Session-based recommendation is a challenging task. Without access to user's historical user-item interactions, the information available in an ongoing session may be very limited. Previous work on session-based has considered sequences of items that users have interacted with sequentially. Such item not fully capture complex transition relationship between go beyond inspection order. Thus graph neural network (GNN) based models been proposed items. However, GNNs typically propagate from...

We propose a Joint Neural Collaborative Filtering (J-NCF) method for recommender systems. The J-NCF model applies joint neural network that couples deep feature learning and interaction modeling with rating matrix. Deep extracts representations of users items architecture based on user-item captures non-linear interactions using the generated by process as input. enables processes to optimize each other through training, which leads improved recommendation performance. In addition, we design...

Session-based recommendation (SBR) , which mainly relies on a user’s limited interactions with items to generate recommendations, is widely investigated task. Existing methods often apply RNNs or GNNs model sequential behavior transition relationship between capture her current preference. For training such models, the supervision signals are merely generated from inside session, neglecting correlations of different sessions, we argue can provide additional supervisions for learning item...

Sequential recommenders that capture users' dynamic intents by modeling sequential behavior, are able to accurately recommend items users. Previous studies on recommendations (SRs) mostly focus optimizing the recommendation accuracy, thus ignoring diversity of recommended items. Many existing methods for improving not applicable SRs because they assume user static and rely post-processing list promote diversity. We consider both accuracy reformulating as a generation task, propose an...

Session-based recommendation is the task of recommending next item a user might be interested in given partially known session information, e.g., part or recent historical sessions. An effective session-based recommender should able to exploit user's evolving preferences, which we assume mixture her short- and long-term interests. Existing methods often embed preference into static representation, plays fixed role when dealing with current short-term This problematic because preferences may...

Sequential recommenders capture dynamic aspects of users’ interests by modeling sequential behavior. Previous studies on recommendations mostly aim to identify main recent optimize the recommendation accuracy; they often neglect fact that users display multiple over extended periods time, which could be used improve diversity lists recommended items. Existing work related diversified typically assumes preferences are static and depend post-processing candidate list However, those conditions...

In recommender systems, leveraging auxiliary behaviors (e.g. view, cart) to enhance the recommendation in target behavior purchase) is crucial for mitigating sparsity issue inherent single-behavior recommendation. This has given rise multi-behavior (MBR). Existing MBR task faces two primary challenges. First, irrelevant that do not align with behavior, can negatively impact prediction accuracy user preference behavior. Second, these methods typically learn coarse-grained preferences, failing...

Link prediction on temporal networks aims to predict the future edges by modeling dynamic evolution involved in graph data. Previous methods relying node/edge attributes or distance structure are not practical due deficiency of and limitation explicit estimation, respectively. Moreover, existing representation learning mostly rely neural (GNNs), which cannot adequately take correlations between nodes into consideration, leading generating inferior node embeddings. Thus, we propose a...

Query suggestions help users of a search engine to refine their queries. Previous work on query suggestion has mainly focused incorporating directly observable features such as co-occurrence and semantic similarity. The structure is often set manually, result which hidden dependencies between queries may be ignored. We propose an AHNQS model that combines hierarchical with session-level neural network user-level the short- long-term history user. An attention mechanism used capture user...

Session-based recommendation aims to generate recommendations merely based on the ongoing session, which is a challenging task. Previous methods mainly focus modeling sequential signals or transition relations between items in current session using RNNs GNNs identify user’s intent for recommendation. Such models generally ignore dynamic connections local and global item patterns, although information taken into consideration by exploiting global-level pair-wise transitions. Moreover,...

Query suggestions help users refine their queries after they input an initial query. We consider the task of generating query that are personalized and diversified. propose a suggestion diversification model (PQSD), where user's long-term search behavior is injected into basic greedy (G-QSD) considers context in current session. aspects identified through clicked documents based on Open Directory Project (ODP). quantify improvement PQSD over state-of-the-art baseline using AOL log show it...

Event detection classifies unlabeled sentences into event labels, which can benefit numerous applications, including information retrieval, question answering and script learning. One of the major obstacles to in reality is insufficient training data. To deal with low-resources problem, we investigate few-shot this paper propose TaLeM, a novel taxonomy-aware learning model, consisting two components, i.e., self-supervised framework (TaSeLF) prototypical networks (TaPN). Specifically, TaSeLF...

Query auto-completion (QAC) is the first step of information retrieval, which helps users formulate entire query after inputting only a few prefixes. Regarding models QAC, traditional method ignores contribution from semantic relevance between queries. However, similar queries always express extremely search intention. In this paper, we propose hybrid model FS-QAC based on similarity as well frequency. We choose word2vec to measure intended and pre-submitted By combining both features, our...