Conventional retrieval-augmented generation(RAG) systems employ rigid retrieval strategies that create: (1) knowledge blind spots across domain boundaries, (2) reasoning fragmentation when processing interdependent concepts, and (3) contradictions from conflicting evidence sources. Motivated by these limitations, we introduce PathwiseRAG, which addresses these challenges through: intent-aware strategy selection to eliminate blind spots, dynamic reasoning networks that capture sub-problem interdependencies to overcome fragmentation, and parallel path exploration with adaptive refinement to resolve conflicts. The framework models query intent across semantic and reasoning dimensions, constructs a directed acyclic graph of interconnected sub-problems, and explores multiple reasoning trajectories while continuously adapting to emerging evidence. Evaluation across challenging benchmarks demonstrates significant improvements over state-of-the-art RAG systems, with average accuracy gains of 4.9% and up to 6.9% on complex queries, establishing a new paradigm for knowledge-intensive reasoning by transforming static retrieval into dynamic, multi-dimensional exploration.

Social network services (SNSs) provide platforms where users engage in social link behavior (e.g., predicting social relationships) and consumption behavior. Recent advancements in deep learning for recommendation and link prediction explore the symbiotic relationships between these behaviors, leveraging social influence theory and user homogeneity, i.e., users tend to accept recommendations from social friends and connect with like-minded users. These studies yield positive feedback for users and platforms, fostering practical applications and economic development. While previous works jointly model these behaviors, most studies often overlook the evolution of social relationships and users’ preferences in dynamic scenes and the correlations inside, as well as the higher order information within the social network and preference network (consumption history). To address this, we propose the dynamic graph neural joint behavior prediction model (DGN-JBP). Specifically, we actively disentangle and initialize user embeddings from multiple perspectives to refine information for modeling. Additionally, we design an attentive graph neural network and combine it with gate recurrent units (GRUs) to extract high-order dynamic information. Finally, we design a dual framework and purposefully fuse embeddings to mutually enhance the effectiveness of predictions on two prediction tasks. Extensive experimental results on two real-world datasets clearly demonstrate the effectiveness of our proposed model.

In recent years, context-aware recommender systems (CARSs), which incorporate contextual information to achieve better recommendations, become a hot topic in the domain of recommender systems. Many context-aware recommendation methods have been proposed in the past decades. Some literatures provide survey of the research on CARSs. However, they mainly focus on context-aware recommendation methods and overlook the evaluation of them. The evaluation methods, evaluation properties and datasets of CARSs are different from those of traditional recommender systems where contexts are not considered. In this paper, we provide a review for evaluation of CARSs. We will introduce the basic concepts of CARSs, propose a new dataset partition method for each category of CARSs according to our classification of CARSs, summarize the evaluation method. Then we summarize the evaluation properties that CARSs pays attention to, which are different from the NCARSs. In addition, we also summarize the datasets specifically for CARSs, its applicable CARSs type and its evaluation dimensions and metrics. Based on our review, we draw some conclusions from evaluation perspective and point out future research directions.

Newly emerging Event-based Social Network (EBSN) concentrates on connecting both online social relationships and offline local events. For the growing amount of events published on EBSNs, personalized event recommendation becomes essential to help users choose attractive events. But most of existing event recommendation algorithms fail to distinguish constraint factors of users’ event participation behaviors from preference factors, which reflects the cost of event participation that hinders users from attending interested events. To take full advantage of influences from contextual information on users’ event participation, we differentiate preference and constraint factors which contribute to users’ decision for event participation, and extract the soft spatial and temporal constraints from event venue and start time contexts respectively. Then we propose the Preference and Constraint Factor Model (PCFM) based on factorization machine model, using attentive mechanism to weight feature interactions and incorporate latent factors of users and contextual features for personalized perference modeling and event recommendation. Moreover, learning-to-rank techniques are utilized to train PCFM as a ranking model for the implicit feedback nature of responses from users. Extensive experiments evaluate the performance of our proposed recommendation model on real-world EBSN datasets, and demonstrate the outperformance than state-of-art event recommendation methods on many metrics.

Event recommendation is essential to help people find attractive events to attend, but it intrinsically faces cold-start problem. The previous studies exploit multiple contextual factors to overcome the cold-start problem in event recommendation. However, they do not consider the correlation among different contextual factors. Moreover, suggesting events for a group of users also has not been well studied. In this paper, we first discover the correlation between organizer and textual content, i.e., the events held by the same organizer tend to have more similar content. Based on this observation, we present a content-venue-aware topic model (CVTM) to capture group interests on an event from two perspectives: content and venue. The correlation between organizer and content is modeled in CVTM to alleviate the sparsity of textual content, and then we can further extract group interests on content of an event more accurately. Finally, a group event recommendation method using CVTM is proposed. We conduct comprehensive experiments to evaluate the recommendation performance of our model on two real-world datasets. The results demonstrate that the proposed model outperforms the state-of-the-art methods that suggest upcoming events for groups. Besides, CVTM can learn semantically coherent latent topics which are useful to explain recommendations.

Event recommendation is an essential means to enable people to find attractive upcoming social events, such as party, exhibition, and concert. While growing line of research has focused on suggesting events to individuals, making event recommendation for a group of users has not been well studied. In this paper, we aim to recommend upcoming events for a group of users. We formalize group recommendation as a ranking problem and propose a group event recommendation framework GERF based on learning-to-rank technique. Specifically, we first analyze different contextual influences on user's event attendance, and extract preference of user to event considering each contextual influence. Then, the preference scores of the users in a group are taken as the features for learningto-rank to model the preference of the group. Moreover, a fast pairwise learning-to-rank algorithm, Bayesian group ranking, is proposed to learn ranking model for each group. Our framework is easily to incorporate additional contextual influences, and can be applied to other group recommendation scenarios. Extensive experiments have been conducted to evaluate the performance of GERF on two real-world datasets and demonstrate the appealing performance of our method on both accuracy and time efficiency.