ABSTRACT Point‐of‐interest (POI) extraction aims to extract text POIs from real‐world data. Existing POI methods, such as social media‐based user generating and web crawling, either require massive human resources or cannot guarantee integrity and reliability. Therefore, in this paper, an end‐to‐end POI extraction framework based on digital city is proposed. It is built of digital models, textures, tiles and other digital assets collected by aircraft. The extraction process for POIs consists of segmenting it into four sequential stages: collecting, segmentation, recognition and cleaning, each enhanced through fine‐tuning on a proposed specialised digital scene dataset or via the development of tailored algorithms. Specifically, in the last stage, the application of large language model (LLM) is explored in the POI data cleaning field. By testing several LLMs of different scales using diverse chain‐of‐thought (CoT) strategies, the relatively optimal prompt scheme for different LLMs is identified regarding noise handling, formatted output and overall cleaning capability. Ultimately, POIs extracted through the proposed methodology exhibit superior quality and accuracy, surpassing the comprehensiveness of existing public commercial POI datasets, with the F1‐score increased by 19.6%, 21.1% and 23.8% on Amap, Baidu and Google POI datasets, respectively.

Cloth-changing person re-identification (ReID) is a newly emerging research topic that aims to retrieve pedestrians whose clothes are changed. Since the human appearance with different clothes exhibits large variations, it is very difficult for existing approaches to extract discriminative and robust feature representations. Current works mainly focus on body shape or contour sketches, but the human semantic information and the potential consistency of pedestrian features before and after changing clothes are not fully explored or are ignored. To solve these issues, in this work, a novel semantic-aware attention and visual shielding network for cloth-changing person ReID (abbreviated as SAVS) is proposed where the key idea is to shield clues related to the appearance of clothes and only focus on visual semantic information that is not sensitive to view/posture changes. Specifically, a visual semantic encoder is first employed to locate the human body and clothing regions based on human semantic segmentation information. Then, a human semantic attention (HSA) module is proposed to highlight the human semantic information and reweight the visual feature map. In addition, a visual clothes shielding (VCS) module is also designed to extract a more robust feature representation for the cloth-changing task by covering the clothing regions and focusing the model on the visual semantic information unrelated to the clothes. Most importantly, these two modules are jointly explored in an end-to-end unified framework. Extensive experiments demonstrate that the proposed method can significantly outperform state-of-the-art methods, and more robust features can be extracted for cloth-changing persons. Compared with multibiometric unified network (MBUNet) (published in TIP2023), this method can achieve improvements of 17.5% (30.9%) and 8.5% (10.4%) on the LTCC and Celeb-reID datasets in terms of mean average precision (mAP) (rank-1), respectively. When compared with the Swin Transformer (Swin-T), the improvements can reach 28.6% (17.3%), 22.5% (10.0%), 19.5% (10.2%), and 8.6% (10.1%) on the PRCC, LTCC, Celeb, and NKUP datasets in terms of rank-1 (mAP), respectively.

Developers often search for application programming interfaces (APIs) and their usage patterns to speed up the efficiency of software development. This paper focuses on the API sequence search task, which refers to using a function-relevant textual query to search for API sequences mined from open-source software repositories that can implement this function. However, the severe semantic gap between text and API makes it challenging to discover the correspondence between natural language queries and desired API sequences. Therefore, we propose a method called documentation-guided API sequence search (DGAS), through which we do not need to worry about the semantic gap between text and API. Specifically, DGAS consists of documentation-guided cross-modal attention (DGCA) and documentation-guided cross-modal matching (DGCM). DGCA calculates the cross-modal attention map using features extracted from the same modality (i.e., API documentation sequence and textual query) instead of from different modalities (i.e., API sequence and textual query) to bridge the semantic gap during the cross-modal attention phase. Besides, DGCM takes API documentation as supplementary information of API sequence to bridge the semantic gap during the cross-modal matching phase. We use the API documentation to extend the existing dataset for API sequence generation to construct a dataset for API sequence search to evaluate DGAS. Experimental results show that DGAS outperforms the baseline methods.

Software cross-modal retrieval is a popular yet challenging direction, such as bug localization and code search. Previous studies generally map natural language texts and codes into a homogeneous semantic space for similarity measurement. However, it is not easy to accurately capture their similar semantics in a homogeneous semantic space due to the semantic gap. Therefore, we propose to map the multi-modal data into heterogeneous semantic spaces to capture their unique semantics. Specifically, we propose a novel software cross-modal retrieval framework named Deep Hypothesis Testing (DeepHT). In DeepHT, to capture the unique semantics of the code’s control flow structure, all control flow paths (CFPs) in the control flow graph are mapped to a CFP sample set in the sample space. Meanwhile, the text is mapped to a CFP correlation distribution in the distribution space to model its correlation with different CFPs. The matching score is calculated according to how well the sample set obeys the distribution using hypothesis testing. The experimental results on two text-to-code retrieval tasks (i.e., bug localization and code search) and two code-to-text retrieval tasks (i.e., vulnerability knowledge retrieval and historical patch retrieval) show that DeepHT outperforms the baseline methods.

To date, only a few works have focused on the cloth-changing person Re-identification (ReID) task, but since it is very difficult to extract generalized and robust features for representing people with different clothes, thus, their performances need to be improved. Moreover, visual-semantic information is also often ignored. To solve these issues, in this work, a novel multigranular visual-semantic embedding algorithm (MVSE) is proposed for cloth-changing person ReID, where visual semantic information and human attributes are embedded into the network, and the generalized features of human appearance can be well learned to effectively solve the problem of cloth-changing. Specifically, to fully represent a person with clothing changes, a multigranular feature representation scheme (MGR) is employed to adaptively extract multilevel and multigranular feature information, and then a cloth desensitization network (CDN) is designed to improve the feature robustness for the person with different clothes, where different high-level human attributes are fully utilized. Moreover, to further solve the issue of pose changes and occlusion under different camera perspectives, a partially semantically aligned network (PSA) is proposed to obtain the visual-semantic information that is used to align the human attributes. Most importantly, these three modules are jointly explored in a unified framework. Extensive experimental results on four cloth-changing person ReID datasets demonstrate that the MVSE algorithm can extract highly robust feature representations of cloth-changing persons, and it can outperform state-of-the-art cloth-changing person ReID approaches.