PubMed^® is an essential resource for the medical domain, but useful concepts are either difficult to extract or are ambiguous, which has significantly hindered knowledge discovery. To address this issue, we constructed a PubMed knowledge graph (PKG) by extracting bio-entities from 29 million PubMed abstracts, disambiguating author names, integrating funding data through the National Institutes of Health (NIH) ExPORTER, collecting affiliation history and educational background of authors from ORCID^®, and identifying fine-grained affiliation data from MapAffil. Through the integration of these credible multi-source data, we could create connections among the bio-entities, authors, articles, affiliations, and funding. Data validation revealed that the BioBERT deep learning method of bio-entity extraction significantly outperformed the state-of-the-art models based on the F1 score (by 0.51%), with the author name disambiguation (AND) achieving an F1 score of 98.09%. PKG can trigger broader innovations, not only enabling us to measure scholarly impact, knowledge usage, and knowledge transfer, but also assisting us in profiling authors and organizations based on their connections with bio-entities.

Advances in machine learning and deep learning methods, together with the increasing availability of large-scale pharmacological, genomic, and chemical datasets, have created opportunities for identifying potentially useful relationships within biochemical networks. Knowledge embedding models have been found to have value in detecting knowledge-based correlations among entities, but little effort has been made to apply them to networks of biochemical entities. This is because such networks tend to be unbalanced and sparse, and knowledge embedding models do not work well on them. However, to some extent, the shortcomings of knowledge embedding models can be compensated for if they are used in association with graph embedding. In this paper, we combine knowledge embedding and graph embedding to represent biochemical entities and their relations as dense and low-dimensional vectors. We build a cascade learning framework which incorporates semantic features from the knowledge embedding model, and graph features from the graph embedding model, to score the probability of linking. The proposed method performs noticeably better than the models with which it is compared. It predicted links and entities with an accuracy of 93%, and its average hits@10 score has an average of 8.6% absolute improvement compared with original knowledge embedding model, 1.1% to 9.7% absolute improvement compared with other knowledge and graph embedding algorithm. In addition, we designed a meta-path algorithm to detect path relations in the biomedical network. Case studies further verify the value of the proposed model in finding potential relationships between diseases, drugs, genes, treatments, etc. Amongst the findings of the proposed model are the suggestion that VDR (vitamin D receptor) may be linked to prostate cancer. This is backed by evidence from medical databases and published research, supporting the suggestion that our proposed model could be of value to biomedical researchers.

Purpose: Internet technology allows millions of people to find high quality medical resources online, with the result that personal healthcare and medical services have become one of the fastest growing markets in China. Data relating to healthcare search behavior may provide insights that could lead to better provision of healthcare services. However, discrepancies often arise between terminologies derived from professional medical domain knowledge and the more colloquial terms that users adopt when searching for information about ailments. This can make it difficult to match healthcare queries with doctors’ keywords in online medical searches. The paper aims to discuss these issues. Design/methodology/approach: To help address this problem, the authors propose a transfer learning using latent factor graph (TLLFG), which can learn the descriptions of ailments used in internet searches and match them to the most appropriate formal medical keywords. Findings: Experiments show that the TLLFG outperforms competing algorithms in incorporating both medical domain knowledge and patient-doctor Q&A data from online services into a unified latent layer capable of bridging the gap between lay enquiries and professionally expressed information sources, and make more accurate analysis of online users’ symptom descriptions. The authors conclude with a brief discussion of some of the ways in which the model may support online applications and connect offline medical services. Practical implications: The authors used an online medical searching application to verify the proposed model. The model can bridge users’ long-tailed description with doctors’ formal medical keywords. Online experiments show that TLLFG can significantly improve the searching experience of both users and medical service providers compared with traditional machine learning methods. The research provides a helpful example of the use of domain knowledge to optimize searching or recommendation experiences. Originality/value: The authors use transfer learning to map online users’ long-tail queries onto medical domain knowledge, significantly improving the relevance of queries and keywords in a search system reliant on sponsored links.

Sales forecasting is a highly practical application of time series prediction. It is used to help enterprises identify and utilize information to reduce costs and maximize profits. For example, in numerous manufacturing enterprises, sales forecasting serves as a key indicator for inventory optimization and directly influences the level of cost savings. However, existing research methods mainly focus on detecting sequences and local correlations from multivariate time series (MTS), but seldom consider modeling the distinct information among the time series within MTS. The prediction accuracy of sales time series is significantly influenced by the dynamic and complex environment, so identifying the distinct signals between different time series within a sales MTS is more important. In order to extract more valuable information from sales series and to enhance the accuracy of sales prediction, we devised a universality–distinction mechanism (UDM) framework that can predict future multi-step sales. Universality represents the instinctive features of sequences and correlation patterns of sales with similar contexts. Distinction corresponds to the fluctuations in a specific time series due to complex or unobserved influencing factors. In the mechanism, a query-sparsity measurement (QSM)-based attention calculation method is proposed to improve the efficiency of the proposed model in processing large-scale sales MTS. In addition, to improve the specific decision-making scenario of inventory optimization and ensure stable accuracy in multi-step prediction, we use a joint Pin-DTW (Pinball loss and Dynamic Time Warping) loss function. Through experiments on the public Cainiao dataset, and via our cooperation with Galanz, we are able to demonstrate the effectiveness and practical value of the model. Compared with the best baseline, the improvements are 57.27%, 50.68%, and 35.26% on the Galanz dataset and 16.58%, 6.07%, and 5.27% on the Cainiao dataset, in terms of the MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error), and RMSE (Root Mean Squared Error).