Personalizing Session-based Recommendations with Hierarchical Recurrent Neural Networks

By Massimo Quadrana, Alexandros Karatzoglou, Balázs Hidasi, Paolo Cremonesi. ACM Conference on Recommender Systems, 2017

711 citations recorded by the source corpus.

Session-based recommendations are highly relevant in many modern on-line services (e.g. e-commerce, video streaming) and recommendation settings. Recently, Recurrent Neural Networks have been shown to perform very well in session-based settings. While in many session-based recommendation domains user identifiers are hard to come by, there are also domains in which user profiles are readily available. We propose a seamless way to personalize RNN models with cross-session information transfer and devise a Hierarchical RNN model that relays end evolves latent hidden states of the RNNs across user sessions. Results on two industry datasets show large improvements over the session-only RNNs.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Information Systems

View Paper: https://www.semanticscholar.org/paper/cb57001315188582bbb79e548dfdb354592fa80e

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Information Systems

Citations in the corpus

A Simple Convolutional Generative Network for Next Item Recommendation (2018, methodology): “…the comparison with content-or context-based sequential recommendation models, such as the 3D CNN recommender [30] and other RNN variants [9, 20, 25, 27]; (2) the GRURec baseline could be regarded as the state-of-the-art Improved GRURec [28] when dealing with the long-range session data sets…”
Contrastive Learning for Sequential Recommendation (2022, methodology): “The following variants modified GRU4Rec by introducing attention mechanism [4, 29, 30], hierarchical structure [36], and user-based gated network [11].”
S3-Rec: Self-Supervised Learning for Sequential Recommendation with Mutual Information Maximization (2020, methodology): “…[3] firstly introduced Gated Recurrent Units (GRU) to the session-based recommendation and a surge of following variants modified this model by introducing pair-wise loss functions [4], memory networks [6, 7], hierarchical structures [17], copy mechanism [18] and reinforcement learning [27], etc.”
Improving Sequential Recommendation with Knowledge-Enhanced Memory Networks (2018, methodology): “As one of the most popular neural networks, Recurrent Neural Networks (RNN) together with its variants LSTM and GRU have been widely applied to sequential recommendation, including session-based GRU [15], user-based GRU [5], dynamic recurrent model [36], recurrent recommender network [35], hierarchi”
Deep Session Interest Network for Click-Through Rate Prediction (2019, background): “Hierarchical RNN [ Quadrana et al. , 2017 ] is proposed to relays end evolves latent hidden states of the RNNs across users’ historical sessions.”
Sequential Recommender Systems: Challenges, Progress and Prospects (2019, methodology): “On the other hand, some hierarchical models including hierarchical embedding models [18], hierarchical RNN [13] and hierarchical attention networks [25] have been devised to incorporate the historical sub-sequences into sequential dependency learning to build more powerful SRSs.”
A Survey on Session-based Recommender Systems (2019, methodology): “To improve GRU4Rec, Tan et al. [107] adopted data augmentation via sequence preprocessing and embedding dropout to enhance the training process and reduce overfitting respectively.”
BERT4Rec: Sequential Recommendation with Bidirectional Encoder Representations from Transformer (2019, background): “Recently, RNN and its variants, Gated Recurrent Unit (GRU) [4] and Long Short-Term Memory (LSTM) [17], are becoming more and more popular for modeling user behavior sequences [7, 14, 15, 28, 36, 56, 59].”
STAMP: Short-Term Attention/Memory Priority Model for Session-based Recommendation (2018, methodology): “In order to verify the performance of our proposed STAMP model and the recent state-of-the-art NARM model in real production environment, where recommendation systems can only suggest a few items at once, the relevant item should be amongst the first few items in the recommendation list[12].”
Recommendation system based on deep learning methods: a systematic review and new directions (2019, methodology): “Most of the RNN based RS method particularly exploit the RNN model for a session-based recommendation (Tan et al. 2016; Wu et al. 2016b; Quadrana et al. 2017; Hidasi et al. 2016a).”
Explainable Interaction-driven User Modeling over Knowledge Graph for Sequential Recommendation (2019, methodology): “Sequence modeling methods for SR mainly belong to Markov Models [4, 17] and RNNs[8, 9, 15, 27].”
PromptRobust: Towards Evaluating the Robustness of Large Language Models on Adversarial Prompts (2023, neutral): “By maintaining a limited history of recent queries (Malhotra et al., 2015; Quadrana et al., 2017), the system can identify patterns or sequences of inputs that might be part of a more extensive adversarial strategy.”
Deep Learning Based Recommender System (2017, background): “Sequential Information w/t User ID [16, 29, 33, 35, 73, 91, 118, 134, 144, 161, 174, 176, 190, 195, 199, 206]”
Hierarchical Temporal Convolutional Networks for Dynamic Recommender Systems (2019, methodology): “These two approaches are first explored in [23], and we refer to them as Init connection and Full connection, respectively.”
Practice on Long Sequential User Behavior Modeling for Click-Through Rate Prediction (2019, background): “pooling operations to summarize user interest representation, (ii) sequential-modeling architecture [21, 30] which treats user behaviors as sequential signals and applies LSTM/GRU operations for user interest summarization.”
Hierarchical Gating Networks for Sequential Recommendation (2019, methodology): “RNN-based methods [12, 13, 28] adopt gated recurrent (GRU) or long short-term memory (LSTM) units to learn the user-item sequence, where the short-term user interests are captured by the hidden states of RNNs.”
AMER: Automatic Behavior Modeling and Interaction Exploration in Recommender System (2020, methodology): “• RNN-based methods [27, 26, 53, 73] mainly apply LSTM or GRU layers to extract sequential representations, which can be covered by recurrent blocks in AMER.”
Sequence-Aware Recommender Systems (2018, methodology): “Better integrated models as used, for example, in [91] are therefore still needed.”