arXiv subjects

Himel Ghosh

Publications and source records attributed to Himel Ghosh.

17 recordsLinked to original sources

arXiv2026

AutoJourn: Multi-Perspective Summarisation, Bias Detection and Bias Neutralisation for LLM-Generated News in Automated Journalism

We present AutoJourn, a demonstration system for multi-perspective news generation and bias-aware evaluation using large language models (LLMs). The system tackles three core challenges in responsible automated journalism: extracting diverse perspectives from unstructured social media discussions, generating summaries that preserve viewpoint diversity, and detecting or mitigating bias in AI-generated news. The pipeline integrates advanced prompt engineering with optional retrieval augmentation to produce semantically diverse perspective sets, a multi-perspective summarisation module that merges conflicting viewpoints into balanced summaries, and a bias analysis suite supporting sentence-level bias detection and type classification in the generated news article, and automatic neutralisation. Users can inspect perspective clusters, compare stance-specific summaries, generate news articles, and apply bias-aware rewrites directly in the interface. We evaluate each component with intrinsic metrics -- semantic diversity, summary quality, and bias reduction and show improvements over strong baselines while maintaining content fidelity. A live, publicly accessible demo accompanies the paper to facilitate reproducibility and further research on socially responsible automated journalism.

cs.CL↗

arXiv2026

LLM BiasScope: A Real-Time Bias Analysis Platform for Comparative LLM Evaluation

As large language models (LLMs) are deployed widely, detecting and understanding bias in their outputs is critical. We present LLM BiasScope, a web application for side-by-side comparison of LLM outputs with real-time bias analysis. The system supports multiple providers (Google Gemini, DeepSeek, MiniMax, Mistral, Meituan, Meta Llama) and enables researchers and practitioners to compare models on the same prompts while analyzing bias patterns. LLM BiasScope uses a two-stage bias detection pipeline: sentence-level bias detection followed by bias type classification for biased sentences. The analysis runs automatically on both user prompts and model responses, providing statistics, visualizations, and detailed breakdowns of bias types. The interface displays two models side-by-side with synchronized streaming responses, per-model bias summaries, and a comparison view highlighting differences in bias distributions. The system is built on Next.js with React, integrates Hugging Face inference endpoints for bias detection, and uses the Vercel AI SDK for multi-provider LLM access. Features include real-time streaming, export to JSON/PDF, and interactive visualizations (bar charts, radar charts) for bias analysis. LLM BiasScope is available as an open-source web application, providing a practical tool for bias evaluation and comparative analysis of LLM behaviour.

cs.CL↗

arXiv2025

Explaining News Bias Detection: A Comparative SHAP Analysis of Transformer Model Decision Mechanisms

Automated bias detection in news text is heavily used to support journalistic analysis and media accountability, yet little is known about how bias detection models arrive at their decisions or why they fail. In this work, we present a comparative interpretability study of two transformer-based bias detection models: a bias detector fine-tuned on the BABE dataset and a domain-adapted pre-trained RoBERTa model fine-tuned on the BABE dataset, using SHAP-based explanations. We analyze word-level attributions across correct and incorrect predictions to characterize how different model architectures operationalize linguistic bias. Our results show that although both models attend to similar categories of evaluative language, they differ substantially in how these signals are integrated into predictions. The bias detector model assigns stronger internal evidence to false positives than to true positives, indicating a misalignment between attribution strength and prediction correctness and contributing to systematic over-flagging of neutral journalistic content. In contrast, the domain-adaptive model exhibits attribution patterns that better align with prediction outcomes and produces 63\% fewer false positives. We further demonstrate that model errors arise from distinct linguistic mechanisms, with false positives driven by discourse-level ambiguity rather than explicit bias cues. These findings highlight the importance of interpretability-aware evaluation for bias detection systems and suggest that architectural and training choices critically affect both model reliability and deployment suitability in journalistic contexts.

cs.CL↗

arXiv2025

Sleepy Chauffeur Detection and Alert Techniques for Road Safety

The most startling of the contemporary problems is the sleepiness of chauffeur which causes lots of car accidents. Prevention of those impending accidents by detecting and alerting the sleepy chauffeur is vital, otherwise that would lead to loss of lives and various traumas along with severe injuries. The slumber or sleep may be caused by huge stress, pressure, relentless work load or alcoholism, for which sleep deprivation occurs and the chauffeur while driving gets drowsy. So far, considerable amount of systems has been developed to detect drowsiness of drivers, most of which mainly depend on image processing algorithms using cameras. Some of them also incorporate artificial intelligence and machine learning based algorithms. This paper presents a review of the existing systems and also proposes an easy and cheap system using sensors and Arduino, capable of detecting sleepiness and generates siren alarm and send alert message to take precautionary measures.

eess.SY↗

arXiv2025

To Bias or Not to Bias: Detecting bias in News with bias-detector

Media bias detection is a critical task in ensuring fair and balanced information dissemination, yet it remains challenging due to the subjectivity of bias and the scarcity of high-quality annotated data. In this work, we perform sentence-level bias classification by fine-tuning a RoBERTa-based model on the expert-annotated BABE dataset. Using McNemar's test and the 5x2 cross-validation paired t-test, we show statistically significant improvements in performance when comparing our model to a domain-adaptively pre-trained DA-RoBERTa baseline. Furthermore, attention-based analysis shows that our model avoids common pitfalls like oversensitivity to politically charged terms and instead attends more meaningfully to contextually relevant tokens. For a comprehensive examination of media bias, we present a pipeline that combines our model with an already-existing bias-type classifier. Our method exhibits good generalization and interpretability, despite being constrained by sentence-level analysis and dataset size because of a lack of larger and more advanced bias corpora. We talk about context-aware modeling, bias neutralization, and advanced bias type classification as potential future directions. Our findings contribute to building more robust, explainable, and socially responsible NLP systems for media bias detection.

cs.CL↗

arXiv2024

Enabling Efficient Serverless Inference Serving for LLM (Large Language Model) in the Cloud

This review report discusses the cold start latency in serverless inference and existing solutions. It particularly reviews the ServerlessLLM method, a system designed to address the cold start problem in serverless inference for large language models. Traditional serverless approaches struggle with high latency due to the size of LLM checkpoints and the overhead of initializing GPU resources. ServerlessLLM introduces a multitier checkpoint loading system, leveraging underutilized GPU memory and storage to reduce startup times by 6--8x compared to existing methods. It also proposes live inference migration and a startup-time-optimized model scheduler, ensuring efficient resource allocation and minimizing delays. This system significantly improves performance and scalability in serverless environments for LLM workloads. Besides ServerlessLLM, several other methods from recent research literature, including Rainbowcake, are reviewed in this paper. Further discussions explore how FaaS providers tackle cold starts and the possible future scopes.

cs.DC↗

arXiv2011

Discovery of Nuclear X-ray Sources in SINGS Galaxies

We present the results of a search for nuclear X-ray activity in nearby galaxies using Chandra archival data in a sample of 62 galaxies from the Spitzer Infrared Nearby Galaxy Survey (SINGS). We detect 37 nuclear X-ray sources; seven of these are new detections. Most of the nuclear X-ray sources are likely to be AGNs. The fraction of galaxies hosting AGNs is thus about 60%, much higher than that found with optical searches, and demonstrates the efficacy of X-ray observations to find hidden AGNs in optically normal galaxies. We find that the nuclear X-ray sources are preferentially present in earlier type galaxies. Unlike what is observed at high redshift, we do not find a strong correlation between the AGN luminosity and the 24 micron luminosity of the host galaxy; we find a strong correlation with the 3.6 micron luminosity instead. This suggests that at the present epoch the accretion rate depends on the total mass of the galaxy, as perhaps does the black hole mass.

astro-ph.GA↗

arXiv2008

The Spectral Energy Distributions of Red 2MASS AGN

We present infrared (IR) to X-ray spectral energy distributions (SEDs) for 44 red AGN selected from the 2MASS survey on the basis of their red J-K$_S$ color (>2 mag) and later observed by Chandra. In comparison with optically-, radio-, and X-ray selected AGN, their median SEDs are red in the optical and near-IR with little/no blue bump. It thus seems that near-IR color selection isolates the reddest subset of AGN that can be classified optically. The shape of the SEDs is generally consistent with modest absorption by gas (in the X-ray) and dust (in the optical-IR). The levels of obscuration, estimated from X-rays, far-IR and our detailed optical/near-IR color modeling are all consistent implying N_H < few*10^{22} cm^{-2}. We present SED models that show how the AGN optical/near-IR colors change due to differing amounts of reddening, AGN to host galaxy ratio, redshift and scattered light emission and apply them to the sources in the sample. We find that the 2MASS AGN optical color, B-R, and to a lesser extent the near-IR color, J-K$_S$, are strongly affected by reddening, host galaxy emission, redshift, and in few, highly polarized objects, also by scattered AGN light. The obscuration/inclination of the AGN allows us to see weaker emission components which are generally swamped by the AGN.

astro-ph↗

arXiv2008

Detecting Low-Mass Supermassive Black Holes

We demonstrate the feasibility of uncovering supermassive black holes in late-type, quiescent spiral galaxies by detecting signs of very low-level nuclear activity. We use a combination of x-ray selection and multi-wavelength follow-up. Here, we apply this technique to NGC 3184 and NGC 5457, both of type Scd, and show that strong arguments can be made that both host AGNs.

astro-ph↗

arXiv2008

Finding Local Low-mass Supermassive Black Holes

The low-mass end of the supermassive black hole mass function is unknown and difficult to determine. Here we discuss our successful program to find active nuclei of late type "normal" galaxies using X-ray detections and multiwavelength identifications. We conclude that most of the Chandra detected nuclear X-ray sources are AGNs. We then outline methods of black hole mass determination when broad emission lines are unobservable.

astro-ph↗

arXiv2008

Low-Level Nuclear Activity in Nearby Spiral Galaxies

We are conducting a search for supermassive black holes (SMBHs) with masses below 10^7 M_sun by looking for signs of extremely low-level nuclear activity in nearby galaxies that are not known to be AGNs. Our survey has the following characteristics: (a) X-ray selection using the Chandra X-ray Observatory, since x-rays are a ubiquitous feature of AGNs; (b) Emphasis on late-type spiral and dwarf galaxies, as the galaxies most likely to have low-mass SMBHs; (c) Use of multiwavelength data to verify the source is an AGN; and (d) Use of the highest angular resolution available for observations in x-rays and other bands, to separate nuclear from off-nuclear sources and to minimize contamination by host galaxy light. Here we show the feasibility of this technique to find AGNs by applying it to six nearby, face-on spiral galaxies (NGC 3169, NGC 3184, NGC 4102, NGC 4647, NGC 4713, NGC 5457) for which data already exist in the Chandra archive. All six show nuclear x-ray sources. The data as they exist at present are ambiguous regarding the nature of the nuclear x-ray sources in NGC 4713 and NGC 4647. We conclude, in accord with previous studies, that NGC 3169 and NGC 4102 are almost certainly AGNs. Most interestingly, a strong argument can be made that NGC 3184 and NGC 5457, both of type Scd, host AGNs.

astro-ph↗

arXiv2007

NGC 5548 in a Low-Luminosity State: Implications for the Broad-Line Region

We describe results from a new ground-based monitoring campaign on NGC 5548, the best studied reverberation-mapped AGN. We find that it was in the lowest luminosity state yet recorded during a monitoring program, namely L(5100) = 4.7 x 10^42 ergs s^-1. We determine a rest-frame time lag between flux variations in the continuum and the Hbeta line of 6.3 (+2.6/-2.3) days. Combining our measurements with those of previous campaigns, we determine a weighted black hole mass of M_BH = 6.54 (+0.26/-0.25) x 10^7 M_sun based on all broad emission lines with suitable variability data. We confirm the previously-discovered virial relationship between the time lag of emission lines relative to the continuum and the width of the emission lines in NGC 5548, which is the expected signature of a gravity-dominated broad-line region. Using this lowest luminosity state, we extend the range of the relationship between the luminosity and the time lag in NGC 5548 and measure a slope that is consistent with alpha = 0.5, the naive expectation for the broad line region for an assumed form of r ~ L^alpha. This value is also consistent with the slope recently determined by Bentz et al. for the population of reverberation-mapped AGNs as a whole.

astro-ph↗

arXiv2006

The Mass of the Black Hole in the Seyfert 1 Galaxy NGC 4593 from Reverberation Mapping

We present new observations leading to an improved black hole mass estimate for the Seyfert 1 galaxy NGC 4593 as part of a reverberation-mapping campaign conducted at the MDM Observatory. Cross-correlation analysis of the H_beta emission-line light curve with the optical continuum light curve reveals an emission-line time delay of 3.73 (+-0.75) days. By combining this time delay with the H_beta line width, we derive a central black hole mass of M_BH = 9.8(+-2.1)x10^6 M_sun, an improvement in precision of a factor of several over past results.

astro-ph↗

arXiv2006

A Reverberation-Based Mass for the Central Black Hole in NGC 4151

We have undertaken a new ground-based monitoring campaign to improve the estimates of the mass of the central black hole in NGC 4151. We measure the lag time of the broad H beta line response compared to the optical continuum at 5100 A and find a lag of 6.6 (+1.1/-0.8) days. We combine our data with the recent reanalysis of UV emission lines by Metzroth et al. to calculate a weighted mean of the black hole mass, M_BH = 4.57 (+0.57/-0.47) x 10^7 M_sun. The absolute calibration of the black hole mass is based on normalization of the AGN black hole mass - stellar velocity dispersion (M_BH - sigma_*) relationship to that of quiescent galaxies by Onken et al. The scatter in the M_BH - sigma_* relationship suggests that reverberation-mapping based mass measurements are typically uncertain by a factor of 3-4.

astro-ph↗

arXiv2006

Chandra Observations of Candidate "True" Seyfert 2 Nuclei

The Unification Model for active galactic nuclei posits that Seyfert 2s are intrinsically like Seyfert 1s, but that their broad-line regions (BLRs) are hidden from our view. A Seyfert 2 nucleus that truly lacked a BLR, instead of simply having it hidden, would be a so-called "true" Seyfert 2. No object has as yet been conclusively proven to be one. We present a detailed analysis of four of the best "true" Seyfert 2 candidates discovered to date: IC 3639, NGC 3982, NGC 5283, and NGC 5427. None of the four has a broad H-alpha emission line, either in direct or polarized light. All four have rich, high-excitation spectra, blue continua, and Hubble Space Telescope (HST) images showing them to be unresolved sources with no host-galaxy obscuration. To check for possible obscuration on scales smaller than that resolvable by HST, we obtained X-ray observations using the Chandra X-ray Observatory. All four objects show evidence of obscuration and therefore could have hidden BLRs. The picture that emerges is of moderate to high, but not necessarily Compton-thick, obscuration of the nucleus, with extra-nuclear soft emission extended on the hundreds-of-parsecs scale that may originate in the narrow-line region. Since the extended soft emission compensates, in part, for the nuclear soft emission lost to absorption, both absorption and luminosity are likely to be severely underestimated unless the X-ray spectrum is of sufficient quality to distinguish the two components. This is of special concern where the source is too faint to produce a large number of counts, or where the source is too far away to resolve the extended soft X-ray emitting region.

astro-ph↗

arXiv2002

Chandra Detection of Highest Redshift (z~6) Quasars in X-rays

We report on Chandra observations of three quasars SDSSP J083643+005453, SDSSP J103027+052455, and SDSSP J130608+035626 at redshifts 5.82, 6.28 and 5.99 respectively. All the three sources are clearly detected in the X-ray band, up to rest frame energies of ~55 keV. These observations demonstrate the unprecedented sensitivity of Chandra to detect faint sources in relatively short exposure times (5.7--8.2 ksec). The broad band X-ray properties of these highest redshift quasars do not appear to be any different from their lower redshift cousins. Spectra of the sources could not be determined with only few counts detected. Observations with XMM-Newton will be able to constrain the spectral shapes, if they are simple. Determination of complex spectra in a reasonable amount of time, however, will have to await next generation of X-ray missions.

astro-ph↗

arXiv2001

X-ray properties of Red Active Galactic Nuclei

The Two Micron All Sky Survey (2MASS) is finding previously unidentified, luminous red active galactic nuclei (AGN). This new sample has a space density similar to, or greater than, previously known AGN, suggesting that a large fraction of the overall population has been missed. Chandra observations of a well-defined subset of these objects reveal that all are X-ray faint, with the reddest sources being the faintest in X-rays. The X-ray hardness ratios cover a wide range, generally indicating log NH ~21-23 /cm*2, but the softest sources show no spectral evidence for intrinsic absorption. These characteristics suggest that a mix of absorbed, direct emission and unabsorbed, scattered and/or extended emission contributes to the X-ray flux, although we cannot rule out the possibility that they are intrinsically X-ray weak. This population of X-ray faint, predominantly broad-line objects could provide the missing population of X-ray absorbed AGN required by current models of the cosmic X-ray background. The existence of AGN which display both broad emission lines and absorbed X-rays has important implications for unification schemes and emphasizes the need for care in assigning classifications to individual AGN.

astro-ph↗