To provide personalized recommendations for users, recommendation systems must predict their unknown preferences. Latent factor models consistently achieve high prediction accuracy for this task. During training, these models encode users and items as latent vectors. Aligning these vectors facilitates predicting the item preference of the user. The training process optimizes the objective function, aiming to minimize the disparity between latent vectors and the collected preferences. In this study, we aim to integrate item descriptions into the construction of the objective function due to the sparsity and inaccuracy of collected preferences. This process is accomplished using Bert for vectorizing item descriptions. We conducted experiments with the proposed approach on datasets Movielens 1M and Yahoo Webscope R4. Our approach demonstrates a reduction in RMSE compared to previous approaches within this research domain.

With the Bag-of-Words model, a document corpus can be originally represented by a Terms-Documents matrix. However, the high-dimensional pure Terms-Documents matrix needs transforming to a lower-dimensional semantic Concepts-Documents matrix in order to not only reduce the feature space dimension but also create more meaningful features. This paper analyzes two feature transformation (FT) models on the Terms-Documents matrix, i.e. the FT model based on Low-Rank Approximation (LRA) and the FT model based on Word Embedding (WE). Both of them have their unique strength and weakness in the text transformation. The LRA-based FT only focuses on the mathematical perspective to statistically cover the original dispersed term set of the corpus as well as possible, while the WE-based FT utilizes the available word embedding vectors to enhance the contextual content of the corpus presentation. Therefore, the combinations of the LRA-based FT and the WE-based FT, named LRAintoWE-based FT and WEintoLRA-based FT, are possibly proposed to obtain comprehensive FTs capturing appropriately both the statistical information and the contextual information. The experiment results on three benchmark datasets show that the information of the WE-based FT and the LRA-based FT can be integrated, and their integration as LRAintoWE-based FT and WEintoLRA-based FT can improve the classification performance compared with that based on only either of them.

The bag-of-words technique is often used to present a document in text categorization. However, for a large set of documents where the dimension of the bag-of-words vector is very high, text categorization becomes a serious challenge as a result of sparse data, over-fitting, and irrelevant features. A filter feature selection method reduces the number of features by eliminating irrelevant features from the bag-of-words vector. In this paper, we analyze the weak points and strong points of two filter feature selection approaches which are the frequency-based approach and the cluster-based approach. Thanks to the analysis, we propose hybrid filter feature selection methods, named the Frequency-Cluster Feature Selection (FCFS) and the Detailed Frequency-Cluster Feature Selection (DtFCFS), to further improve the performance of the filter feature selection process in text categorization. The FCFS is a combination of the Frequency-based approach and the Cluster-based approach, while the DtFCFS, a detailed version of the FCFS, is a comprehensively hybrid clusterbased method. We do experiments with four benchmark datasets (the Reuters-21578 and Newsgroup dataset for news classification, the Ohsumed dataset for medical document classification, and the LingSpam dataset for email classification) to compare the proposed methods with six related wellknown methods such as the Comprehensive Measurement Feature Selection (CMFS), the Optimal Orthogonal Centroid Feature Selection (OCFS), the Crossed Centroid Feature Selection (CIIC), the Information Gain (IG), the Chi-square (CHI), and the Deviation from Poisson Feature Selection (DFPFS). In terms of the Micro-F1, the Macro-F1, and the dimension reduction rate, the DtFCFS is superior to the other methods, while the FCFS shows competitive and even superior performance to the good methods, especially for the Macro-F1.

For a large set of documents, bag-of-words vector can reach thousands of features. Document classification faces many difficulties in high dimensionality of bag-of-words vector. High dimensionality not only increases computation cost but also degrades the accuracy of classification process. The aim of filter feature selection is to remove irrelevant features by selecting a subset of the original feature set. In this paper, we analyze two filter feature selection approaches which are the frequency-based approach and the cluster-based approach. We propose a hybrid filter Feature Selection method for the combination of these approaches, named FCFS, in order to exploit their strong points. We experiment on FCFS and related filter feature selection methods as CMFS, OCFS, CIIC, IG, CHI with two datasets about news and medicine. Regarding Macro-F1, FCFS is superior to the other methods, while FCFS shows comparable and even better performance than the other methods in term of Micro-F1