Sparsity-aware neural user behavior modeling in online interaction platforms

Modern online platforms offer users an opportunity to participate in a variety of content-creation, social networking, and shopping activities. With the rapid proliferation of such online services, learning data-driven user behavior models is indispensable to enable personalized user experiences. Recently, representation learning has emerged as an effective strategy for user modeling, powered by neural networks trained over large volumes of interaction data. Despite their enormous potential, we encounter the unique challenge of data sparsity for a vast majority of entities, e.g., sparsity in ground-truth labels for entities and in entity-level interactions (cold-start users, items in the long-tail, and ephemeral groups). In this dissertation, we develop generalizable neural representation learning frameworks for user behavior modeling designed to address different sparsity challenges across applications. Our problem settings span transductive and inductive learning scenarios, where transductive learning models entities seen during training and inductive learning targets entities that are only observed during inference. We leverage different facets of information reflecting user behavior (e.g., interconnectivity in social networks, temporal and attributed interaction information) to enable personalized inference at scale. Our proposed models are complementary to concurrent advances in neural architectural choices and are adaptive to the rapid addition of new applications in online platforms. First, we examine two transductive learning settings: inference and recommendation in graph-structured and bipartite user-item interactions. In chapter 3, we formulate user profiling in social platforms as semi-supervised learning over graphs given sparse ground-truth labels for node attributes. We present a graph neural network framework that exploits higher-order connectivity structures (network motifs) to learn attributed structural roles of nodes that identify structurally similar nodes with co-varying local attributes. In chapter 4, we design neural collaborative filtering models for few-shot recommendations over user-item interactions. To address item interaction sparsity due to heavy-tailed distributions, our proposed meta-learning framework learns-to-recommend few-shot items by knowledge transfer from arbitrary base recommenders. We show that our framework consistently outperforms state-of-art approaches on overall recommendation (by 5% Recall) while achieving significant gains (of 60-80% Recall) for tail items with fewer than 20 interactions. Next, we explored three inductive learning settings: modeling spread of user-generated content in social networks; item recommendations for ephemeral groups; and friend ranking in large-scale social platforms. In chapter 5, we focus on diffusion prediction in social networks where a vast population of users rarely post content. We introduce a deep generative modeling framework that models users as probability distributions in the latent space with variational priors parameterized by graph neural networks. Our approach enables massive performance gains (over 150% recall) for users with sparse activities while being faster than state-of-the-art neural models by an order of magnitude. In chapter 6, we examine item recommendations for ephemeral groups with limited or no historical interactions together. To overcome group interaction sparsity, we present self-supervised learning strategies that exploit the preference co-variance in observed group memberships for group recommender training. Our framework achieves significant performance gains (over 30% NDCG) over prior state-of-the-art group recommendation models. In chapter 7, we introduce multi-modal inference with graph neural networks that captures knowledge from multiple feature modalities and user interactions for multi-faceted friend ranking. Our approach achieves notable higher performance gains for critical populations of less-active and low degree users.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Sankar, Aravind

Contributors dc:contributor

• Sundaram, Hari
• Han, Jiawei
• Tong, Hanghang
• Shah, Neil

• Computer science

Statement dc:rights

• Copyright 2021 Aravind Sankar

Language dc:language

en, eng

Handle dc:identifier

http://hdl.handle.net/2142/113910

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F113910