[Paper Review] A Correspondence Analysis Framework for Author-Conference Recommendations

Rahul Radhakrishnan Iyer, Manish Sharma+1 more|arXiv (Cornell University)|Jan 8, 2020

Recommender Systems and Techniques51 references4 citations

TL;DR

This paper proposes a novel recommender system for author-conference matching using Correspondence Analysis (CA) to integrate paper content and author social networks, outperforming traditional methods like content-based and collaborative filtering. By modeling paper-conference relationships through frequency-based matrices and reducing dimensionality via CA, the approach achieves superior recommendation accuracy, especially for new authors with no publication history.

ABSTRACT

For many years, achievements and discoveries made by scientists are made aware through research papers published in appropriate journals or conferences. Often, established scientists and especially newbies are caught up in the dilemma of choosing an appropriate conference to get their work through. Every scientific conference and journal is inclined towards a particular field of research and there is a vast multitude of them for any particular field. Choosing an appropriate venue is vital as it helps in reaching out to the right audience and also to further one's chance of getting their paper published. In this work, we address the problem of recommending appropriate conferences to the authors to increase their chances of acceptance. We present three different approaches for the same involving the use of social network of the authors and the content of the paper in the settings of dimensionality reduction and topic modeling. In all these approaches, we apply Correspondence Analysis (CA) to derive appropriate relationships between the entities in question, such as conferences and papers. Our models show promising results when compared with existing methods such as content-based filtering, collaborative filtering and hybrid filtering.

Motivation & Objective

• To address the challenge of recommending appropriate conferences to researchers, especially new authors lacking publication history.
• To overcome limitations of existing methods that rely solely on content or social networks by integrating both into a unified framework.
• To develop a dimensionality reduction-based approach using Correspondence Analysis (CA) that effectively models relationships between papers, authors, and conferences.
• To empirically evaluate the proposed method against established filtering techniques (content-based, collaborative, hybrid) on real-world ACM conference data.

Proposed method

• Constructs frequency-based matrices: Paper × Words, Paper × Conference, and Words × Conference from paper abstracts and conference publication records.
• Applies Correspondence Analysis (CA) to reduce dimensionality and extract latent relationships between entities, avoiding the limitations of PCA on discrete data.
• Uses tf-idf and topic modeling (LDA) to represent paper content, with CA applied to the resulting matrices for dimensionality reduction.
• Combines content and social network features in a hybrid model, with similarity computed using cosine, Euclidean, and Pearson metrics.
• Employs a multi-stage recommendation pipeline: matrix construction → CA transformation → similarity computation → top-k conference ranking.
• Validates performance using standard metrics: MAP@5, MRR, MNDCG@5, MR-P, MF-M@5, MP@5, and MR@5 on a real dataset of ACM publications.

Experimental results

Research questions

• RQ1 Can Correspondence Analysis effectively model paper-conference relationships when applied to discrete, frequency-based data matrices?
• RQ2 How does integrating author social network data improve conference recommendation accuracy, especially for new authors?
• RQ3 Does content-based filtering using CA outperform traditional collaborative filtering and hybrid methods in author-conference recommendation?
• RQ4 To what extent does the choice of representation (tf-idf vs. topic modeling) affect recommendation performance?
• RQ5 Which similarity measure (cosine, Euclidean, Pearson) yields the most accurate conference recommendations in the proposed framework?

Key findings

• The content-based CA method significantly outperforms collaborative filtering, with MAP@5 of 0.1531 (cosine) compared to 0.1083 for collaborative filtering using tf-idf.
• The first method, which relies only on conference frequency and not content, performs poorly for new authors, indicating the need for content-aware models.
• Content-based filtering using CA achieves higher accuracy than CA-based methods alone, suggesting some information loss during dimensionality reduction.
• tf-idf representation outperforms topic modeling (400 topics), likely due to higher feature granularity and better preservation of lexical information.
• Cosine similarity consistently yields the best performance across all metrics, with MRR of 0.3862 in collaborative filtering and 0.3237 in hybrid filtering.
• The hybrid filtering method combining content and collaborative signals does not significantly improve results over content-based filtering alone, indicating that content remains the dominant factor.