Are Topics Interesting or Not? An LDA-based Topic-graph Probabilistic Model for Web Search Personalization

Web content topic modeling using LDA and HTML tags

An immense volume of digital documents exists online and offline with content that can offer useful information and insights. Utilizing topic modeling enhances the analysis and understanding of digital documents. Topic modeling discovers latent semantic structures or topics within a set of digital textual documents. The Internet of Things, Blockchain, recommender system, and search engine optimization applications use topic modeling to handle data mining tasks, such as classification and clustering. The usefulness of topic models depends on the quality of resulting term patterns and topics with high quality. Topic coherence is the standard metric to measure the quality of topic models. Previous studies build topic models to generally work on conventional documents, and they are insufficient and underperform when applied to web content data due to differences in the structure of the conventional and HTML documents. Neglecting the unique structure of web content leads to missing otherwise coherent topics and, therefore, low topic quality. This study aims to propose an innovative topic model to learn coherence topics in web content data. We present the HTML Topic Model (HTM), a web content topic model that takes into consideration the HTML tags to understand the structure of web pages. We conducted two series of experiments to demonstrate the limitations of the existing topic models and examine the topic coherence of the HTM against the widely used Latent Dirichlet Allocation (LDA) model and its variants, namely the Correlated Topic Model, the Dirichlet Multinomial Regression, the Hierarchical Dirichlet Process, the Hierarchical Latent Dirichlet Allocation, the pseudo-document based Topic Model, and the Supervised Latent Dirichlet Allocation models. The first experiment demonstrates the limitations of the existing topic models when applied to web content data and, therefore, the essential need for a web content topic model. When applied to web data, the overall performance dropped an average of five times and, in some cases, up to approximately 20 times lower than when applied to conventional data. The second experiment then evaluates the effectiveness of the HTM model in discovering topics and term patterns of web content data. The HTM model achieved an overall 35% improvement in topic coherence compared to the LDA.

Predicting the Mortality of ICU Patients by Topic Model with Machine-Learning Techniques

Predicting clinical patients’ vital signs is a leading critical issue in intensive care units (ICUs) related studies. Early prediction of the mortality of ICU patients can reduce the overall mortality and cost of complication treatment. Some studies have predicted mortality based on electronic health record (EHR) data by using machine learning models. However, the semi-structured data (i.e., patients’ diagnosis data and inspection reports) is rarely used in these models. This study utilized data from the Medical Information Mart for Intensive Care III. We used a Latent Dirichlet Allocation (LDA) model to classify text in the semi-structured data of some particular topics and established and compared the classification and regression trees (CART), logistic regression (LR), multivariate adaptive regression splines (MARS), random forest (RF), and gradient boosting (GB). A total of 46,520 ICU Patients were included, with 11.5% mortality in the Medical Information Mart for Intensive Care III group. Our results revealed that the semi-structured data (diagnosis data and inspection reports) of ICU patients contain useful information that can assist clinical doctors in making critical clinical decisions. In addition, in our comparison of five machine learning models (CART, LR, MARS, RF, and GB), the GB model showed the best performance with the highest area under the receiver operating characteristic curve (AUROC) (0.9280), specificity (93.16%), and sensitivity (83.25%). The RF, LR, and MARS models showed better performance (AUROC are 0.9096, 0.8987, and 0.8935, respectively) than the CART (0.8511). The GB model showed better performance than other machine learning models (CART, LR, MARS, and RF) in predicting the mortality of patients in the intensive care unit. The analysis results could be used to develop a clinically useful decision support system.