Augmenting Sequential Recommendation with Pseudo-Prior Items via Reversely Pre-training Transformer

Abstract

Sequential Recommendation characterizes the evolving patterns by modeling item sequences chronologically. The essential target of it is to capture the item transition correlations. The recent developments of transformer inspire the community to design effective sequence encoders, e.g., SASRec and BERT4Rec. However, we observe that these transformer-based models suffer from the cold-start issue, i.e., performing poorly for short sequences. Therefore, we propose to augment short sequences while still preserving original sequential correlations. We introduce a new framework for Augmenting Sequential Recommendation with Pseudo-prior items (ASReP). We firstly pre-train a transformer with sequences in a reverse direction to predict prior items. Then, we use this transformer to generate fabricated historical items at the beginning of short sequences. Finally, we fine-tune the transformer using these augmented sequences from the time order to predict the next item. Experiments on two real-world datasets verify the effectiveness of ASReP. The code is available on https://github.com/DyGRec/ASReP. CCS CONCEPTS• Information systems → Recommender systems; • Computing methodologies → Neural networks.

Sequential Recommendation via Stochastic Self-Attention

“…SR encodes users' dynamic interests by modeling item-item transition relationships in sequences. Recent advancements in Transformer [24,42] introduce the self-attention mechanism to reveal the position-wise item-item relationships, which leads to the stateof-the-art performance in SR. SASRec is the pioneering work in proposing Transformer for sequential recommendation, which applies scaled dot-product self-attention to learn item-item correlation weights. BERT4Rec [38] adopts bi-directional modeling in sequences.…”

Improving Contrastive Learning with Model Augmentation

“…The sequential recommendation (Fan et al, 2021;Liu et al, 2021c;Chen et al, 2018;Tang & Wang, 2018;Zheng et al, 2019) aims at predicting future items in sequences, where the crucial part is to characterize item relationships in sequences. Recent developments in sequence modeling (Fan et al, 2021;Liu et al, 2021c) verify the superiority of Transform (Vaswani et al, 2017), i.e. the selfattention mechanism, in revealing item correlations in sequences.…”

“…However, the data sparsity issue (Liu et al, 2021c) and noise in sequences undermine the performance of a model in sequential recommendation. The former hinders performance due to insufficient training since the complex structure of a sequential model requires a dense corpus to be adequately trained.…”

Pseudo Siamese Network for Few-shot Intent Generation

“…If we can generate high-quality pseudo-labeled examples for these few-shot intents, we can effectively alleviate this issue and improve the performance. There are only a few previous works [11,12,16,26] that try to augment the training data with generation methods and alleviate the scarce annotation problem. However, these models utilize simple neural networks with limited model capacity, like LSTMs [7], to do text generation.…”