Topic-Enhanced Memory Networks for Personalised Point-of-Interest Recommendation

Abstract

Point-of-Interest (POI) recommender systems play a vital role in people's lives by recommending unexplored POIs to users and have drawn extensive attention from both academia and industry. Despite their value, however, they still suffer from the challenges of capturing complicated user preferences and fine-grained user-POI relationship for spatio-temporal sensitive POI recommendation. Existing recommendation algorithms, including both shallow and deep approaches, usually embed the visiting records of a user into a single latent vector to model user preferences: this has limited power of representation and interpretability. In this paper, we propose a novel topic-enhanced memory network (TEMN), a deep architecture to integrate the topic model and memory network capitalising on the strengths of both the global structure of latent patterns and local neighbourhood-based features in a nonlinear fashion. We further incorporate a geographical module to exploit user-specific spatial preference and POI-specific spatial influence to enhance recommendations. The proposed unified hybrid model is widely applicable to various POI recommendation scenarios. Extensive experiments on real-world WeChat datasets demonstrate its effectiveness (improvement ratio of 3.25% and 29.95% for contextaware and sequential recommendation, respectively). Also, qualitative analysis of the attention weights and topic modeling provides insight into the model's recommendation process and results.

Personalized Route Recommendation With Neural Network Enhanced Search Algorithm

“…More recently, Recurrent Neural Network (RNN) together with its variant Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) have been widely used for modeling sequential trajectory data. Zhou et al [31] proposed topic-enhanced memory networks for POI recommendation problem; Zhen et al [32] utilized hierarchical RNN to capture high-level information in trajectories; Gao et al [33] proposed variational RNN to model the latent variable of trajectories data; Wu et al [34] took road network constraints into consideration when designing recurrent neural network; Feng et al [35] proposed a multi-modal embedding RNN with attention mechanism to predict human mobility; Liu et al [36] proposed spatial-temporal RNN to model spatial-temporal context information; Ai et al [11] proposed a Space time feature-based RNN to model spatial-temporal information. These studies mainly focus on short-term trajectory behaviors, e.g., one-step location recommendation [36], which are not suitable for solving the current task.…”

Curriculum Meta-Learning for Next POI Recommendation

“…The techniques for next POI recommendation include Markov Chain based [4,48] and Neural Network based. The latter, including the models derived from RNNs [14,20], LSTMs [18,49], Memory Network [52], and GANs [50], has gained wide attention due to the superior performance. However, these models require large-scale training data which are not available in coldstart cities in our scenario.…”

Learning Graph-Based Geographical Latent Representation for Point-of-Interest Recommendation

“…. In many research studies on POI recommendation, auxiliary information such as social relationships [29] and temporal information [10,28,33] are used for improving POI recommendation performance. However, in this paper, we focus on only geographical influences among POIs for POI recommendation.…”

“…We exclude several state-of-the-art POI recommendation models such as TEMN [33] and STGN [30], which utilize POI sequences as inputs for recommending the next POI, since they are designed for successive POI recommendation.…”