Search

Accurate short-term forecasts of passenger flow in metro systems under delay conditions are crucial for emergency response and service recovery, which pose significant challenges and are currently under-researched. Due to the rare occurrence of delay events, the limited sample size under delay condictions make it difficult for conventional models to effectively capture the complex impacts of delays on passenger flow, resulting in low forecasting accuracy. Recognizing the strengths of large language models (LLMs) in few-shot learning due to their powerful pre-training, contextual understanding, ability to perform zero-shot and few-shot reasoning, to address the issues that effectively generalize and adapt with minimal data, we propose a passenger flow forecasting framework under delay conditions that synthesizes an LLM with carefully designed prompt engineering. By Refining prompt design, we enable the LLM to understand delay event information and the pattern from historical passenger flow data, thus overcoming the challenges of passenger flow forecasting under delay conditions. The propmpt engineering in the framework consists of two main stages: systematic prompt generation and prompt refinement. In the prompt generation stage, multi-source data is transformed into descriptive texts understandable by the LLM and stored. In the prompt refinement stage, we employ the multidimensional Chain of Thought (CoT) method to refine the prompts. We verify the proposed framework by conducting experiments using real-world datasets specifically targeting passenger flow forecasting under delay conditions of Shenzhen metro in China. The experimental results demonstrate that the proposed model performs particularly well in forecasting passenger flow under delay conditions., Comment: 14 pages, 2 figures

Train delays can propagate rapidly throughout the Urban Rail Transit (URT) network under networked operation conditions, posing significant challenges to operational departments. Accurately predicting passenger travel choices under train delays can provide interpretable insights into the redistribution of passenger flow, offering crucial decision support for emergency response and service recovery. However, the diversity of travel choices due to passenger heterogeneity and the sparsity of delay events leads to issues of data sparsity and sample imbalance in the travel choices dataset under metro delays. It is challenging to model this problem using traditional machine learning approaches, which typically rely on large, balanced datasets. Given the strengths of large language models (LLMs) in text processing, understanding, and their capabilities in small-sample and even zero-shot learning, this paper proposes a novel Passenger Travel Choice prediction framework under metro delays with the Large Language Model (DelayPTC-LLM). The well-designed prompting engineering is developed to guide the LLM in making and rationalizing predictions about travel choices, taking into account passenger heterogeneity and features of the delay events. Utilizing real-world data from Shenzhen Metro, including Automated Fare Collection (AFC) data and detailed delay logs, a comparative analysis of DelayPTC-LLM with traditional prediction models demonstrates the superior capability of LLMs in handling complex, sparse datasets commonly encountered under disruption of transportation systems. The results validate the advantages of DelayPTC-LLM in terms of predictive accuracy and its potential to provide actionable insights for big traffic data., Comment: 15 pages,4 figures

Time position embeddings capture the positional information of time steps, often serving as auxiliary inputs to enhance the predictive capabilities of time series models. However, existing models exhibit limitations in capturing intricate time positional information and effectively utilizing these embeddings. To address these limitations, this paper proposes a novel model called D2Vformer. Unlike typical prediction methods that rely on RNNs or Transformers, this approach can directly handle scenarios where the predicted sequence is not adjacent to the input sequence or where its length dynamically changes. In comparison to conventional methods, D2Vformer undoubtedly saves a significant amount of training resources. In D2Vformer, the Date2Vec module uses the timestamp information and feature sequences to generate time position embeddings. Afterward, D2Vformer introduces a new fusion block that utilizes an attention mechanism to explore the similarity in time positions between the embeddings of the input sequence and the predicted sequence, thereby generating predictions based on this similarity. Through extensive experiments on six datasets, we demonstrate that Date2Vec outperforms other time position embedding methods, and D2Vformer surpasses state-of-the-art methods in both fixed-length and variable-length prediction tasks.

W49A is a prominent giant molecular cloud (GMC) that exhibits strong star formation activities, yet its structural and kinematic properties remain uncertain. Our study aims to investigate the large-scale structure and kinematics of W49A, and elucidate the role of filaments and hub-filament systems (HFSs) in its star formation activity. We utilized continuum data from Herschel and the James Clerk Maxwell Telescope (JCMT) as well as the molecular lines 12CO (3-2), 13CO (3-2), and C18O (3-2) to identify filaments and HFS structures within W49A. Further analysis focused on the physical properties, kinematics, and mass transport within these structures. Additionally, recombination line emission from the H I/OH/Recombination (THOR) line survey was employed to trace the central H II region and ionized gas. Our findings reveal that W49A comprises one blue-shifted (B-S) HFS and one red-shifted (R-S) HFS, each with multiple filaments and dense hubs. Notably, significant velocity gradients were detected along these filaments, indicative of material transport toward the hubs. High mass accretion rates along the filaments facilitate the formation of massive stars in the HFSs. Furthermore, the presence of V-shaped structures around clumps in position-velocity diagrams suggests ongoing gravitational collapse and local star formation within the filaments. Our results indicate that W49A consists of one R-S HFS and one B-S HFS, and that the material transport from filaments to the hub promotes the formation of massive stars in the hub. These findings underscore the significance of HFSs in shaping the star formation history of W49A., Comment: 19 pages, 22 figures. Accepted to A&A

Recent advances in diffusion models can generate high-quality and stunning images from text. However, multi-turn image generation, which is of high demand in real-world scenarios, still faces challenges in maintaining semantic consistency between images and texts, as well as contextual consistency of the same subject across multiple interactive turns. To address this issue, we introduce TheaterGen, a training-free framework that integrates large language models (LLMs) and text-to-image (T2I) models to provide the capability of multi-turn image generation. Within this framework, LLMs, acting as a "Screenwriter", engage in multi-turn interaction, generating and managing a standardized prompt book that encompasses prompts and layout designs for each character in the target image. Based on these, Theatergen generate a list of character images and extract guidance information, akin to the "Rehearsal". Subsequently, through incorporating the prompt book and guidance information into the reverse denoising process of T2I diffusion models, Theatergen generate the final image, as conducting the "Final Performance". With the effective management of prompt books and character images, TheaterGen significantly improves semantic and contextual consistency in synthesized images. Furthermore, we introduce a dedicated benchmark, CMIGBench (Consistent Multi-turn Image Generation Benchmark) with 8000 multi-turn instructions. Different from previous multi-turn benchmarks, CMIGBench does not define characters in advance. Both the tasks of story generation and multi-turn editing are included on CMIGBench for comprehensive evaluation. Extensive experimental results show that TheaterGen outperforms state-of-the-art methods significantly. It raises the performance bar of the cutting-edge Mini DALLE 3 model by 21% in average character-character similarity and 19% in average text-image similarity.

We report the detection of the ordered alignment between the magnetic field and kpc-scale bubbles in the nearby spiral galaxy, NGC\,628. Applying the Velocity Gradient Technique (VGT) on CO spectroscopic data from the ALMA-PHANGS, the magnetic field of NGC\,628 is measured at the scale of 191\,pc ($\sim$ 4\,$''$). The large-scale magnetic field is oriented parallel to the spiral arms and curves around the galactic bubble structures in the mid-infrared emission observed by the James Webb Space Telescope (JWST). Twenty-one bubble structures have been identified at the edges of spiral arms with scales over 300\,pc, which includes two kpc-scale structures. These bubbles are caused by supernova remnants and prolonged star formation and are similar to the outflow chimneys found in neutral hydrogen in galactic disks. At the edge of the bubbles, the shocks traced by the OIII emission present a curved magnetic field that parallels the bubble's shell. The magnetic field follows the bubble expansion and binds the gas in the shell to trigger further star formation. By analyzing the larger sample of 1694 bubbles, we found a distinct radial-size distribution of bubbles in NGC\,628 indicating the star formation history in the galaxy., Comment: 15 pages, 7 figures, Accepted by ApJ

Single-pointing observations of NH$_3$ (1,1) and (2,2) were conducted towards 672 Planck Early Release Cold Cores (ECCs) using the Nanshan 26-m radio telescope. Out of these sources, a detection rate of 37% (249 cores) was achieved, with NH$_3$(1,1) hyperfine structure detected in 187 and NH$_3$(2,2) emission lines detected in 76 cores. The detection rate of NH3 is positively correlated with the continuum emission fluxes at a frequency of 857 GHz. Among the observed 672 cores, ~22% have associated stellar and IR objects within the beam size (~2$\arcmin$). This suggests that most of the cores in our sample may be starless. The kinetic temperatures of the cores range from 8.9 to 20.7 K, with an average of 12.3 K, indicating a coupling between gas and dust temperatures. The ammonia column densities range from 0.36 to 6.07$\times10^{15}$ cm$^{-2}$, with a median value of 2.04$\times10^{15}$ cm$^{-2}$. The fractional abundances of ammonia range from 0.3 to 9.7$\times10^{-7}$, with an average of 2.7 $\times10^{-7}$, which is one order of magnitude larger than that of Massive Star-Forming (MSF) regions and Infrared Dark Clouds (IRDCs). The correlation between thermal and non-thermal velocity dispersion of the NH$_3$(1,1) inversion transition indicates the dominance of supersonic non-thermal motions in the dense gas traced by NH$_3$, and the relationship between these two parameters in Planck cold cores is weaker, with lower values observed for both parameters relative to other samples under our examination. The cumulative distribution shapes of line widths in the Planck cold cores closely resemble those of the dense cores found in regions of Cepheus, and Orion L1630 and L1641, with higher values compared to Ophiuchus. A comparison of NH3 line-center velocities with those of $^{13}$CO and C$^{18}$O shows small differences (0.13 and 0.12 km s$^{-1}$ ), suggesting quiescence on small scales., Comment: 29 pages, 13 figures

An increasing number of individuals are willing to post states and opinions in social media, which has become a valuable data resource for studying human health. Furthermore, social media has been a crucial research point for healthcare now. This paper outlines the methods in our participation in the #SMM4H 2023 Shared Tasks, including data preprocessing, continual pre-training and fine-tuned optimization strategies. Especially for the Named Entity Recognition (NER) task, we utilize the model architecture named W2NER that effectively enhances the model generalization ability. Our method achieved first place in the Task 3. This paper has been peer-reviewed and accepted for presentation at the #SMM4H 2023 Workshop.

We study the magnetic field structures in six giant filaments associated with the spiral arms of the Milky Way by applying the Velocity Gradient technique (VGT) to the 13CO spectroscopic data from GRS, Fugin, and SEDIGSM surveys. Compared to dust polarized emission, the VGT allows us to separate the foreground and background using the velocity information, from which the orientation of the magnetic field can be reliably determined. We find that in most cases, the magnetic fields stay aligned with the filament bodies, which are parallel to the disk midplane. Among these, G29, G47, and G51 exhibit smooth magnetic fields, and G24, G339, and G349 exhibit discontinuities. The fact that most filaments have magnetic fields that stay aligned with the Galactic disk midplane suggests that Galactic shear can be responsible for shaping the filaments. The fact that the magnetic field can stay regular at the resolution of our analysis (<= 10 pc) where the turbulence crossing time is short compared to the shear time suggests that turbulent motion can not effectively disrupt the regular orientation of the magnetic field. The discontinuities found in some filaments can be caused by processes including filament reassembly, gravitational collapse, and stellar feedback., Comment: 18 pages, 11 figures, accepted by ApJ

Gas infall motions play a crucial role in high-mass star formation and are characterized by observable signatures in the form of blue-shifted asymmetric spectral line profiles ("blue profiles"). However, the connection between blue profiles and infall motions is unclear due to complex gas motions at parsec scales. In this study, we present the results of an HCN (4-3) mapping survey conducted with the JCMT, towards 38 massive clumps exhibiting blue profiles in HCO+ (3-2). We extract 34 HCN cores from the 38 observed fields. The core-averaged spectra show various line profiles, indicating that blue-profile HCO+ (3-2) does not guarantee the same in HCN (4-3). Through non-LTE radiation transfer calculations, we attribute the low detection rate of high-$J$ blue profiles to a combination of insufficient HCN (4-3) opacity and intricate gas motion across different density layers. The comparison between the MALT90 and BGPS line surveys highlights the importance of appropriate tracers, high spectral resolution, and column density thresholds when searching for blue profiles. We select 11 reliable infall candidates and adopt the Hill5 model to fit the infall velocity of 0.2-1.9 km/s, corresponding to 5% to 74% of free-fall velocity. Assuming a spherically collapsing model, we estimate the median and mean mass infall rates to be 4.5E-3 and 7.6E-3 Msun/year, respectively. The consistency of the mass infall rates among different transitions suggests a steady accretion process from the clump gas envelope to the inner region., Comment: 36 pages, 9 figures, 5 tables; accepted for publication in ApJS

In this paper, we investigate the asymptotic behavior of the non-simple systole, which is the length of a shortest non-simple closed geodesic, on a random closed hyperbolic surface on the moduli space $\mathcal{M}_g$ of Riemann surfaces of genus $g$ endowed with the Weil-Petersson measure. We show that as the genus $g$ goes to infinity, the non-simple systole of a generic hyperbolic surface in $\mathcal{M}_g$ behaves exactly like $\log g$., Comment: 47 pages, 11 figures. Comments welcome

We report high angular resolution observations, made with the Atacama Large Millimeter Array in band 6, of high excitation molecular lines of $\rm CH_3CN$ and $\rm SO_2$ and of the H29$\alpha$ radio recombination line towards the G345.0061+01.794 B HC H II region, in order to investigate the physical and kinematical characteristics of its surroundings. Emission was detected in all observed components of the J=14$\rightarrow$13 rotational ladder of $\rm CH_3CN$ and in the $30_{4,26}-30_{3,27}$ and $32_{4,28}-32_{3,29}$ lines of $\rm SO_2$. The peak of the velocity integrated molecular emission is located $\sim$0$\,.\!\!^{\prime\prime}$4 northwest of the peak of the continuum emission. The first-order moment images and channel maps show a velocity gradient, of 1.1 km s$^{-1}$ arcsec$^{-1}$, across the source, and a distinctive spot of blueshifted emission towards the peak of the zero-order moment. The rotational temperature is found to decrease from 252$\pm24$ Kelvin at the peak position to 166$\pm16$ Kelvin at its edge, indicating that our molecular observations are probing a hot molecular core that is internally excited. The emission in the H29$\alpha$ line arises from a region of 0$\,.\!\!^{\prime\prime}$65 in size, where its peak coincides with that of the dust continuum. We model the kinematical characteristics of the "central blue spot" feature as due to infalling motions, suggesting a central mass of 172.8$\pm8.8 M_{\odot}$. Our observations indicate that this HC H II region is surrounded by a compact structure of hot molecular gas, which is rotating and infalling toward a central mass, that is most likely confining the ionized region. The observed scenario is reminiscent of a "butterfly pattern" with an approximately edge-on torus and ionized gas roughly parallel to its rotation axis., Comment: Accepted by ApJ

We studied the hub filament system G323.46-0.08 based on archival molecular line data from the SEDIGISM 13CO survey and infrared data from the GLIMPSE, MIPS, and Hi-GAL surveys. G323.46-0.08 consists of three filaments, F-north, F-west, and F-south, that converge toward the central high_mass clump AGAL 323.459-0.079. F-west and Part 1 of the F-south show clear large-scale velocity gradients 0.28 and 0.44 km s-1 pc-1, respectively. They seem to be channeling materials into AGAL 323.459-0.079. The minimum accretion rate was estimated to be 1216 M Myr-1. A characteristic V-shape appears around AGAL 323.459-0.079 in the PV diagram, which traces the accelerated gas motions under gravitational collapse. This has also been supported by model fitting results. All three filaments are supercritical and they have fragmented into many dense clumps. The seesaw patterns near most dense clumps in the PV diagram suggests that mass accretion also occurs along the filament toward the clumps. Our results show that filamentary accretion flows appear to be an important mechanism for supplying the materials necessary to form the central high-mass clump AGAL 323.459-0.079 and to propel the star forming activity taking place therein.

Event co-occurrences have been proved effective for event extraction (EE) in previous studies, but have not been considered for event argument extraction (EAE) recently. In this paper, we try to fill this gap between EE research and EAE research, by highlighting the question that ``Can EAE models learn better when being aware of event co-occurrences?''. To answer this question, we reformulate EAE as a problem of table generation and extend a SOTA prompt-based EAE model into a non-autoregressive generation framework, called TabEAE, which is able to extract the arguments of multiple events in parallel. Under this framework, we experiment with 3 different training-inference schemes on 4 datasets (ACE05, RAMS, WikiEvents and MLEE) and discover that via training the model to extract all events in parallel, it can better distinguish the semantic boundary of each event and its ability to extract single event gets substantially improved. Experimental results show that our method achieves new state-of-the-art performance on the 4 datasets. Our code is avilable at https://github.com/Stardust-hyx/TabEAE., Comment: Accepted to ACL 2023 main conference

Increasingly, cyber aggression becomes the prevalent phenomenon that erodes the social media environment. However, due to subjective and expense, the traditional self-reporting questionnaire is hard to be employed in the current cyber area. In this study, we put forward the prediction model for cyber aggression based on the cutting-edge deep learning algorithm. Building on 320 active Weibo users' social media activities, we construct basic, dynamic, and content features. We elaborate cyber aggression on three dimensions: social exclusion, malicious humour, and guilt induction. We then build the prediction model combined with pretrained BERT model. The empirical evidence shows outperformance and supports a stronger prediction with the BERT model than traditional machine learning models without extra pretrained information. This study offers a solid theoretical model for cyber aggression prediction. Furthermore, this study contributes to cyber aggression behaviors' probing and social media platforms' organization., Comment: 11 pages, 2 tables

In this paper, we first construct a sequence of hyperbolic surfaces with connected geodesic boundary such that the first normalized Steklov eigenvalue $\tilde{\sigma}_1$ tends to infinity. We then prove that as $g\rightarrow \infty$, a generic $\Sigma\in \mathcal{M}_{g,n}(L_g)$ satisfies $\tilde{\sigma}_1(\Sigma)>C\cdot \|L_g\|_1$ where $C$ is a positive universal constant. Here $\mathcal{M}_{g,n}(L_g)$ is the moduli space of hyperbolic surfaces of genus $g$ and $n$ boundary components of length $L_g=(L_g^1,\cdots, L_g^n)$ endowed with the Weil-Petersson metric where $\|L_g\|_1\rightarrow\infty$ satisfies certain conditions., Comment: 20pages, new results added, second theorem is improved

In this article, we study the second eigenvalues of closed hyperbolic surfaces for large genus. We show that for every closed hyperbolic surface $X_g$ of genus $g$ $(g\geq 3)$, up to uniform positive constants multiplications, the second eigenvalue $\lambda_2(X_g)$ of $X_g$ is greater than $\frac{\mathcal{L}_2(X_g)}{g^2}$ and less than $\mathcal{L}_2(X_g)$; moreover these two bounds are optimal as $g\to \infty$. Here $\mathcal{L}_2(X_g)$ is the shortest length of simple closed multi-geodesics separating $X_g$ into three components. Furthermore, we also investigate the quantity $\frac{\lambda_2(X_g)}{\mathcal{L}_2(X_g)}$ for random hyperbolic surfaces of large genus. We show that as $g\to \infty$, a generic hyperbolic surface $X_g$ has $\frac{\lambda_2(X_g)}{\mathcal{L}_2(X_g)}$ uniformly comparable to $\frac{1}{\ln(g)}$., Comment: Journal of Differential Geometry, to appear, 38 pages, 6 figures, comments are welcome

Magnetic fields play an important role in the evolution of molecular clouds and star formation. Using the Velocity Gradient Technique (VGT) model, we measured the magnetic field in Orion A using the 12CO, 13CO, and C18O (1-0) emission lines at a scale of 0.07 pc. The measured B-field shows an east-west orientation that is perpendicular to the integral shaped filament of Orion A at large scale. The VGT magnetic fields obtained from 13CO and C18O are in agreement with the B-field that is measured from the Planck 353 GHz dust polarization at a scale of 0.55 pc. Removal of density effects by using a Velocity Decomposition Algorithm can significantly improve the accuracy of the VGT in tracing magnetic fields with the 12CO (1-0) line. The magnetic field strength of seven sub-clouds, OMC-1, OMC-2, OMC-3, OMC-4, OMC-5, L 1641-N, and NGC 1999 has also been estimated with the Davis-Chandrasekhar-Fermi (DCF) and MM2 technique, and these are found to be in agreement with previous results obtained from dust polarization at far-infrared and sub-millimeter wavelengths. At smaller scales, the VGT proves a good method to measure magnetic fields., Comment: 15 pages, 6 figures, accepted to ApJ

We studied the cloud-cloud collision candidate G323.18+0.15 based on signatures of induced filaments, clumps, and star formation. We used archival molecular spectrum line data from the SEDIGISM $^{13}$CO($J$\,=\,2--1) survey, from the Mopra southern Galactic plane CO survey, and infrared to radio data from the GLIMPSE, MIPS, Hi-GAL, and SGPS surveys. Our new result shows that the G323.18+0.15 complex is 3.55kpc away from us and consists of three cloud components, G323.18a, G323.18b, and G323.18c. G323.18b shows a perfect U-shape structure, which can be fully complemented by G323.18a, suggesting a collision between G323.18a and the combined G323.18bc filamentary structure. One dense compressed layer (filament) is formed at the bottom of G323.18b, where we detect a greatly increased velocity dispersion. The bridge with an intermediate velocity in a position-velocity diagram appears between G323.18a and G323.18b, which corresponds to the compressed layer. G323.18a plus G323.18b as a whole are probably not gravitationally bound. This indicates that high-mass star formation in the compressed layer may have been caused by an accidental event. The column density in the compressed layer of about $1.36 \times 10^{22}$cm$^{-2}$ and most of the dense clumps and high-mass stars are located there. The average surface density of classI and classII young stellar objects (YSOs) inside the G323.18+0.15 complex is much higher than the density in the surroundings. The timescale of the collision between G323.18a and G323.18b is $1.59$Myr. This is longer than the typical lifetime of classI YSOs and is comparable to the lifetime of classII YSOs.

Entity Set Expansion (ESE) is a promising task which aims to expand entities of the target semantic class described by a small seed entity set. Various NLP and IR applications will benefit from ESE due to its ability to discover knowledge. Although previous ESE methods have achieved great progress, most of them still lack the ability to handle hard negative entities (i.e., entities that are difficult to distinguish from the target entities), since two entities may or may not belong to the same semantic class based on different granularity levels we analyze on. To address this challenge, we devise an entity-level masked language model with contrastive learning to refine the representation of entities. In addition, we propose the ProbExpan, a novel probabilistic ESE framework utilizing the entity representation obtained by the aforementioned language model to expand entities. Extensive experiments and detailed analyses on three datasets show that our method outperforms previous state-of-the-art methods., Comment: Accepted by SIGIR 2022 (Full Paper)