Ranking-based Fusion Algorithms for Extreme Multi-label Text Classification (XMTC)

TL;DR

This paper proposes ranking-based fusion algorithms that combine sparse and dense retrieval methods to improve extreme multi-label text classification, effectively addressing the long-tail label distribution challenge.

Contribution

It introduces novel fusion algorithms that leverage the complementary strengths of sparse and dense retrievers for better label ranking in XMTC.

Findings

Fusion algorithms improve overall classification accuracy.

Enhanced performance on tail labels compared to individual retrievers.

Effective balancing of head and tail label predictions.

Abstract

In the context of Extreme Multi-label Text Classification (XMTC), where labels are assigned to text instances from a large label space, the long-tail distribution of labels presents a significant challenge. Labels can be broadly categorized into frequent, high-coverage \textbf{head labels} and infrequent, low-coverage \textbf{tail labels}, complicating the task of balancing effectiveness across all labels. To address this, combining predictions from multiple retrieval methods, such as sparse retrievers (e.g., BM25) and dense retrievers (e.g., fine-tuned BERT), offers a promising solution. The fusion of \textit{sparse} and \textit{dense} retrievers is motivated by the complementary ranking characteristics of these methods. Sparse retrievers compute relevance scores based on high-dimensional, bag-of-words representations, while dense retrievers utilize approximate nearest neighbor (ANN) algorithms on dense text and label embeddings within a shared embedding space. Rank-based fusion algorithms leverage these differences by combining the precise matching capabilities of sparse retrievers with the semantic richness of dense retrievers, thereby producing a final ranking that improves the effectiveness across both head and tail labels.