Chinese Named Entity Recognition (CNER) is a fundamental step in extracting structured information from unstructured Chinese text. To address the issue of insufficient word boundary information in Chinese text, most existing CNER methods introduce external lexicon information into deep learning models by adding additional network layers, thereby solving the CNER problem through sequence labeling, which has achieved relatively good results. However, these methods treat all possible words in the input sentence equally, and the redundant words introduced has an adverse effect on the accuracy of CNER task. Moreover, such methods struggle to handle nested entity problems. Therefore, this paper proposes a novel CNER method named UCBsLM, which consists of two steps: In the first step, to fully utilize the word boundary information provided by the external lexicon to obtain candidate named entities, UCBsLM models the utilization of lexical information in the external lexicon as the "exploration-exploitation" dilemma in reinforcement learning, and then uses UCB algorithm to obtain the optimal candidate entity segmentation schemes, thereby eliminating the influence of redundant words. In the second step, UCBsLM employs the levitated entity marker technique to process the input text of the pre-trained language model, and completes the classification of candidate named entities based on the contextualized representation vectors of the candidate entities, thereby addressing potential nested entity problems and improving the efficiency of CNER. Experiments show that the F1 of the proposed method exceed 96% on MSRA and Resume dataset, comparable to benchmark methods and achieve 71.24% on Weibo dataset, outperforming all comparison methods, which reflect the feasibility of the method.

Entity matching aims to determine whether entity information from different sources refers to the same object in the real world.It is crucial for downstream applications such as data fusion and information retrieval.Existing methods typically model the entity matching task as a binary classification problem, focusing solely on the overall probability outputs of positive and negative classes.This approach lacks sufficient attention to subtle differences, making it difficult to capture important nuances and underutilizing the samples.As a result, it often leads to incorrect matching of similar but nonmatching entity pairs.To address this issue, we proposes an entity matching method based on contrastive learning of similar samples.We improve the contrastive learning model by constructing "similar but non-matching sample pairs" in addition to positive and negative samples.The subtle differences between entities are amplified through an interactive attention mechanism.Subsequently, the improved contrastive learning loss accurately learns the similarity and dissimilarity features of the three types of sample pairs.This approach effectively solves the problem of identifying similar but non-matching entities.Experiments conducted on the publicly available WDC product dataset with four categories demonstrate the superiority of our method.Compared with the state-of-the-art methods, our approach achieves an average improvement of 2.01%, 0.14%, 1.01%, and 3.83% in F1 scores across the four product datasets of different sizes.

The network with attributes at the nodes is called an attribute network. Detecting abnormal nodes in the attribute network has a wide range of applications in real life, such as detecting malicious actors in social networks, fraudsters in financial networks, and traffic congestion in transportation networks. Existing methods mainly use contrastive learning to detect abnormal nodes, and have achieved better results than traditional and compressed-reconstruction methods. However, existing contrastive learning-based abnormal node detection methods have limitations in ensuring sufficient contrast between positive and negative samples and in capturing comprehensive abnormal features, and the detection accuracy needs to be further improved. This paper proposes a novel anomaly detection model called FGAnomAES, which aims to simultaneously discover anomalies at the full granularity level. The model builds an attribute-enhanced network, allowing nodes with attribute similarity to have edges connected in the original network. Based on this network, positive and negative sample subgraphs with sufficient contrast in both structure and attribute can be obtained by interval sampling. On the basis of node-to-subgraph and node-to-node contrastive learning, node-to-global contrast is added to learn subgraph-level, node-level, and global abnormal features. Finally, the graded comparison results of each node are combined to calculate the full-graded abnormal score of each node. Based on the abnormal scores, abnormal nodes are detected. Comparative experiments were conducted on four real-world datasets, including BlogCatalog, Cora, Flickr, and ACM, compared with models such as GRADATE, CoLA, and DOMINANT. The results show that compared with the benchmark models, the AUC values of FGAnomAES increased by 1.75%-9.23%, 1.33%-4.46%, 3.17%-11.46%, and 1.61-7.01%.

Mobile IP geolocation aims to obtain a mobile device’s geographic location by IP. This technology is widely used in preventing financial risk, investigating cybercrime, and delivering targeted information. Currently, there are three types of IP geolocation: based on cooperation, querying in database, or network measurement. However, since restricted cooperation, low-reliability databases, and unresponsive mobile IPs, existing technologies are hard to geolocate fine-grained location of mobile IP. In this paper, we propose the concept of district anchor, and propose a non-cooperative mobile IP geolocation scheme, including three parts: acquiring district anchors by clustering, evaluating the reliability of district anchors, and geolocating mobile IPs. We also give implemented approach of this scheme. Instead of using existing clustering algorithms treating IPs and geolocations in no particular order, we propose two-stages clustering algorithm (IPG2C) to acquire district anchors, and establish reliability evaluation mechanism by IP distribution and spatial distribution of cluster. Eventually, using obtained reliable district anchors, we use “subnet geolocation” strategy to geolocate mobile IPs. The experimental results in 10 cities show that: 1) our scheme can be used to geolocate mobile IPs without cooperation; 2) the mean geolocation error is 12.47km, where precision of 56.67% of mobile IPs is street-level and minimum error is only 13m; 3) that the mean geolocation error of the anchor-based method is smaller than that of the landmark-based method; 4) compared with 13 clustering algorithms (e.g., K-Means++, Mean Shift, DBSCAN, and GMM), mean geolocation error using IPG2C’s district anchors is reduced by 26.62%~50.77%.

Street-level landmarks are an important basis for street-level IP geolocation, and the web-based landmark is one of the main sources of street-level landmarks. Considering the existing street-level landmark evaluation methods having low accuracy and strict constraints, this paper analyses the causes and evaluation idea of invalid web-based candidate landmarks and proposes Evaluator, a web-based landmark evaluation approach. Evaluator adopts the idea of the decision tree to filter invalid landmarks layer by layer and comprehensively estimates the quantitative reliability of candidate landmarks with public data and services to obtain reliable landmarks. This paper proposes the domain name system (DNS) distributed query algorithm to effectively resolve all IP addresses of a domain name, which provides data support for Evaluator to filter candidate landmarks. Meanwhile, this paper also proposes a reverse verification algorithm to obtain all domain names of an IP address, which provides an important reference to calculate the reliability of a reliable landmark. In addition, gradient descent is used to assess the parameters of the reliability estimating model, which effectively improves the robustness of Evaluator. Experiments show that reliable landmarks from Evaluator reduce the geolocation error of 100 targets in Hong Kong from 7.30 km to 3.91 km, compared with the landmark verifying method (LVM), one of the latest web-based landmark evaluation methods. Moreover, Evaluator significantly improves the evaluation coverage based on the same geolocation accuracy with street-level landmark evaluation (SLE), one of the latest landmark evaluation methods.