A New Paradigm for E-commerce Retrieval

View the PDF file for the paper entitled Recover and Coordination of a New Form: A New Model for e -commerce retrieval, by Ming Pang and 10 other authors

PDF HTML (experimental) view

a summary:Traditional and dense recovery methods are combined to take advantage of general knowledge in the world and often fail to capture the exact features of information and products. With the appearance of large LLMS models, industrial research systems have begun to employ LLMS to create identifiers to retrieve the product. Common identifiers include (1) fixed/semantic and (2) product sets. The first approach requires the creation of the product identifier system from the zero point, and the loss of global knowledge included within the LLMS. Although the second approach benefits from this general knowledge, the great difference in the distribution of words between information and products means that the product based on the product is not well in line with the user search queries, which leads to the summons of lost products. Moreover, when the information contains many features, these algorithms generate a large number of identifiers, which makes it difficult to evaluate their quality, which leads to a decrease in the efficiency of the summons in general.

To face these challenges, this paper offers a new model for e -commerce retrieval: Gram. Gram employs a joint training on text information from each of the queries and products to create common text identifiers, and effectively fill the gap between queries and products. This approach is not only reinforced by the relationship between information and products, but also improves the efficiency of reasoning. The model uses a joint determination strategy to create improved symbols to increase recovery efficiency. In addition, it provides a product recording mechanism for inquiries to compare the values of the product through various symbols, which increases the efficiency of retrieval. The A/B test on the Internet explains that a gram greatly outperforms traditional models and the latest gym, which confirms its effectiveness and practical diversity.