Mind the Gap: Cross-Lingual Information Retrieval with Hierarchical Knowledge Enhancement

Authors: Fuwei Zhang, Zhao Zhang, Xiang Ao, Dehong Gao, Fuzhen Zhuang, Yi Wei, Qing He4345-4353

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, experimental results demonstrate that HIKE achieves substantial improvements over state-of-the-art competitors.
Researcher Affiliation Collaboration 1 Key Lab of Intelligent Information Processing of Chinese Academy of Sciences (CAS), Institute of Computing Technology, CAS, Beijing 100190, China 2 University of Chinese Academy of Sciences, Beijing 100049, China 3 Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China 4 Institute of Intelligent Computing Technology, Suzhou, CAS 5 Alibaba Group, Hangzhou, China 6 Institute of Artificial Intelligence, Beihang University, Beijing 100191, China 7 SKLSDE, School of Computer Science, Beihang University, Beijing 100191, China
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code.
Open Datasets Yes We evaluate the HIKE model in a public CLIR dataset CLIRMatrix (Sun and Duh 2020). Specifically, we use the MULTI-8 set in CLIRMatrix, in which queries and documents are jointly aligned in 8 different languages.
Dataset Splits Yes The training sets of every language pair contain 10,000 queries, while the validation and the test sets contain 1,000 queries.
Hardware Specification No The paper does not provide specific hardware details used for running its experiments.
Software Dependencies No The paper mentions using 'multilingual BERT' and 'BERT-base, multilingual cased' but does not provide specific version numbers for software dependencies or frameworks.
Experiment Setup Yes In the training stage, the number of heads for the multi-head attention mechanism in knowledge-level fusion is set to 6. The learning rates are divided into two parts: the BERT lr1 and the other modules lr2. And we set lr1 to 1e-5 and lr2 to 1e-3. We set the number of neighboring entities in KG as 3. We randomly sample 1600 query-document pairs as our training data per epoch. The maximum training epochs are set to 15.