FiBiNET: Combining Feature Importance and Bilinear feature Interaction for Click-Through Rate Prediction (1905.09433v1)

Abstract: Advertising and feed ranking are essential to many Internet companies such as Facebook and Sina Weibo. Among many real-world advertising and feed ranking systems, click through rate (CTR) prediction plays a central role. There are many proposed models in this field such as logistic regression, tree based models, factorization machine based models and deep learning based CTR models. However, many current works calculate the feature interactions in a simple way such as Hadamard product and inner product and they care less about the importance of features. In this paper, a new model named FiBiNET as an abbreviation for Feature Importance and Bilinear feature Interaction NETwork is proposed to dynamically learn the feature importance and fine-grained feature interactions. On the one hand, the FiBiNET can dynamically learn the importance of features via the Squeeze-Excitation network (SENET) mechanism; on the other hand, it is able to effectively learn the feature interactions via bilinear function. We conduct extensive experiments on two real-world datasets and show that our shallow model outperforms other shallow models such as factorization machine(FM) and field-aware factorization machine(FFM). In order to improve performance further, we combine a classical deep neural network(DNN) component with the shallow model to be a deep model. The deep FiBiNET consistently outperforms the other state-of-the-art deep models such as DeepFM and extreme deep factorization machine(XdeepFM).

• The paper presents a detailed guide to using the acmart document class to facilitate consistent and compliant ACM publications.
• It outlines technical aspects including margin settings, typography choices, and metadata extraction to enhance digital library management.
• The paper explains how different template styles and parameters address various submission contexts, reducing the challenges of academic formatting.

Overview of "The Name of the Title is Hope"

"The Name of the Title is Hope" is a detailed exposition centered on the utilization of the "acmart" document class, specifically designed to streamline the preparation of articles for publication within the ACM ecosystem. This paper does not engage with novel research findings, but rather provides a comprehensive guide to employing the document class in handling various publication formats. The authors, Ben Trovato, G.K.M. Tobin, Lars Thörväld, and a cohort of contributors, offer insights into the stylistic and functional aspects of preparing documentation for ACM's diverse suite of journals and conferences.

Technical and Functional Aspects

The core contribution of the paper lies in elucidating the "acmart" document class capabilities, which integrates accessibility features and is structured to facilitate metadata extraction for enhanced digital library management. The document class supports intricate formatting tasks, such as setting appropriate margins, typeface requirements (specifically the "Libertine" typeface), and enforcing the strict prohibition of unauthorized modifications. Emphasis is placed on conforming to ACM's rigorous standards in structuring content for review, as well as for final publication states, ranging from double-blind conference submissions to camera-ready journal articles.

Analytical Features and Contributions

Prominently, the paper outlines the distinctions between template styles applicable to different types of publications—such as acmsmall, acmlarge, and acmtog for journals, and acmconf, sigchi, and sigplan for conference proceedings—which are pivotal in aligning an article's presentation with ACM's guidelines. In addition to template styles, a set of parameters is detailed to modify the selected template style, offering options catering to double-blind submissions or author versions suitable for distribution.

The paper also systematically explores the metadata-related elements such as CCS concepts and user-defined keywords, highlighting their significance for enhancing the discoverability of research outputs. These thematic taxonomies, critical for academic searchability, are complemented by directives for properly utilizing tables and figures, ensuring consistency across submissions.

Practical and Theoretical Implications

While the paper is primarily a practical guide, its implications are notably significant for the academic community engaged in ACM publications. By consolidating various formatting protocols into a uniform template, it reduces the cognitive load on researchers tasked with adhering to publication standards, allowing them to focus on content generation. The theoretical implications, though not the primary focus, are embedded in ensuring clarity, accessibility, and systemic coherence of disseminated research, thereby impacting the scientific communication process positively.

Speculative Future Directions

Potential future developments involve further integration of the "acmart" template with evolving digital platforms to accommodate emerging modes of scholarly communication, such as enhanced interactive content or preprint repositories. Additionally, continued improvements in the template may address dynamically changing requirements of accessibility standards and responsive design to cater to a diverse global audience.

In summation, "The Name of the Title is Hope" serves as an essential resource for those involved in preparing and submitting research manuscripts within the ACM framework, emphasizing systematic adherence to best practices in academic publishing. While the paper itself does not purport new knowledge in computer science, its contribution to academic scholarship through rigorous guidelines for document preparation is indisputable.