Smart contract classification holds significant application value in the field of blockchain. However, existing methods suffer from inefficiencies and high computational complexity when dealing with smart contract data. To address these issues, this paper proposes a Cluster-BERT model based on neural clustering techniques. The model reduces the computational burden of self-attention mechanisms by clustering attention heads, thereby improving training efficiency. The Cluster-BERT model comprises multiple modules. Module 1 preprocesses smart contract data, converting abstract syntax trees and graph structure features into text representations suitable for BERT models. Module 2 serves as the core of the model, introducing neural clustering methods to reduce computational complexity. Module 3 further optimizes the model by finding the optimal number of centroids, achieving a balance between training efficiency and classification accuracy. Experimental results show that our proposed Cluster-BERT achieved an accuracy of 91.42%, a recall of 91.44%, and an F1 score of 91.43%, which indicates a noticeable improvement over the baseline model. Our model reduces computational complexity from quadratic to linear, resulting in an average reduction of 8.48% in training time and 7.88% in prediction time compared to the baseline model. On the smart contract dataset, the accuracy and precision of our model outperformed other models proposed in recent years by 1% to 2% points on average.

ABSTRACT With the advancement of information technology, social networking platforms have been widely integrated into people's daily lives, significantly altering their lifestyle and consumption habits. Currently, these platforms are abundant with texts possessing economic attributes, known as social finance texts. The ability to timely and accurately identify social finance texts with potential value holds considerable economic and practical significance. Addressing the existing challenges: (1) the presence of key information of varying lengths and sparse text features in texts, where single‐layer text content fails to provide sufficient semantic information; (2) the deficiencies of traditional static language models in expressing text features; (3) the cluttered content leading to unclear text themes, thereby diminishing the classification effectiveness. This paper proposes a social finance text classification method based on the DLMFSE (Dynamic Language Model with Multi‐Feature Fusion and Semantic Enhancement) algorithm. It involves semantic feature training of texts through a pre‐trained model based on dynamic word vector representation, obtaining document‐level and character‐level feature vectors. Subsequently, the MKC and Bi‐GRU networks are utilized for multidimensional feature extraction from character‐level vector representations of texts, acquiring multidimensional shallow features and deep semantic features of social finance texts, and merging them with document‐level feature vectors to obtain a multidimensional vector representation of social finance texts. A semantic feature extension algorithm is introduced to extract and expand key text features, resulting in semantically enhanced word vector representations. By integrating the aforementioned vector representations, a multidimensional semantic feature vector for social finance texts is constructed. Experimental results demonstrate that our DLMFSE model achieves a macro‐ F 1 score of 94.2% on a self‐constructed dataset, outperforming strong baselines including BERT by 2.9%. This significant improvement not only validates the effectiveness of our multi‐feature fusion strategy but also indicates substantial practical value for real‐time financial monitoring and opportunity discovery in social media platforms.

Predicting interactions between drugs and their targets is vital for drug discovery and repositioning. Conventional techniques are slow and labor-intensive, while deep learning algorithms offer efficient solutions. However, deep learning often focus on single drug representations or simplistic combinations, leading to suboptimal feature representation. Moreover, the prevalent use of convolutional neural networks (CNNs) in drug image representation neglects the necessity for both local and global drug information in Drug-Target Interaction (DTI) tasks. To address these challenges, we propose TriCvT-DTI, a novel approach that combines molecular images, chemical sequence features, and graph representations of drugs to comprehensively capture structural, spatial, and functional aspects. TriCvT-DTI introduces a bidirectional multi-head attention mechanism for interactive feature learning between drugs and targets, enhancing performance by modeling complex relationships. By using Convolutional Vision Transformers (CvTs), TriCvT-DTI can effectively extract structural and spatial features from drug images. We evaluate our model on three datasets: Human, C. elegans, and Davis, and we compare it with state-of-the-art methods. Then we train TriCvT-DTI with uni-modality and bi-modality to compare then extract the impact of each modality on TriCvT-DTI. Experimental results demonstrate that TriCvT-DTI outperforms existing methods on both balanced and unbalanced datasets. Moreover, it presents impressive generalization capabilities on the Drug-Target Interaction (DTI) task.

Abstract How to quickly and accurately retrieve relevant smart contracts from a huge amount of smart contracts has become an urgent need for users. The classification of smart contracts offers a solution by narrowing down the search space. Existing smart contract classification methods suffer from incomplete semantic feature extraction and a lack of consideration of the existence of rich semantics in existing smart contracts of the same class. To address the above problems, we propose a contrast learning and semantic feature embedding approach to enhance K-Nearest Neighbor (CL-SFE-IKNN). Our method fuses local features, global features, and account transaction features of the smart contract source code to perfect the semantics of the contract. Our method adopts KNN to retrieve multiple instances of contracts in the same class and assigns weights to the model output based on their labels. Meanwhile, we introduce contrastive learning and semantic feature embedding to enhance KNN retrieval to high-quality nearest neighbors of the same class. Experimental results show that by combining a KNN classifier with a traditional linear classifier, our model achieves the best performance compared with other baseline models.

Abstract In the absence of sufficient user behavior data, game recommendation systems face the cold-start problem. To address this issue, this paper proposes a solution based on the Graph Neural Network pre-training model to alleviate the cold-start problem. The proposed model directly reconstructs cold-start user/game embeddings using a meta-learning setup based on dataset training simulations and uses an adaptive neighbor sampler to improve user interaction relations and thereby to improve game recommendation performance. Experimental results demonstrate the effectiveness and practicality of the recommendation model proposed in this study. Moreover, the proposed model is embedded in the game recommendation system to visualize the recommendation results.

Abstract Short texts on social platforms often have the problems of diverse categories and semantic sparsity, making it challenging to identify the diverse intentions of users. To address this issue, this article proposes a multi‐label social short text classification method (IML‐CL) based on contrastive learning and improved ml‐KNN. First, a contrastive learning approach is employed to train a multi‐label text classification model. This approach improves semantic sparsity by leveraging the knowledge from the existing samples to enrich the feature representation of short texts. Simultaneously, an improved ml‐KNN algorithm is developed to enhance the accuracy of label prediction. This algorithm utilizes a two‐layer nearest neighbor rule and introduces a penalty function and weight optimization. Next, the model generates the feature representation for the test sample and predicts its label. Additionally, the improved ml‐KNN algorithm retrieves neighbors of the test sample and uses their label information for prediction. Finally, the two predictions are combined to obtain the final prediction, which accurately identifies the user's intention. The experimental results demonstrate that, on the dataset constructed in this article, the IML‐CL method effectively boosts the performance of the baseline model.

The classification of the smart contract can effectively reduce the search space and improve retrieval efficiency. The existing classification methods are based on natural language processing technologies. Because the processing of source code by these technologies lacks extraction and processing in the software engineering field, there is still a lot of room for improvement in their methods of feature extraction. Therefore, this paper proposes a multi-feature fusion method for smart contract classification (MFF-SC) based on the code processing technology. From the source code perspective, source code processing method and attention mechanism are used to extract local code features. Structure-based traversal method are used to extract global code features from abstract syntax tree. Local and global code features introduce attention mechanism to generate code semantic features. From the perspective of account transaction, the feature of account transaction is extracted by using TransR. Next, the code semantic features and account transaction features generate smart contract semantic features by an attention mechanism. Finally, the smart contract semantic features are fed into a stacked denoising autoencoder and a softmax classifier for classification. Experimental results on a real dataset show that MFF-SC achieves an accuracy rate of 83.9%, compared with other baselines and variants.

Compared with the traditional microwave synthetic aperture radar (SAR), the terahertz SAR is much more sensitive to the high frequency vibration of the platform which needs to be compensated. This paper proposes a novel method for the terahertz SAR imaging in which compensation for these two motion errors (i.e., high frequency vibration and the traditional low frequency motion error) are considered simultaneously. The effectiveness of the proposed method is verified by measured terahertz SAR data.

The goal of aspect-level sentiment analysis is to identify the sentiment polarity of a specific opinion target expressed; it is a fine-grained sentiment analysis task. Most of the existing works study how to better use the target information to model the sentence without using the interactive information between the sentence and target. In this article, we argue that the prediction of aspect-level sentiment polarity depends on both context and target. First, we propose a new model based on LSTM and the attention mechanism to predict the sentiment of each target in the review, the matrix-interactive attention network (M-IAN) that models target and context, respectively. M-IAN use an attention matrix to learn the interactive attention of context and target and generates the final representations of target and context. Then we introduce two gate networks based on M-IAN to build a deep interactive memory network to capture multiple interactions of target and context. The deep interactive memory network can excellently formulate specific memory for different targets, which is helpful in sentiment analysis. The experimental results of Restaurant and Laptop datasets of SemEval 2014 validate the effectiveness of our model.

Although distant supervised relation extraction has been widely used, its performance is weakened by the wrong labeling problem. During eliminating noise, many previous work failed to grasp the trade-off between long dependencies and computational complexity while using neural networks, and ignored the potential noise in external knowledge. In this paper, we propose a neural model incorporating dilated convolutional neural networks with soft entity type constraints to jointly tackle the above two issues. Rather than using traditional convolutional neural networks or recurrent neural networks, we propose dilated convolutional networks as the sentence encoder so as to capture long dependencies in large scale context and improve the robustness against local noise while keeping the efficiency of computation. Furthermore, we utilize entity types, which is considerd as external knowledge with noise, to denoise in relation classification by taking the constraints between entity types and relations into account and learning more precise entity types and attention weights simultaneously. Experimental results on the New York Times dataset show that, our model incorporating dilated convolutional networks with soft entity type constraints promotes distant supervised relation extraction and achieves state-of-art results compared with baselines.

Summary Web service recommendation becomes an increasingly important issue when more and more services are published on the Internet. Many Web service recommendation methods have been proposed in recent years, most of which adopted collaborative filtering (CF) techniques. In general, these approaches have two limitations. Firstly, they rarely leverage user ratings since this kind of explicit feedback is difficult to collect for Web services. Secondly, the new user cold‐start problem is an inherent limitation of CF because the new users have not yet cast sufficient numbers of votes. In this paper, pseudo ratings of services constructed based on plenty of user‐service interactions, also known as a kind of implicit feedback, are provided to represent users' preferences on services. Based on these pseudo ratings, we present a novel Web service recommendation approach, which can alleviate the cold‐start problem by integrating contextual information and an online learning model. Experiments conducted on a real world data set show that, compared with the method without contextual information, our proposed approach that handles the cold start problem by integrating contextual information can achieve better F‐Measure performance (5.08 times increase on average). Moreover, the proposed online recommendation approach can dramatically decrease the time overhead while keeping the similar recommendation performance.

Sparse topic models (STMs) are widely used for learning a semantically rich latent sparse representation of short texts in large scale, mainly by imposing sparse priors or appropriate regularizers on topic models. However, it is difficult for these STMs to model the sparse structure and pattern of the corpora accurately, since their sparse priors always fail to achieve real sparseness, and their regularizers bypass the prior information of the relevance between sparse coefficients. In this paper, we propose a novel Bayesian hierarchical topic models called Bayesian Sparse Topical Coding with Poisson Distribution (BSTC-P) on the basis of Sparse Topical Coding with Sparse Groups (STCSG). Different from traditional STMs, it focuses on imposing hierarchical sparse prior to leverage the prior information of relevance between sparse coefficients. Furthermore, we propose a sparsity-enhanced BSTC, Bayesian Sparse Topical Coding with Normal Distribution (BSTC-N), via mathematic approximation. We adopt superior hierarchical sparse inducing prior, with the purpose of achieving the sparsest optimal solution. Experimental results on datasets of Newsgroups and Twitter show that both BSTC-P and BSTC-N have better performance on finding clear latent semantic representations. Therefore, they yield better performance than existing works on document classification tasks.

Min Peng, Qianqian Xie, Yanchun Zhang, Hua Wang, Xiuzhen Zhang, Jimin Huang, Gang Tian. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2018.

Summary Search engine based Web service discovery model suffers from the semantic sparsity problem due to the fact that Web services are described in short texts, which in turn leads to poor recall. To address this issue, external information that enriches the semantics of the Web service and improves discovery performance has been highly concerned. In light of this, we propose a novel Web service discovery approach that uses the neural topic model, which seamlessly integrates tagging information and word embedding for semantic sparsity problem. More specifically, instead of clustering Web services as done in most existing service discovery approaches, we use word embedding to map the words as continuous embeddings to embody external semantics of the service description. We also leverage the neural topic model in service discovery, which takes continuous word distribution as the input and interprets the Web service description as a hierarchical model. Based on the neural topic model and word embedding, we propose an efficient Web service query and ranking approach. Experiments conducted on a real‐world Web service dataset demonstrate the effectiveness of the proposed approach.

This article presents an unsupervised multi-view hierarchical embedding (UMHE) framework to sufficiently reveal the intrinsic topical knowledge in social events. Event-oriented topics are highly related to such events as it can provide explicit descriptions of what have happened in social community. In many real-world cases, however, it is difficult to include all attributes of microblogs, more often, textual aspects only are available. Traditional topic modelling methods have failed to generate event-oriented topics with the textual aspects, since the inherent relations between topics are often overlooked in these methods. Meanwhile, the metrics in original word vocabulary space might not effectively capture semantic distances. Our UMHE framework overcomes the severe information deficiency and poor feature representation. The UMHE first develops a multi-view Bayesian rose tree to preliminarily generate prior knowledge for latent topics and their relations. With such prior knowledge, we design an unsupervised translation-based hierarchical embedding method to make a better representation of these latent topics. By applying self-adaptive spectral clustering on the embedding space and the original space concomitantly, we eventually extract event-oriented topics in word distributions to express social events. Our framework is purely data-driven and unsupervised, without any external knowledge. Experimental results on TREC Tweets2011 dataset and Sina Weibo dataset demonstrate that the UMHE framework can construct hierarchical structure with high fitness, but also yield topic embeddings with salient semantics; therefore, it can derive event-oriented topics with meaningful descriptions.

BACKGROUND: Drug-drug interactions (DDIs) are one of the major concerns in drug discovery. Accurate prediction of potential DDIs can help to reduce unexpected interactions in the entire lifecycle of drugs, and are important for the drug safety surveillance. RESULTS: Since many DDIs are not detected or observed in clinical trials, this work is aimed to predict unobserved or undetected DDIs. In this paper, we collect a variety of drug data that may influence drug-drug interactions, i.e., drug substructure data, drug target data, drug enzyme data, drug transporter data, drug pathway data, drug indication data, drug side effect data, drug off side effect data and known drug-drug interactions. We adopt three representative methods: the neighbor recommender method, the random walk method and the matrix perturbation method to build prediction models based on different data. Thus, we evaluate the usefulness of different information sources for the DDI prediction. Further, we present flexible frames of integrating different models with suitable ensemble rules, including weighted average ensemble rule and classifier ensemble rule, and develop ensemble models to achieve better performances. CONCLUSIONS: The experiments demonstrate that different data sources provide diverse information, and the DDI network based on known DDIs is one of most important information for DDI prediction. The ensemble methods can produce better performances than individual methods, and outperform existing state-of-the-art methods. The datasets and source codes are available at https://github.com/zw9977129/drug-drug-interaction/ .

Compressing textstreams generated by social networks can both reduce storage consumption and improve efficiency such as fast searching. However, the compression process is a challenge due to the large scale of textstreams. In this article, we propose a textstream compression framework based on compressed sensing theory and design a series of matching parallel procedures. The new approach uses a linear projection technique in the textstream compression process, achieving fast compression speed and low compression ratio. Two processes are executed by designing elaborated parallel procedures for efficient compressing and decompressing of large-scale textstreams. The decompression process is implemented for approximate solutions of underdetermined linear systems. Experimental results show that the new method can efficiently achieve the compression and decompression tasks on a large amount of text generated by social networks.

With the rapid development of SOA (Service Oriented Architecture), an increasing number of Web services have been published on the Internet. How to recommend suitable Web services to users becomes a challenging problem. Existing Web services recommendation approaches based on collaborative filtering mainly focus on QoS (Quality of Service) prediction. Recommending services based on users' ratings on services are seldom reported since such explicit feedback data is difficult to collect. In this paper, we report a dataset of implicit feedback on real-world Web services, which consist of more than 280,000 user-service interaction records, 65,000 service users and 15,000 Web services or mashups. In addition, time is becoming an increasingly important factor in recommenders since time effects influence users' preferences to a large extent. Based on the collected dataset, we propose a time-aware service recommendation approach. Temporal information is sufficiently considered in our approach, where three time effects are analyzed and modeled including user bias shifting, Web service bias shifting, and user preference shifting. Experimental results show that the proposed approach outperforms seven existing collaborative filtering approaches on the prediction accuracy.

With the increasing number of Web services available on the Internet, how to recommend Web services to interested users effectively and efficiently remains to be a big challenge. At present, collaborative filtering (CF) is the most widely used technique in the design of recommender systems to handle information overload. For Web services, however, it is difficult for user to collect personalized QoS (Quality of Service)data and other explicit feedbacks such as ratings. In most cases, only a part of the implicit feedbacks (e.g., watchlist) is available in service registry. In this paper, we leverage implicit feedback from user's watchlist to build a CF-based recommender system for Web service. Our main contribution is to transform implicit feedbacks into explicit ratings to improve the accuracy of service recommendation. More specifically, we first construct binary user-service rating matrix according to the implicit feedback from the watchlist. Then, temporal and tag preference are combined into the original rating matrix to generate a more accurate pseudo rating matrix, which can reflect users' different preference on services in their own watchlists. Finally, we use traditional user-based CF method to produce a personalized service recommendation list with corresponding pseudo ratings. Moreover, the empirical experiments based on ProgrammableWeb show that compared with traditional log-based CF method, the recommender system with temporal and tag preference is more accurate and precise.

Image annotation has been widely investigated to discover the semantics of an image. However, most of the existing algorithms focus on noun tags (e.g. concepts and objects). Since an image is a snapshot of the real world event, annotating images with verbs will enable richer understanding of an image. In this paper, we propose a data-driven approach to verb oriented image annotation. At first, we obtain verb candidates by generating search queries for a given image with initial noun tags and establishing a sentence corpus from those queries. We utilize visualness to filter tags which are not visually presentable (e.g. pain) and differentiate tags into two categories (i.e. scene based and object based) to impose linguistic rules in verb extraction. Then we further re-rank the candidate verbs with the tag context discovered from the images which are both semantically and visually similar to the given image in the MIRFlickr dataset. Our experimental results from user study demonstrate that our proposed approach is promising.