DeepCAT: Deep Category Representation for Query Understanding in E-commerce Search

Mapping a search query to a set of relevant categories in the product is a significant challenge in e-commerce search for two reasons: 1) data exhibits severe class imbalance problem due to biased click, and 2) queries with little customer feedback (e.g., tail queries) are well-represented in the training set, and cause difficulties for query. To address these problems, we propose a deep learning model,, which learns joint word-category representations to enhance the query process. We believe learning category interactions helps to the performance of category mapping on minority classes, tail and torso. DeepCAT contains a novel word-category representation model that the category representations based on word-category co-occurrences in training set. The category representation is then leveraged to introduce a loss function to estimate the category-category co-occurrences for refining word-category embeddings. To demonstrate our model's effectiveness on categories and tail queries, we conduct two sets of experiments. The show that DeepCAT reaches a 10% improvement on minority classes and a7.1% improvement on tail queries over a state-of-the-art label embedding model. findings suggest a promising direction for improving e-commerce search by modeling of taxonomy hierarchies.