Stellenausschreibung220/en
Call for Master Thesis: Benchmark Graph Algorithms attributed by large language model
Job Advertisement
Improving text features is a fundamental aspect of academic recommendation networks. Within academic citation networks, it’s essential to grasp the essence of texts, assess the innovativeness of new documents, and uncover connections between texts. Existing relationship predictions mostly rely on using Bag of Words (BoW)[3] to extract text features, which cannot comprehend and extract concept-level text features based on understanding. However, for text citation networks, comprehension is a necessary algorithmic capability, especially when the citation data itself contains some noisy and missing links. Besides, it also enables the generation of human-readable interpretations for error analysis and optimization of the datasets. However, recent works have indicated that there is a certain trade-off between the effective utilization of features and the effective utilization of structural features [4]. In other words, relationship prediction algorithms based on message passing algorithms can either fully extract structural features or efficiently utilize node features, but not both simultaneously. Therefore, in this project, our aim is to delve deeply into the maximum accuracy achievable by these two strategies: namely, utilizing language models exclusively and relying solely on the structural characteristics of graphs for relationship prediction.
Student Assistants / Tutors
No information available