View in EDS

Text-based inductive twitter user geolocation via tweet-level graph construction.

Saved in:
Bibliographic Details
Title: Text-based inductive twitter user geolocation via tweet-level graph construction.
Authors: Qiao, Yaqiong1 (AUTHOR) qiaoyq@nankai.edu.cn, Wei, Qiongya2 (AUTHOR) z20231090844@stu.ncwu.edu.cn, Luo, Xiangyang3 (AUTHOR) xiangyangluo@126.com, Li, Chenliang4 (AUTHOR) lichenliang.whu@gmail.com, Ma, Jiangtao5 (AUTHOR) majiangt@139.com, Li, Xiang1 (AUTHOR) lixiang@nankai.edu.cn
Source: Knowledge & Information Systems. 6/11/2026, Vol. 68 Issue 1, p1-29. 29p.
Subjects: Location data, Message passing (Computer science), Machine learning, Data mining
Abstract: The geographical location of social media users is crucial for various applications such as online marketing and event detection. Existing text-based geolocation methods primarily locate users by mining location-indicative words or statistical features within tweets. However, most of them fail to capture the contextual relationships among words within each document and lack inductive learning capabilities for new words, limiting their geolocation performance. To tackle the problems, this paper proposes a novel text-based inductive Twitter user geolocation method via tweet-level graph construction, called ITGC. ITGC constructs a tweet-level graph for each user's tweets, where the nodes in the graph consist of geographical entities, dialect entities, and mentioned users obtained from tweets via NER, and location labels derived from k-means clustering. Connections between nodes are established based on the values representing the adjacent relationships of words in the globally shared matrices. By combining the globally shared matrices with the tweet-level graph, the model can capture key information and contextual relationships in tweets at a finer granularity while effectively integrating globally and local information. Furthermore, we propose an improved message-passing mechanism (IMPM), enabling ITGC to generate fine-grained node representations based on local structures and dynamically update the node representations within the tweet graph. In this way, ITGC can effectively generate embeddings for unseen words from new tweets, thereby generalizing learned knowledge to previously unseen texts, achieving inductive learning of new words. Experiments on two public datasets show that our method outperforms state-of-the-art baselines. [ABSTRACT FROM AUTHOR]
Copyright of Knowledge & Information Systems is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:The geographical location of social media users is crucial for various applications such as online marketing and event detection. Existing text-based geolocation methods primarily locate users by mining location-indicative words or statistical features within tweets. However, most of them fail to capture the contextual relationships among words within each document and lack inductive learning capabilities for new words, limiting their geolocation performance. To tackle the problems, this paper proposes a novel text-based inductive Twitter user geolocation method via tweet-level graph construction, called ITGC. ITGC constructs a tweet-level graph for each user's tweets, where the nodes in the graph consist of geographical entities, dialect entities, and mentioned users obtained from tweets via NER, and location labels derived from k-means clustering. Connections between nodes are established based on the values representing the adjacent relationships of words in the globally shared matrices. By combining the globally shared matrices with the tweet-level graph, the model can capture key information and contextual relationships in tweets at a finer granularity while effectively integrating globally and local information. Furthermore, we propose an improved message-passing mechanism (IMPM), enabling ITGC to generate fine-grained node representations based on local structures and dynamically update the node representations within the tweet graph. In this way, ITGC can effectively generate embeddings for unseen words from new tweets, thereby generalizing learned knowledge to previously unseen texts, achieving inductive learning of new words. Experiments on two public datasets show that our method outperforms state-of-the-art baselines. [ABSTRACT FROM AUTHOR]
ISSN:02191377
DOI:10.1007/s10115-026-02813-2