Cross-Lingual Low-Resource Set-to-Description Retrieval for Global E-Commerce
Authors: Juntao Li, Chang Liu, Jian Wang, Lidong Bing, Hongsong Li, Xiaozhong Liu, Dongyan Zhao, Rui Yan8212-8219
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results indicate that our proposed CLMN yields impressive results on the challenging task and the context-dependent cross-lingual mapping on BERT yields noticeable improvement over the pre-trained multi-lingual BERT model. |
| Researcher Affiliation | Collaboration | Peking University, Beijing, China 2Wangxuan Institute of Computer Technology, Peking University, Beijing, China 3DAMO Academy, Alibaba Group 4Indiana University, Bloomington, USA |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks labeled as 'Algorithm' or 'Pseudocode'. |
| Open Source Code | Yes | The code is available on https://github.com/Liu Chang97/CLMN. |
| Open Datasets | No | We manually collect a new and high-quality paired dataset, where each pair contains an unordered product attribute set in the source language and an informative product description in the target language. As the dataset construction process is both time-consuming and costly, the new dataset only comprises of 13.5k pairs, which is a low-resource setting and can be viewed as a challenging testbed for model development and evaluation in cross-border e-commerce. The dataset will be given by asking. |
| Dataset Splits | Yes | We randomly split the alignment dataset into 10,500/1,000/2,000 for training/validating/testing the CLMN model. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper mentions using 'NLTK for tokenization' and 'BERT-based CLMN model' but does not provide specific version numbers for these software components. |
| Experiment Setup | Yes | We limit the length of product attributes and descriptions to 50 words and 100 words, respectively. The pre-trained static monolingual and bilingual word embeddings on Taobao and e Bay corpora used by all the non-BERT models have the dimension size 300. For our BERT-based CLMN model, we use the 768-dimension outputs of the second last BERT layer and keep the dimension of the encoder outputs as well as interaction outputs the same as the dimension of BERT. We use L = 2 stacked attention modules for attributes and descriptions. The parameters of encoders for attributes and descriptions are shared. We set the mini-batch size to 50 and adopt Adam optimizer (Kingma and Ba 2014) with the initial learning rate of 3e-4 to for training. |