Multi-data Fusion Based Marketing Prediction of Listed Enterprise Using MS-LSTM Model

The intelligent analysis and marketing prediction of high-tech enterprises based on artificial intelligence is a hot topic in the field. Most of the existing researches mainly focus on taking the internal structural features of the enterprise as the starting point to study the influencing factors of enterprise marketing trends. Different with previous studies, This research attempts to simulate the analyzing and decision processes of domain experts towards issues of enterprise operations by using artificial intelligent. One main challenge of the research is to simulate domain experts’ behaviors of analyzing multi-data including both structured and unstructured data, especially how to extract knowledge, patterns and import factors from unstructured data to support enterprise decisions. In order to solve the challenge mentioned above, an intelligent analysis framework MS-LSTM based on business management theory is proposed. Firstly, MS-LSTM collects, processes and analyzes multi-data by using an encoder strategy module, which contains more than 10 strategy models such as normalization, one-hot, distribution fitting, time series completion, semantic encoding, Bert, etc., providing high quality input for downstream tasks. Finally, a LSTM based time series processing model is proposed to make marketing prediction based on upstream processed multi-source data. Extensive experiments are conducted to verify the proposed model. Compared with traditional benchmark model, the proposed MS-LSTM could efficiently extract meaningful knowledge and patterns, which could be explained to a certain extent by business management theory, from multi-source data. The model has improved the accuracy of enterprise trend prediction by 19.3 times compared with state-of-art baselines, which further verify the application values of the research.