With the increasing demand for personalization in modern digital ecosystems, Recommender Systems (RS) have become essential for delivering user-centric services. However, traditional RS approaches, which are primarily driven by algorithmic optimization over historical interaction data, still face well-known challenges such as the cold-start problem, limited interpretability, and insufficient personalization quality. Meanwhile, the rapid evolution of Large Language Models (LLMs) has showcased remarkable capabilities in semantic understanding, reasoning, and generalization, presenting a promising opportunity to rethink the design of recommendation paradigms. A central research question is how to systematically leverage LLMs to address fundamental RS challenges while avoiding unnecessary complexity, inefficiency, or unintended consequences. In this paper, we provide the first problem-driven synthesis of LLM-empowered RS. We propose a comprehensive taxonomy that organizes existing integration efforts into four major problem domains: Cold-start problem, poor interpretability, suboptimal user experience, and novel challenges introduced by LLMs themselves (e.g., hallucination, bias, and inefficiency). We begin by formally characterizing each problem domain and reviewing its significance in recommendation research. We then categorize recent advances by methodological perspective. Finally, we offer insights and future directions for building adaptive, efficient, and value-aligned RS enhanced by LLM capabilities.

Sequence-based Point-Of-Interest (POI) recommendations are increasingly crucial for location-based services and social platforms, offering nuanced insights into predicting user preferences from historical interaction patterns. However, a significant challenge arises from the non-uniform distribution of user-POI interaction sequences, where user preferences are often obscured by irregular and sporadic activities. This paper proposes an innovative Uniform Sequence Balancing (USB) strategy, addressing the critical issue of non-uniform sequences by utilizing the standard deviation of time intervals to achieve uniformity. Our approach transforms non-uniform sequences into uniform ones, thereby facilitating more accurate preference capture. We leverage the Transformer eXtra Long (Transformer-XL) model, known for its ability to discern long-term dependencies, and integrate it with our USB strategy to propose the Sequential Transformer-XL Recommender (STR). Our comprehensive experiments on two widely used public datasets demonstrate the effectiveness of STR, which significantly outperforms state-of-the-art models. The proposed STR not only optimizes recommendation performance but also paves the way for future research on sequence-based recommendation systems.

Lithological facies classification is a pivotal task in petroleum geology, underpinning reservoir characterization and influencing decision-making in exploration and production operations. Traditional classification methods, such as support vector machines and Gaussian process classifiers, often struggle with the complexity and nonlinearity of geological data, leading to suboptimal performance. Moreover, numerous prevalent approaches fail to adequately consider the inherent dependencies in the sequence of measurements from adjacent depths in a well. A novel approach leveraging an attention-based gated recurrent unit (AGRU) model is introduced in this paper to address these challenges. The AGRU model excels by exploiting the sequential nature of well-log data and capturing long-range dependencies through an attention mechanism. This model enables a flexible and context-dependent weighting of different parts of the sequence, enhancing the discernment of key features for classification. The proposed method was validated on two publicly available datasets. Results demonstrate a considerably improvement over traditional methods. Specifically, the AGRU model achieved superior performance metrics considering precision, recall, and F1-score.

As the main parent and guardian, mothers are often concerned with the study performance of their children. More specifically, most mothers are eager to know the concrete examination scores of their children. However, with the continuous progress of modern education systems, most schools or teachers have now been forbidden to release sensitive student examination scores to the public due to privacy concerns, which has made it infeasible for mothers to know the real study level or examination performance of their children. Therefore, a conflict has come to exist between teachers and mothers, which harms the general growing up of students in their study. In view of this challenge, we propose a Privacy-aware Examination Results Ranking (PERR) method to attempt at balancing teachers’ privacy disclosure concerns and the mothers’ concerns over their children’s examination performance. By drawing on a relevant case study, we prove the effectiveness of the proposed PERR method in evaluating and ranking students according to their examination scores while at the same time securing sensitive student information.