Acute Complication Prediction and Diagnosis Model CLSTM-BPR:A Fusion Method of Time Series Deep Learning and Bayesian Personalized Ranking

Acute complication prediction model is of great importance for the overall reduction of premature death in chronic diseases.The CLSTM-BPR proposed in this paper aims to improve the accuracy,interpretability,and generalizability of the existing disease prediction models.Firstly,through its complex neural network structure,CLSTM-BPR considers both disease commonality and patient characteristics in the prediction process.Secondly,by splicing the time series prediction algorithm and classifier,the judgment basis is given along with the prediction results.Finally,this model introduces the pairwise algorithm Bayesian Personalized Ranking(BPR)into the medical field for the first time,and achieves a good result in the diagnosis of six acute complications.Experiments on the Medical Information Mart for Intensive Care IV(MIMIC-IV)dataset show that the average Mean Absolute Error(MAE)of biomarker value prediction of the CLSTM-BPR model is 0.26,and the average accuracy(ACC)of the CLSTM-BPR model for acute complication diagnosis is 92.5%.Comparison experiments and ablation experiments further demonstrate the reliability of CLSTM-BPR in the prediction of acute complication,which is an advancement of current disease prediction tools.

DP-BPR:Destination prediction based on Bayesian personalized ranking

Destination prediction has attracted widespread attention because it can help vehicle-aid systems recommend related services in advance to improve user driving experience.However,the relevant research is mainly based on driving trajectory of vehicles to predict the destinations,which is challenging to achieve the early destination prediction.To this end,we propose a model of early destination prediction,DP-BPR,to predict the destinations by users’travel time and locations.There are three challenges to accomplish the model:1)the extremely sparse historical data make it challenge to predict destinations directly from raw historical data;2)the destinations are related to not only departure points but also departure time so that both of them should be taken into consideration in prediction;3)how to learn destination preferences from historical data.To deal with these challenges,we map sparse high-dimensional data to a dense low-dimensional space through embedding learning using deep neural networks.We learn the embeddings not only for users but also for locations and time under the supervision of historical data,and then use Bayesian personalized ranking(BPR)to learn to rank destinations.Experimental results on the Zebra dataset show the effectiveness of DP-BPR.

Exploiting Geo-Social Correlations to Improve Pairwise Ranking for Point-of-Interest Recommendation

Recently, as location-based social network(LBSN) rapidly grow, point-of-interest(POI) recommendation has become an important way to help people locate interesting places. Nowadays, there have been deep studies conducted on the geographical and social influence in the point-of-interest recommendation model based on the rating prediction. The fact is, however, relying solely on the rating fails to reflect the user's preferences very accurately, because the users are most concerned with the list of ranked point-of-interests(POIs) on the actual output of recommender systems. In this paper, we propose a co-pairwise ranking model called Geo-Social Bayesian Personalized Ranking model(GSBPR), which is based on the pairwise ranking with the exploiting geo-social correlations by incorporating the method of ranking learning into the process of POI recommendation. In this model, we develop a novel BPR pairwise ranking assumption by injecting users' geo-social preference. Based on this assumption, the POI recommendation model is reformulated by a three-level joint pairwise ranking model. And the experimental results based on real datasets show that the proposed method in this paper enjoys better recommendation performance compared to other state-of-the-art POI recommendation models.

Social Trust Aware Item Recommendation for Implicit Feedback

Social trust aware recommender systems have been well studied in recent years. However, most of existing methods focus on the recommendation scenarios where users can provide explicit feedback to items. But in most cases, the feedback is not explicit but implicit. Moreover, most of trust aware methods assume the trust relationships among users are single and homogeneous, whereas trust as a social concept is intrinsically multi-faceted and heterogeneous. Simply exploiting the raw values of trust relations cannot get satisfactory results. Based on the above observations, we propose to learn a trust aware personalized ranking method with multi-faceted trust relations for implicit feedback. Specifically, we first introduce the social trust assumption -- a user＇s taste is close to the neighbors he/she trusts into the Bayesian Personalized Ranking model. To explore the impact of users＇ multi-faceted trust relations, we further propose a category- sensitive random walk method CRWR to infer the true trust value on each trust link. Finally, we arrive at our trust strength aware item recommendation method SocialBPRcawn by replacing the raw binary trust matrix with the derived real-valued trust strength. Data analysis and experimental results on two real-world datasets demonstrate the existence of social trust influence and the effectiveness of our social based ranking method SocialBPRcawR in terms of AUC （area under the receiver operating characteristic curve）.