Multi-Relational Graph based Heterogeneous Multi-Task Learning in Community Question Answering

TL;DR

This paper introduces HMTGIN, a novel multi-relational graph-based multi-task learning model designed to jointly solve heterogeneous community question answering tasks, demonstrating superior performance on a large-scale dataset.

Contribution

HMTGIN is the first MTL model to effectively handle CQA tasks using multi-relational graphs, integrating domain knowledge and graph isomorphism networks for improved learning.

Findings

HMTGIN outperforms all baseline models on five CQA tasks.

The multi-relational graph approach enhances task performance.

Cross-task constraints significantly improve joint learning effectiveness.

Abstract

Various data mining tasks have been proposed to study Community Question Answering (CQA) platforms like Stack Overflow. The relatedness between some of these tasks provides useful learning signals to each other via Multi-Task Learning (MTL). However, due to the high heterogeneity of these tasks, few existing works manage to jointly solve them in a unified framework. To tackle this challenge, we develop a multi-relational graph based MTL model called Heterogeneous Multi-Task Graph Isomorphism Network (HMTGIN) which efficiently solves heterogeneous CQA tasks. In each training forward pass, HMTGIN embeds the input CQA forum graph by an extension of Graph Isomorphism Network and skip connections. The embeddings are then shared across all task-specific output layers to compute respective losses. Moreover, two cross-task constraints based on the domain knowledge about tasks' relationships are used to regularize the joint learning. In the evaluation, the embeddings are shared among different task-specific output layers to make corresponding predictions. To the best of our knowledge, HMTGIN is the first MTL model capable of tackling CQA tasks from the aspect of multi-relational graphs. To evaluate HMTGIN's effectiveness, we build a novel large-scale multi-relational graph CQA dataset with over two million nodes from Stack Overflow. Extensive experiments show that: $(1)$ HMTGIN is superior to all baselines on five tasks; $(2)$ The proposed MTL strategy and cross-task constraints have substantial advantages.