Fine-tune BERT for E-commerce Non-Default Search Ranking
This work addresses relevance issues in e-commerce search for platforms, but it is incremental as it applies existing methods like BERT to a specific domain.
The paper tackled the problem of poor relevance in non-default e-commerce search rankings by proposing a two-stage ranking scheme that combines refined keyword matching with BERT-Large fine-tuning, achieving first place in a supervised phase and second place in a final phase based on F1 scores.
The quality of non-default ranking on e-commerce platforms, such as based on ascending item price or descending historical sales volume, often suffers from acute relevance problems, since the irrelevant items are much easier to be exposed at the top of the ranking results. In this work, we propose a two-stage ranking scheme, which first recalls wide range of candidate items through refined query/title keyword matching, and then classifies the recalled items using BERT-Large fine-tuned on human label data. We also implemented parallel prediction on multiple GPU hosts and a C++ tokenization custom op of Tensorflow. In this data challenge, our model won the 1st place in the supervised phase (based on overall F1 score) and 2nd place in the final phase (based on average per query F1 score).