, the time between consecutive user interactions is notably longer compared to other domains like movies, posing significant challenges for sequential recommender systems. To address this phenomenon in the mobile app domain, we introduce INTERPOS, an Interaction Rhythm Guided Positional Morphing strategy for autoregressive mobile app recommender systems. INTERPOS incorporates rhythm-guided position embeddings, providing a more comprehensive representation that considers both the sequential order of interactions and the temporal gaps between them. This approach enables a deep understanding of users' rhythms at a fine-grained level, capturing the intricacies of their interaction patterns over time. We propose three strategies to incorporate the morphed positional embeddings in two transformer-based sequential recommendation system architectures. Our extensive evaluations show that INTERPOS outperforms state-of-the-art models using 7 mobile app recommendation datasets on NDCG@K and HIT@K metrics. The source code of INTERPOS is available at https://github.com/dlgrad/INTERPOS.

Adib Mosharrof, Moghis Fereidouni, A. B. Siddique

Traditional task-oriented dialog (ToD) systems rely heavily on labor-intensive turn-level annotations, such as dialogue states and policy labels, for training. This work explores whether large language models (LLMs) can be fine-tuned solely on natural language dialogs to perform ToD tasks, without requiring such annotations. We evaluate their ability to generalize to unseen domains and compare their performance with models trained on fully annotated data. Through extensive experiments with three open-source LLMs of varying sizes and two diverse ToD datasets, we find that models fine-tuned without turn-level annotations generate coherent and contextually appropriate responses. However, their task completion performance - measured by accurate execution of API calls - remains suboptimal, with the best models achieving only around 53% success in unseen domains. To improve task completion, we propose ZeroToD, a framework that incorporates a schema augmentation mechanism to enhance API call accuracy and overall task completion rates, particularly in out-of-domain settings. We also compare ZeroToD with fine-tuning-free alternatives, such as prompting off-the-shelf LLMs, and find that our framework enables smaller, fine-tuned models that outperform large-scale proprietary LLMs in task completion. Additionally, a human study evaluating informativeness, fluency, and task completion confirms our empirical findings. These findings suggest the feasibility of developing cost-effective, scalable, and zero-shot generalizable ToD systems for real-world applications.

Muhammad Umair Haider, Hammad Rizwan, Hassan Sajjad et al.

Circuit discovery aims to identify minimal subnetworks that are responsible for specific behaviors in large language models (LLMs). Existing approaches primarily rely on iterative edge pruning, which is computationally expensive and limited to coarse-grained units such as attention heads or MLP blocks, overlooking finer structures like individual neurons. We propose a node-level pruning framework for circuit discovery that addresses both scalability and granularity limitations. Our method introduces learnable masks across multiple levels of granularity, from entire blocks to individual neurons, within a unified optimization objective. Granularity-specific sparsity penalties guide the pruning process, allowing a comprehensive compression in a single fine-tuning run. Empirically, our approach identifies circuits that are smaller in nodes than those discovered by prior methods; moreover, we demonstrate that many neurons deemed important by coarse methods are actually irrelevant, while still maintaining task performance. Furthermore, our method has a significantly lower memory footprint, 5-10x, as it does not require keeping intermediate activations in the memory to work.

Muhammad Umair Haider, Hammad Rizwan, Hassan Sajjad et al.

Interpreting the internal mechanisms of large language models (LLMs) is crucial for improving their trustworthiness and utility. Prior work has primarily focused on mapping individual neurons to discrete semantic concepts. However, such mappings struggle to handle the inherent polysemanticity in LLMs, where individual neurons encode multiple, distinct concepts. Through a comprehensive analysis of both encoder and decoder-based LLMs across diverse datasets, we observe that even highly salient neurons, identified via various attribution techniques for specific semantic concepts, consistently exhibit polysemantic behavior. Importantly, activation magnitudes for fine-grained concepts follow distinct, often Gaussian-like distributions with minimal overlap. This observation motivates a shift from neuron attribution to range-based interpretation. We hypothesize that interpreting and manipulating neuron activation ranges would enable more precise interpretability and targeted interventions in LLMs. To validate our hypothesis, we introduce NeuronLens, a novel range-based interpretation and manipulation framework that provides a finer view of neuron activation distributions to localize concept attribution within a neuron. Extensive empirical evaluations demonstrate that NeuronLens significantly reduces unintended interference, while maintaining precise manipulation of targeted concepts, outperforming neuron attribution.

Moghis Fereidouni, Muhammad Umair Haider, Peizhong Ju et al.

A key barrier to interpreting large language models is polysemanticity, where neurons activate for multiple unrelated concepts. Sparse autoencoders (SAEs) have been proposed to mitigate this issue by transforming dense activations into sparse, more interpretable features. While prior work suggests that SAEs promote monosemanticity, no quantitative comparison has examined how concept activation distributions differ between SAEs and their base models. This paper provides the first systematic evaluation of SAEs against base models through activation distribution lens. We introduce a fine-grained concept separability score based on the Jensen-Shannon distance, which captures how distinctly a neuron's activation distributions vary across concepts. Using two large language models (Gemma-2-2B and DeepSeek-R1) and multiple SAE variants across five datasets (including word-level and sentence-level), we show that SAEs reduce polysemanticity and achieve higher concept separability. To assess practical utility, we evaluate concept-level interventions using two strategies: full neuron masking and partial suppression. We find that, compared to base models, SAEs enable more precise concept-level control when using partial suppression. Building on this, we propose Attenuation via Posterior Probabilities (APP), a new intervention method that uses concept-conditioned activation distributions for targeted suppression. APP achieves the smallest perplexity increase while remaining highly effective at concept removal.

Vinaik Chhetri, Krishna Upadhyay, A. B. Siddique et al.

Artificial Intelligence (AI)-powered features have rapidly proliferated across mobile apps in various domains, including productivity, education, entertainment, and creativity. However, how users perceive, evaluate, and critique these AI features remains largely unexplored, primarily due to the overwhelming volume of user feedback. In this work, we present the first comprehensive, large-scale study of user feedback on AI-powered mobile apps, leveraging a curated dataset of 292 AI-driven apps across 14 categories with 894K AI-specific reviews from Google Play. We develop and validate a multi-stage analysis pipeline that begins with a human-labeled benchmark and systematically evaluates large language models (LLMs) and prompting strategies. Each stage, including review classification, aspect-sentiment extraction, and clustering, is validated for accuracy and consistency. Our pipeline enables scalable, high-precision analysis of user feedback, extracting over one million aspect-sentiment pairs clustered into 18 positive and 15 negative user topics. Our analysis reveals that users consistently focus on a narrow set of themes: positive comments emphasize productivity, reliability, and personalized assistance, while negative feedback highlights technical failures (e.g., scanning and recognition), pricing concerns, and limitations in language support. Our pipeline surfaces both satisfaction with one feature and frustration with another within the same review. These fine-grained, co-occurring sentiments are often missed by traditional approaches that treat positive and negative feedback in isolation or rely on coarse-grained analysis. To this end, our approach provides a more faithful reflection of the real-world user experiences with AI-powered apps. Category-aware analysis further uncovers both universal drivers of satisfaction and domain-specific frustrations.

Moghis Fereidouni, Md Sajid Ahmed, Adib Mosharrof et al.

Task-oriented dialog (TOD) systems facilitate users in accomplishing complex, multi-turn tasks through natural language. While traditional approaches rely on extensive fine-tuning and annotated data for each domain, instruction-tuned large language models (LLMs) offer a more flexible alternative. However, LLMs struggle to reliably handle multi-turn task completion, particularly with accurately generating API calls and adapting to new domains without explicit demonstrations. To address these challenges, we propose RealTOD, a novel framework that enhances TOD systems through prompt chaining and fine-grained feedback mechanisms. Prompt chaining enables zero-shot domain adaptation via a two-stage prompting strategy, eliminating the need for human-curated demonstrations. Meanwhile, the fine-grained feedback mechanism improves task completion by verifying API calls against domain schemas and providing precise corrective feedback when errors are detected. We conduct extensive experiments on the SGD and BiTOD benchmarks using four LLMs. RealTOD improves API accuracy, surpassing AutoTOD by 37.74% on SGD and SimpleTOD by 11.26% on BiTOD. Human evaluations further confirm that LLMs integrated with RealTOD achieve superior task completion, fluency, and informativeness compared to existing methods.

A. B. Siddique, Fuad Jamour, Luxun Xu et al.

Identifying user intents from natural language utterances is a crucial step in conversational systems that has been extensively studied as a supervised classification problem. However, in practice, new intents emerge after deploying an intent detection model. Thus, these models should seamlessly adapt and classify utterances with both seen and unseen intents -- unseen intents emerge after deployment and they do not have training data. The few existing models that target this setting rely heavily on the scarcely available training data and overfit to seen intents data, resulting in a bias to misclassify utterances with unseen intents into seen ones. We propose RIDE: an intent detection model that leverages commonsense knowledge in an unsupervised fashion to overcome the issue of training data scarcity. RIDE computes robust and generalizable relationship meta-features that capture deep semantic relationships between utterances and intent labels; these features are computed by considering how the concepts in an utterance are linked to those in an intent label via commonsense knowledge. Our extensive experimental analysis on three widely-used intent detection benchmarks shows that relationship meta-features significantly increase the accuracy of detecting both seen and unseen intents and that RIDE outperforms the state-of-the-art model for unseen intents.

A. B. Siddique, Fuad Jamour, Vagelis Hristidis

Slot filling is identifying contiguous spans of words in an utterance that correspond to certain parameters (i.e., slots) of a user request/query. Slot filling is one of the most important challenges in modern task-oriented dialog systems. Supervised learning approaches have proven effective at tackling this challenge, but they need a significant amount of labeled training data in a given domain. However, new domains (i.e., unseen in training) may emerge after deployment. Thus, it is imperative that these models seamlessly adapt and fill slots from both seen and unseen domains -- unseen domains contain unseen slot types with no training data, and even seen slots in unseen domains are typically presented in different contexts. This setting is commonly referred to as zero-shot slot filling. Little work has focused on this setting, with limited experimental evaluation. Existing models that mainly rely on context-independent embedding-based similarity measures fail to detect slot values in unseen domains or do so only partially. We propose a new zero-shot slot filling neural model, LEONA, which works in three steps. Step one acquires domain-oblivious, context-aware representations of the utterance word by exploiting (a) linguistic features; (b) named entity recognition cues; (c) contextual embeddings from pre-trained language models. Step two fine-tunes these rich representations and produces slot-independent tags for each word. Step three exploits generalizable context-aware utterance-slot similarity features at the word level, uses slot-independent tags, and contextualizes them to produce slot-specific predictions for each word. Our thorough evaluation on four diverse public datasets demonstrates that our approach consistently outperforms the SOTA models by 17.52%, 22.15%, 17.42%, and 17.95% on average for unseen domains on SNIPS, ATIS, MultiWOZ, and SGD datasets, respectively.

Umar Farooq, A. B. Siddique, Fuad Jamour et al.

Responding to user reviews promptly and satisfactorily improves application ratings, which is key to application popularity and success. The proliferation of such reviews makes it virtually impossible for developers to keep up with responding manually. To address this challenge, recent work has shown the possibility of automatic response generation. However, because the training review-response pairs are aggregated from many different apps, it remains challenging for such models to generate app-specific responses, which, on the other hand, are often desirable as apps have different features and concerns. Solving the challenge by simply building a model per app (i.e., training with review-response pairs of a single app) may be insufficient because individual apps have limited review-response pairs, and such pairs typically lack the relevant information needed to respond to a new review. To enable app-specific response generation, this work proposes AARSynth: an app-aware response synthesis system. The key idea behind AARSynth is to augment the seq2seq model with information specific to a given app. Given a new user review, it first retrieves the top-K most relevant app reviews and the most relevant snippet from the app description. The retrieved information and the new user review are then fed into a fused machine learning model that integrates the seq2seq model with a machine reading comprehension model. The latter helps digest the retrieved reviews and app description. Finally, the fused model generates a response that is customized to the given app. We evaluated AARSynth using a large corpus of reviews and responses from Google Play. The results show that AARSynth outperforms the state-of-the-art system by 22.2% on BLEU-4 score. Furthermore, our human study shows that AARSynth produces a statistically significant improvement in response quality compared to the state-of-the-art system.

A. B. Siddique, Samet Oymak, Vagelis Hristidis

Paraphrasing is expressing the meaning of an input sentence in different wording while maintaining fluency (i.e., grammatical and syntactical correctness). Most existing work on paraphrasing use supervised models that are limited to specific domains (e.g., image captions). Such models can neither be straightforwardly transferred to other domains nor generalize well, and creating labeled training data for new domains is expensive and laborious. The need for paraphrasing across different domains and the scarcity of labeled training data in many such domains call for exploring unsupervised paraphrase generation methods. We propose Progressive Unsupervised Paraphrasing (PUP): a novel unsupervised paraphrase generation method based on deep reinforcement learning (DRL). PUP uses a variational autoencoder (trained using a non-parallel corpus) to generate a seed paraphrase that warm-starts the DRL model. Then, PUP progressively tunes the seed paraphrase guided by our novel reward function which combines semantic adequacy, language fluency, and expression diversity measures to quantify the quality of the generated paraphrases in each iteration without needing parallel sentences. Our extensive experimental evaluation shows that PUP outperforms unsupervised state-of-the-art paraphrasing techniques in terms of both automatic metrics and user studies on four real datasets. We also show that PUP outperforms domain-adapted supervised algorithms on several datasets. Our evaluation also shows that PUP achieves a great trade-off between semantic similarity and diversity of expression.