Exploiting Positional Information for Session-Based Recommendation

For present e-commerce platforms, it is important to accurately predict users’ preference for a timely next-item recommendation. To achieve this goal, session-based recommender systems are developed, which are based on a sequence of the most recent user-item interactions to avoid the influence raised from outdated historical records. Although a session can usually reflect a user’s current preference, a local shift of the user’s intention within the session may still exist. Specifically, the interactions that take place in the early positions within a session generally indicate the user’s initial intention, while later interactions are more likely to represent the latest intention. Such positional information has been rarely considered in existing methods, which restricts their ability to capture the significance of interactions at different positions. To thoroughly exploit the positional information within a session, a theoretical framework is developed in this paper to provide an in-depth analysis of the positional information. We formally define the properties of forward-awareness and backward-awareness to evaluate the ability of positional encoding schemes in capturing the initial and the latest intention. According to our analysis, existing positional encoding schemes are generally forward-aware only, which can hardly represent the dynamics of the intention in a session. To enhance the positional encoding scheme for the session-based recommendation, a dual positional encoding (DPE) is proposed to account for both forward-awareness and backward-awareness . Based on DPE, we propose a novel Positional Recommender (PosRec) model with a well-designed Position-aware Gated Graph Neural Network module to fully exploit the positional information for session-based recommendation tasks. Extensive experiments are conducted on two e-commerce benchmark datasets, Yoochoose and Diginetica and the experimental results show the superiority of the PosRec by comparing it with the state-of-the-art session-based recommender models.

Entity recognition of Chinese medical text based on multi-head self-attention combined with BILSTM-CRF

<abstract> <p>Named entities are the main carriers of relevant medical knowledge in Electronic Medical Records (EMR). Clinical electronic medical records lead to problems such as word segmentation ambiguity and polysemy due to the specificity of Chinese language structure, so a Clinical Named Entity Recognition (CNER) model based on multi-head self-attention combined with BILSTM neural network and Conditional Random Fields is proposed. Firstly, the pre-trained language model organically combines char vectors and word vectors for the text sequences of the original dataset. The sequences are then fed into the parallel structure of the multi-head self-attention module and the BILSTM neural network module, respectively. By splicing the output of the neural network module to obtain multi-level information such as contextual information and feature association weights. Finally, entity annotation is performed by CRF. The results of the multiple comparison experiments show that the structure of the proposed model is very reasonable and robust, and it can effectively improve the Chinese CNER model. The model can extract multi-level and more comprehensive text features, compensate for the defect of long-distance dependency loss, with better applicability and recognition performance.</p> </abstract>

Oilseed Rape Shares Abundant and Generalized Pollinators with Its Co-Flowering Plant Species

Mass-flowering crops, such as Oilseed Rape (OSR), provide resources for pollinators and benefit from pollination services. Studies that observe the community of interactions between plants and pollinators are critical to understanding the resource needs of pollinators. We observed pollinators on OSR and wild plants in adjacent semi-natural areas in Sachsen-Anhalt, Germany to quantify (1) the co-flowering plants that share pollinators with OSR, (2) the identity and functional traits of plants and pollinators in the network module of OSR, and (3) the identity of the plants and pollinators that act as network connectors and hubs. We found that four common plants share a high percentage of their pollinators with OSR. OSR and these plants all attract abundant pollinators in the community, and the patterns of sharing were not more than would be expected by chance sampling. OSR acts as a module hub, and primarily influences the other plants in its module that have similar functional traits. However, the plants that most influence the pollination of OSR have different functional traits and are part of different modules. Our study demonstrates that supporting the pollination of OSR requires the presence of semi-natural areas with plants that can support a high abundances of generalist pollinators.

Application of U-Net with Global Convolution Network Module in Computer-Aided Tongue Diagnosis

The rapid development of intelligent manufacturing provides strong support for the intelligent medical service ecosystem. Researchers are committed to building Wise Information Technology of 120 (WIT 120) for residents and medical personnel with the concept of simple smart medical care and through core technologies such as Internet of Things, Big Data Analytics, Artificial Intelligence, and microservice framework, to improve patient safety, medical quality, clinical efficiency, and operational benefits. Among them, how to use computers and deep learning technology to assist in the diagnosis of tongue images and realize intelligent tongue diagnosis has become a major trend. Tongue crack is an important feature of tongue states. Not only does change of tongue crack states reflect objectively and accurately changed circumstances of some typical diseases and TCM syndrome but also semantic segmentation of fissured tongue can combine the other features of tongue states to further improve tongue diagnosis systems’ identification accuracy. Although computer tongue diagnosis technology has made great progress, there are few studies on the fissured tongue, and most of them focus on the analysis of tongue coating and body. In this paper, we do systematic and in-depth researches and propose an improved U-Net network for image semantic segmentation of fissured tongue. By introducing the Global Convolution Network module into the encoder part of U-Net, it solves the problem that the encoder part is relatively simple and cannot extract relatively abstract high-level semantic features. Finally, the method is verified by experiments. The improved U-Net network has a better segmentation effect and higher segmentation accuracy for fissured tongue image dataset. It can be used to design a computer-aided tongue diagnosis system.

Agent Clustering Strategy Based on Metabolic Flux Distribution and Transcriptome Expression for Novel Drug Development

The network module-based method has been used for drug repositioning. The traditional drug repositioning method only uses the gene characteristics of the drug but ignores the drug-triggered metabolic changes. The metabolic network systematically characterizes the connection between genes, proteins, and metabolic reactions. The differential metabolic flux distribution, as drug metabolism characteristics, was employed to cluster the agents with similar MoAs (mechanism of action). In this study, agents with the same pharmacology were clustered into one group, and a total of 1309 agents from the CMap database were clustered into 98 groups based on differential metabolic flux distribution. Transcription factor (TF) enrichment analysis revealed the agents in the same group (such as group 7 and group 26) were confirmed to have similar MoAs. Through this agent clustering strategy, the candidate drugs which can inhibit (Japanese encephalitis virus) JEV infection were identified. This study provides new insights into drug repositioning and their MoAs.

Drug repurposing improves disease targeting 11-fold and can be augmented by network module targeting, applied to COVID-19

This analysis presents a systematic evaluation of the extent of therapeutic opportunities that can be obtained from drug repurposing by connecting drug targets with disease genes. When using FDA-approved indications as a reference level we found that drug repurposing can offer an average of an 11-fold increase in disease coverage, with the maximum number of diseases covered per drug being increased from 134 to 167 after extending the drug targets with their high confidence first neighbors. Additionally, by network analysis to connect drugs to disease modules we found that drugs on average target 4 disease modules, yet the similarity between disease modules targeted by the same drug is generally low and the maximum number of disease modules targeted per drug increases from 158 to 229 when drug targets are neighbor-extended. Moreover, our results highlight that drug repurposing is more dependent on target proteins being shared between diseases than on polypharmacological properties of drugs. We apply our drug repurposing and network module analysis to COVID-19 and show that Fostamatinib is the drug with the highest module coverage.

Expandable Substrate Module and Storage Network Compatible w/ BitBadges

(THIS IS THE MIRROR OF A GRANT PROPOSAL). We are working on creating a storage network module on substrate compatible w/ the BitBadges blockchain. This will allow for substrate to interact in a meaningful way with the BitBadges ecosystem, which follows an experimental PoCP or Proof of Computation proof.Other projects like Crust or Subspace are within the same realm of trying to create some form of storage network. The difference is how BitBadges goes about accomplishing this and what a substrate module as an integration will provide. BitBadges as a whole is working on offline-centric networks and integrating variations of distributing sharding algorithms. Their data is transmitted w/ CouchDB being an integral part of its core for node syncronization. BitBadges is also integrating w/ various third party networks viewed as either pegs or applications.We believe that the substrate ecosystem can provide another integration w/ our network and we might even be able to create something similar to what Crust is doing on their mainnet, but w/ BitBadges as the storage method or a mirrored peg for the data. Our team, which is currently just two people, are passionate and have years of blockchain development experience under our belts. We think Substrate can work great as one of our third party integrations or pegged chains.

Feature Fusion-Based Siamese Region Proposal Network for Ultrasound Tracking

Object tracking based on ultrasound image navigation can effectively reduce damage to healthy tissues in radiotherapy. In this study, we propose a deep Siamese network based on feature fusion. Whilst adopting MobileNetV2 as the backbone, an unsupervised training strategy is introduced to enrich the volume of the samples. The region proposal network module is designed to predict the location of the target, and a non-maximum suppression-based postprocessing algorithm is designed to refine the tracking results. Moreover, the proposed method is evaluated in the Challenge on Liver Ultrasound Tracking dataset and the self-collected dataset, which proves the need for the improvement and the effectiveness of the algorithm.

Design of Deep Learning Model for Task-Evoked fMRI Data Classification

Machine learning methods have been successfully applied to neuroimaging signals, one of which is to decode specific task states from functional magnetic resonance imaging (fMRI) data. In this paper, we propose a model that simultaneously utilizes characteristics of both spatial and temporal sequential information of fMRI data with deep neural networks to classify the fMRI task states. We designed a convolution network module and a recurrent network module to extract the spatial and temporal features of fMRI data, respectively. In particular, we also add the attention mechanism to the recurrent network module, which more effectively highlights the brain activation state at the moment of reaction. We evaluated the model using task-evoked fMRI data from the Human Connectome Project (HCP) dataset, the classification accuracy got 94.31%, and the experimental results have shown that the model can effectively distinguish the brain states under different task stimuli.