Fusion of text and graph information for machine learning problems on networks

Abstract

Today, increased attention is drawn towards network representation learning, a technique that maps nodes of a network into vectors of a low-dimensional embedding space. A network embedding constructed this way aims to preserve nodes similarity and other specific network properties. Embedding vectors can later be used for downstream machine learning problems, such as node classification, link prediction and network visualization. Naturally, some networks have text information associated with them. For instance, in a citation network, each node is a scientific paper associated with its abstract or title; in a social network, all users may be viewed as nodes of a network and posts of each user as textual attributes. In this work, we explore how combining existing methods of text and network embeddings can increase accuracy for downstream tasks and propose modifications to popular architectures to better capture textual information in network embedding and fusion frameworks.

Temporal network embedding framework with causal anonymous walks representations

“…Recent studies show that a combination of deep learning models with semi-supervised techniques gives state-of-the-art results in terms of scalability, speed, and quality in downstream tasks ( Makarov et al, 2021 ; Makarov, Makarov & Kiselev, 2021 ; Makarov, Korovina & Kiselev, 2021 ). However, static models are limited by the necessity to retrain the model with each significant change of graph structure.…”

Exploration in Sequential Recommender Systems via Graph Representations

“…There are a lot of applications of GNNs for specific domains in recommender systems like knowledge-aware [28], [29] or social recommendations [30]- [32]. However, our work is concentrated on a classic sequential formulation of recommender systems.…”

Hotel Recognition via Latent Image Embedding

“…Ko et al [10] take a different approach for generating hard points: they generate points by linearly interpolating between negative data points during training and picking the hardest point. Other approaches may directly learn metric space transformation via graph embeddings [14,15].…”