Consistency-Aware Recommendation for User-Generated Item List Continuation

Abstract

User-generated item lists are popular on many platforms. Examples include video-based playlists on YouTube, image-based lists (or "boards") on Pinterest, book-based lists on Goodreads, and answerbased lists on question-answer forums like Zhihu. As users create these lists, a common challenge is in identifying what items to curate next. Some lists are organized around particular genres or topics, while others are seemingly incoherent, reflecting individual preferences for what items belong together. Furthermore, this heterogeneity in item consistency may vary from platform to platform, and from sub-community to sub-community. Hence, this paper proposes a generalizable approach for user-generated item list continuation. Complementary to methods that exploit specific content patterns (e.g., as in song-based playlists that rely on audio features), the proposed approach models the consistency of item lists based on human curation patterns, and so can be deployed across a wide range of varying item types (e.g., videos, images, books). A key contribution is in intelligently combining two preference models via a novel consistency-aware gating network -a general user preference model that captures a user's overall interests, and a current preference priority model that captures a user's current (as of the most recent item) interests. In this way, the proposed consistency-aware recommender can dynamically adapt as user preferences evolve. Evaluation over four datasets (of songs, books, and answers) confirms these observations and demonstrates the effectiveness of the proposed model versus stateof-the-art alternatives. Further, all code and data are available at https://github.com/heyunh2015/ListContinuation_WSDM2020.

AdaMCT: Adaptive Mixture of CNN-Transformer for Sequential Recommendation

“…Furthermore, we devise a personalized scheme to adaptively aggregate the output of these modules. This is in contrast to many existing works that combine local and global modules by simple summation and concatenation [3,17,19,30,55,56,58] either without personalized adaptation or without learnable importance adjustment.…”

“…Moreover, the coefficients differ for different users, which shows that it self-adaptively learns different patterns for personalized users. This further illustrates that it is not an ideal way to set a static and equal trade-off between local and global branches, as many works [3,17,19,30,55,56,58] did in the past. In our adaptive paradigm, the model can fuse information with the dynamical use of locality bias (CNNs) to improve the performance of the Transformer.…”

“…Many prior works [21,36,44] propose and design local-global memory mechanisms to learn different-level item sequences, while they use a simple concatenation of neural modules without an adaptive learnable preference factor. However, some works [17,29] [31,62]. These works design user-specific strategies to fuse local and global information into the recommendation model [6,47].…”

Decoupled Side Information Fusion for Sequential Recommendation

“…Specifically, self-attention based methods, such as SAS-Rec [11] and BERT4Rec [25] have been regarded with strong potential due to their ability to capture long-range dependencies between items. Recently, many improvements on self-attention based solutions are proposed, taking into consideration personalization [28], item similarity [15], consistency [7], multiple interests [5], information dissemination [29], pseudo-prior items augmentation [18], motifs [3], etc. However, most of the current SR methods often assume that only the item IDs are available and do not take the side information like item attributes into consideration, ignoring the fact that such highly-related information could provide extra supervision signals.…”

Learning Context-Aware Outfit Recommendation

“…Sequential recommendation methods usually discover the internal connections between previously purchased items to accurately predict the next item. Generally, sequence models are divided into two main types: (1) the order-based models view user sequences as item orders, mainly mining the relationship between commodity sequences, such as Markov Chains [22], recurrent neural network RNN [23,24], convolutional neural network [CNN] [25,26] and attention mechanism [27]; GRU4Rec used a gated recurrent unit (GRU) to build session-based recommendations [28] Modular click sequence, and the improved version further improves its Top-N recommendation performance [29]. However, in each time step, the RNN takes the state and current operation of the last step as its input.…”