Personalized exercise recommendation systems aim to enhance learning efficiency by dynamically guiding students toward content aligned with their evolving knowledge states. Among various approaches, Reinforcement Learning (RL) has emerged as an effective framework for modeling student-environment interactions as sequential decision-making processes. However, most existing RL-based methods typically reward recommendations that target unmastered knowledge concepts.Such reward-driven strategies are prone to local optima, restricting exploration of unattempted or less familiar knowledge concepts. In addition, they often overlook key learning factors, such as forgetting dynamics and exercise difficulty, leading to suboptimal outcomes. To address these issues, we propose a novelIntegratingEntropyRegulation andDual-objectiveOptimizationExerciseRecommendation (IERDO-ER)method. Specifically, we introduce an entropy-based function to encourage broader exploration in the exercise space. We also design an end-to-end policy network that generates candidate exercises as next-step actions for the recommendation agent. To enable more adaptive and pedagogically sound recommendations, we develop a dual-objective reward mechanism that incorporates both anti-forgetting and gentleness objectives. This mechanism continuously balances policy optimization through a synergy of rewards and penalties. Experiments on three real-world educational datasets show that our approach consistently outperforms strong baselines, validating its effectiveness.

Abstract Cardinality estimation is a crucial component in query optimizers. After decades of research, employing autoregressive models for cardinality estimation has demonstrated remarkable accuracy. However, when queries involve attributes with large domain sizes, autoregressive model-based estimators struggle to accurately capture the data distribution, leading to poor performance. Furthermore, these models often exhibit significant errors when handling queries with low-selectivity predicates. To address these challenges, we propose a self-adaptive cardinality estimator named AdaCard. Initially, we employ a self-adaptive smoothing factor selection strategy to variably adjust the original data, thereby mitigating the impact of large domain sizes. Secondly, to correct errors stemming from Monte Carlo sampling, we utilize resampling to refine the handling of low-selectivity predicates, thereby improving accuracy. Through evaluation using four real-world benchmarks, we compared AdaCard with mainstream baselines. The final results show that our estimator has the lowest tail estimation error and improves accuracy by nearly 10$\times $ over the second-best method, with similar latency and model size.

Abstract In recent years, with the rapid development of mobile applications, user check-in histories have been increasing. Successive point-of-interest (POI) recommendation has gained growing attention. Existing successive point-of-interest recommendation methods learn long- and short-term user preferences through historical check-in sequences to provide more personalized services. However, due to sparse data and complicated temporal patterns, the application of such technique is still limited by two challenges: 1) difficulty meeting user travel needs in time; 2) difficulty capturing users complicated behavior patterns. To address this problem, we propose a new Influence-Aware successive POI recommendation Model (InfAM), which can learn the influence of POIs in a short-term sequence fragment for next point-of-interest recommendation. To capture periodic patterns of user movements, InfAM takes a user’s check-in data within a day as an input sequence to address the current travel needs of the user. In addition, based on multihead attention mechanism and user embedding, InfAM focuses on the influence of POIs in short-term sequences and general user preferences in these sequences. Therefore, InfAM integrates three specific dependencies, which can fully learn the dynamic interaction between short-term preferences: the influence of POIs in short-term sequence fragments (POI-poi), user preferences (POI-user), and the periodicity of check-ins (POI-time). Evaluation results on real-world datasets show that InfAM achieves state-of-the-art recommendation performance.

The data for recommendations, usually a matrix composed of users and items, include a large number of missing data, noise data, etc, which have a negative effect on the accuracy of recommendations.In order to improve recommendation performance, this paper put forwards an improved model named Stacked Denoising AutoEncoder (SDAE), which improves the autoencoder by both indicators and denoising parts to construct an effective stacked autoencoding network.The first layer of the encoding network is responsible for dealing with missing data with the help of indicators and to get a new encoding for features, and then stacked denoising is applied to process noise data for a further optimization.SDAE's output can be accepted by collaborative filtering methods to provide a more accurate recommendation.Three data sets are used to verify the proposed model, the experimental results show that the proposed model presents an active ability on improving recommendation performance and mitigates the negative influence caused by missing data, noise data, etc.Index Terms-recommendation, data preprocessing, stacked encoding I. .

Extracting web content is to obtain the required data embedded in web pages, usually including structured records, such as product information, and text content, such as news. Web pages use a large number of HTML tags to organize and to present various information. Both knowing little about the structures of web pages and mixing kinds of information in web pages are making the extraction process very challenging to guarantee extraction performance and extraction adaptability. This study proposes a unified web content extraction framework that can be applied in various web environments to extract both structured records and text content. First, we construct a characteristic container to hold kinds of characteristics related with extraction objectives, including visual text information, content semantics(instead of HTML tag semantics), web page structures, etc. Second, the above characteristics are integrated into an extraction framework for extraction decisions on different web sites. Especially, we put forward different strategies, path aggregation for extracting text content and HMM model for structured records, to locate the extraction area by exploiting both those extraction characteristics. Comparative experiments on multiple web sites with popular extraction methods, including CETR, CETD and CNBE, show that our proposed extraction method can provide better extraction precision and extraction adaptability.

Detecting review spam is important for current e-commerce applications. However, the posted order of review has been neglected by the former work. In this paper, we explore the issue on fake review detection in review sequence, which is crucial for implementing online anti-opinion spam. We analyze the characteristics of fake reviews firstly. Based on review contents and reviewer behaviors, six time sensitive features are proposed to highlight the fake reviews. And then, we devise supervised solutions and a threshold-based solution to spot the fake reviews as early as possible. The experimental results show that our methods can identify the fake reviews orderly with high precision and recall.