Commodity Recommendation Techniques Based on Multimodal Data

With the rapid development of live banding, there is an increasing demand for user recommendations. However, traditional goods recommendations cannot dynamically adjust user preferences. To solve the above problems, we proposes an intelligent recommendation technology based on multimodal data. First, the data are acquired and processed through the e-commerce platform and divided into user side and commodity side; second, encoding and feature extraction is performed for the data, while weights are assigned to the historical comments according to the comment timestamps, which reflect the user's recent interests; third, the characteristics of the commodity image information and textual comment information are fused, and the characteristics of the data information on the user side and commodity side are fused; lastly, commodity recommendation is carried out through a fully-connected neural network. Recommendation. This technology can more accurately understand the user's interest and commodity characteristics, provide users with a more satisfactory recommendation experience, solve the problem of existing technology ignoring the change of user's and commodity's interest and characteristics over time and the problem of single data modality, and provide an effective solution for commodity recommendation.