Few-shot named entity recognition is to identify named entities in scenarios where labeled data is scarce. Existing prototype building methods ignore the use of class semantic and it is difficult to obtain accurate prototype representations only by relying on few support samples. In this paper, we propose a class semantic prompts enhanced prototypical fusion method (CSFP). Specifically, we design a class-semantic prototype that adapts to current task through prompts. In order to add intra-class similarity to the existing prototype and obtain a more accurate and stable prototype representation, we consider the samples distribution and fuse class-semantic prototype with existing prototype through a weighted strategy. Experimental results on two few-shot NER benchmarks show that our method outperforms previous SOTA methods. The analysis further verifies the effectiveness of our method.

Knowledge graphs often face the issue of missing links. Addressing the problem of reasoning about and completing these missing entities or relations has become a key research focus. However, existing graph attention networks rely on connections within the graph for information propagation and aggregation, limiting their ability to effectively utilize disconnected graph structures, which leads to a poor performance. In this paper, we propose a maximum Mutual Information Estimation based Graph Attention Network (MIEGAT). This model aims to capture both local connected information and global non-connected information from knowledge graphs, enabling unsupervised extraction of relational graph structural information through mutual information maximization. Experimental results demonstrate that the MIEGAT model achieves the state-of-the-art performance across four datasets, effectively extracting non-connected information and representing sparse entities.

Compared with Static Knowledge Graphs, Temporal Knowledge Graphs need to pay more attention to the time when facts occur and these facts will change over time. However, existing models lack the capture of entity and relation and timestamp feature interactions, which is mainly reflected in the temporal multi-relation pattern and some relations exhibit persistence. To address the above issues, we propose a new TKGC model, which is Multi Transformation of Entities in Quaternion Vector Space (MTE). Specifically, we embed entities into 3D space and represent timestamps and relations as quaternions. MTE learns a pair of entity relation-aware vectors for each relation and an additional coordinate offset vector for each timestamp. In this way, MTE can capture the feature interactions between entities and relations and strengthen the interaction between entities and timestamps. Extensive experiment results show that MTE produces state-of-the-art performances on well-known benchmark datasets for Temporal Knowledge Graph Completion.

Existing research on temporal knowledge graph completion treats temporal information as supplementary, without simulating various features of facts from a temporal perspective. This work summarizes features of temporalized facts from both diachronic and synchronic perspectives: (1) Diachronicity. Facts often exhibit varying characteristics and trends across different temporal domains; (2) Synchronicity. In specific temporal contexts, various relations between entities influence each other, generating latent semantics. To track above issues, we design a quaternion-based model, TeDS, which divides timestamps into diachronic and synchronic timestamps to support dual temporal perception: (a) Two composite quaternions fusing time and relation information are generated by reorganizing synchronic timestamp and relation quaternions, and Hamilton operator achieves their interaction. (b) Each time point is sequentially mapped to an angle and converted to scalar component of a quaternion using trigonometric functions to build diachronic timestamps. We then rotate relation by using Hamilton operator between it and diachronic timestamp. In this way, TeDS achieves deep integration of relations and time while accommodating different perspectives. Empirically, TeDS significantly outperforms SOTA models on six benchmarks.

Distant supervision is widely used in relation extraction because it can automatically annotate data based on existing Knowledge Graph and corpus. Inevitably, it also results in noisy labels problem. In order to address the problem, the usual method is to put all sentences with the same entity pair in a bag, set bag-level label for them, and perform relation prediction on bag-level. However, in some downstream tasks such as question answering and semantic parsing, accurate sentence-level prediction is more important. So in this paper, we conduct study on the sentence-level and propose a novel and efficient sentence-level distant supervision relation extraction framework, SEDSL. Specifically, we adopt soft labels that can be dynamically updated during training phase to provide more accurate supervision signals to alleviate the influence of noisy labels and propose a tighter noise-filtering and re-labeling strategy to identify noisy instances and re-label them. Moreover, SEDSL is independent of the backbone network structure, so it is more general and can be applied to various sentence encoders. Extensive experimental results on NYT-10 dataset show the significant improvement of the proposed framework over all baseline methods on sentence-level relation extraction and noise reduction effect.

The distant supervision method automatically brings a large amount of training data to the relation extraction task through machine labelling, but inevitably introduces massive noise. Current methods aim to obtain precise bag representations to address the issue of noise by integrating multi-level information. However, they overlook the failure of point embeddings to capture the entailment among relations. In this paper, we propose a model called DSPrompt, which applies prompt learning to the DSRE task for the first time. To model the entailment among models, we use Gaussian embeddings, which indicate the generalized representations of relations, instead of traditional point embeddings. However, this can lead to the vanishing of contextual semantics, so we inject bag-specific contextual information into the Gaussian embeddings to offset the embeddings toward the actual semantics. We conduct extensive experiments on three widely used datasets, and the results show that our proposed model brings significant performance improvement compared with state-of-the-art DSRE methods. The codes are available at https://github.com/zc-meng/DSPrompt.

Multi-behavior sequential recommendation (MBSR) predicts a user's next item of interest based on their interaction history across different behavior types. Although existing studies have proposed capturing the correlation between different types of behavior, two important challenges have not been explored: i) Dealing with heterogeneous item transitions (both global and local perspectives). ii) Mitigating the issue of noise that arises from the incorporation of auxiliary behaviors. To address these issues, we propose a novel solution, Global Heterogeneous Graph and Target Interest Denoising for Multi-behavior Sequential Recommendation (GHTID). In particular, we view the transitions between behavior types of items as different relationships and propose two heterogeneous graphs. By considering the relationship between items under different behavioral types of transformations, we propose two heterogeneous graph convolution modules and explicitly learn heterogeneous item transitions. Moreover, we utilize two attention networks to integrate long-term and short-term interests associated with the target behavior to alleviate the noisy interference of auxiliary behaviors. Extensive experiments on four real-world datasets demonstrate that our method outperforms other state-of-the-art methods.

Knowledge concept recommendation is a kind of fine-grained recommendation in massive open online courses (MOOCs) scenario, user interaction data has the characteristics of strong collaborative signals and imbalanced interactions. This leads to a single recommendation and reduced accuracy. Recently, the ability of contrastive learning (CL) in mitigating interaction imbalance in recommender systems has received widespread attention. CL requires the use of augmentation methods to generate different views. Existing augmentation methods (1) augment only at the topology or feature level ignoring semantic or structural information, and (2) undifferentiated augmentation tends to lose the critical information. In this paper, we propose Graph Contrastive Learning with Adaptive Augmentation for Knowledge Concept Recommendation (GCARec). Specifically, (1) topology level adaptive augmentation based on degree centrality captures critical structural information, and then (2) feature level adaptive augmentation based on degree centrality captures critical semantic information. Comprehensive experiments show that our proposed approach can outperform other baselines. Our implementations are available at https://github.com/DingZhaoyuan/GCARec.

In the real world, the Knowledge Graph(KG) is dynamic and new entities are added at any time. Therefore, open-world Knowledge Graph Completion(KGC) was proposed to approach new-added entities, but previous approaches often introduced too much noise when introducing external text resources for new entities. To alleviate this problem, knowledge graph zero-shot relational learning (KGZSL) has recently attracted more attention. Generative Adversarial Networks (GANs) are frequently used in KGZSL to connect existing relation descriptions to the domain of knowledge graphs. However, these methods ignore the impact of existing entities on embeddings for unidentified relational representations and in-stead concentrate on examining the connection between relational textual texts and knowledge network structures. In this work, we propose a multi-block attention framework using relation Description Generative Adversarial Networks (desGAN) jointing KG and text representation and address model collapse and training stability problems in previous studies. The core idea of our method is to obtain the background knowledge graph information and the relation representation through and multi-block attention layer and the desGAN, then a connection between the structured KG semantic space and the unstructured text semantic space of the new entity is established, forcing the entity pair to be closer to their real relation. Experimental results on the knowledge graph zeroshot relational learning dataset demonstrate that our MASZSL has a faster convergence speed and achieves state-of-the-art performance on this task.

Compared with static knowledge graphs (KGs) temporal KGs record the dynamic relations between entities over time, therefore, research on temporal Knowledge Graph Completion (KGC) attracts much attention. Temporal KGs exhibit complex temporal relation patterns, such as multiple relations. However, existing methods can hardly model all the relation patterns and apply to the temporal KGs. In this paper, we propose a novel temporal KGC method that Combining Translation and Rotation (ComTR) in Dual Quaternion Space for temporal KGC. Specifically, we use dual-quaternion-based multiplication to model timestamps and relations as the combination of translation and rotation operations. We analyze the relation patterns of temporal KGs in detail and demonstrate that our method can model all the relation patterns in temporal KGs. Empirically, we show that ComTR can achieve the state-of-the-art performances over four temporal KGC benchmarks datasets.

Deep Neural Network (DNN) testing is one of the most widely-used ways to guarantee the quality of DNNs. However, labeling test inputs to check the correctness of DNN prediction is very costly, which could largely affect the efficiency of DNN testing, even the whole process of DNN development. To relieve the labeling-cost problem, we propose a novel test input prioritization approach (called PRIMA) for DNNs via intelligent mutation analysis in order to label more bug-revealing test inputs earlier for a limited time, which facilitates to improve the efficiency of DNN testing. PRIMA is based on the key insight: a test input that is able to kill many mutated models and produce different prediction results with many mutated inputs, is more likely to reveal DNN bugs, and thus it should be prioritized higher. After obtaining a number of mutation results from a series of our designed model and input mutation rules for each test input, PRIMA further incorporates learning-to-rank (a kind of supervised machine learning to solve ranking problems) to intelligently combine these mutation results for effective test input prioritization. We conducted an extensive study based on 36 popular subjects by carefully considering their diversity from five dimensions (i.e., different domains of test inputs, different DNN tasks, different network structures, different types of test inputs, and different training scenarios). Our experimental results demonstrate the effectiveness of PRIMA, significantly outperforming the state-of-the-art approaches (with the average improvement of 8.50%~131.01% in terms of prioritization effectiveness). In particular, we have applied PRIMA to the practical autonomous-vehicle testing in a large motor company, and the results on 4 real-world scene-recognition models in autonomous vehicles further confirm the practicability of PRIMA.

With the growth of software systems, logs have become an important data to aid system maintenance. Log-based anomaly detection is one of the most important methods for such purpose, which aims to automatically detect system anomalies via log analysis. However, existing log-based anomaly detection approaches still suffer from practical issues due to either depending on a large amount of manually labeled training data (supervised approaches) or unsatisfactory performance without learning the knowledge on historical anomalies (unsupervised and semi-supervised approaches). In this paper, we propose a novel practical log-based anomaly detection approach, PLELog, which is semi-supervised to get rid of time-consuming manual labeling and incorporates the knowledge on historical anomalies via probabilistic label estimation to bring supervised approaches' superiority into play. In addition, PLELog is able to stay immune to unstable log data via semantic embedding and detect anomalies efficiently and effectively by designing an attention-based GRU neural network. We evaluated PLELog on two most widely-used public datasets, and the results demonstrate the effectiveness of PLELog, significantly outperforming the compared approaches with an average of 181.6% improvement in terms of F1-score. In particular, PLELog has been applied to two real-world systems from our university and a large corporation, further demonstrating its practicability