Self-Attentive Neural Network for Hashtag Recommendation

Abstract

A hashtag is a type of metadata tag used on social networks and can help people search for specific topics or content. To capture the interactive information between words and understand the content of microblog posts deeply, this study proposed a neural network model based on a word-level self-attention mechanism. Given a microblog post, the weight of each word was calculated through a self-attention mechanism, and then the representation of a microblog post was obtained through the weighted summation of words. Finally, a multi-layer perceptron was adopted on the representation of a microblog post to perform the classification. The effectiveness of the proposed model was verified through experiments of large-scale datasets. Results show that: (1) introducing word-level self-attention mechanism into hashtag recommendation is effective. (2) In comparison with the baseline methods used in previous studies, such as convolutional neural network or long short-term memory network, the proposed self-attentive neural networks can provide a more accurate representation of a microblog post and significantly improve the F-score of hashtag recommendation on the same dataset. This study provides references for the methods and evaluation of short-text hashtag recommendations, such as microblogs.

Text-to-hashtag Generation using Seq2Seq Learning

“…On the Internet, there is an abundance of content regarding the implementation of neural networks for text translation and summarization similar to this work [Kaviani and Rahmani 2020;Li et al 2016Li et al , 2019Yang et al 2019].…”

PAC-MAN: Multi-Relation Network in Social Community for Personalized Hashtag Recommendation

“…Due to their powerful ability in the natural language processing domain, various approaches have been proposed for non-personalized hashtag recommendation. For example, SANN [6] leverages a multihead attention mechanism for finding microblog representation.…”

“…Transformer and its variants, such as BERT [22], are applied and show improvement in some hashtag recommendations. SANN [6] leverages the self-attention mechanism to obtain microblog representation for making recommendations. EmHash [7] applies BERT embedding to obtain microblog representation for hashtag recommendation.…”

“…Users have dynamic relations with their interacted hashtags based on relevance levels between them. To obtain the type-r interaction-based user embedding q a i,r with regard to the dynamic relations, we then aggregate all type-r interaction messages m a i←j ,r by applying the multi-head attentive aggregation MHA(•) with the user embedding u a i as query Q, and the type-r interaction message m a i←j ,r as key K and value V, as shown in Equation (6).…”

Fenomena Tagar Foodporn di Instagram Berbahasa Indonesia: Sebuah Tinjauan Gramatika Visual

“…Penggunaan gambar yang menarik ditambah tagar (tanda pagar) yang unik menambah ketertarikan peselancar dunia maya dalam berselancar di aplikasi Instagram ini. Tagar dalam bahasa Indonesia dikenal luas dengan sebutan hastag merupakan jenis tanda metadata yang digunakan di jejaring sosial dan dapat membantu orang mencari topik atau konten tertentu (Yang et al, 2019). Tagar digunakan untuk mengatur informasi secara akurat dan efektif sehingga banyak layanan mikroblog yang memungkinkan pengguna membuat dan menggunakan tanda pagar depan kata atau frasa tanpa spasi dalam sebuah unggahan.…”