Research on Cultural Tourism Product Portfolio Optimization and Profit Attribution Based on Representation Learning and Causal Inference

In order to cope with the revenue uncertainty and strategic decision-making complexity faced by the cultural tourism industry in the digital operation process, this paper proposes a cultural tourism product portfolio optimization method that integrates representation learning and causal inference. This method builds an explainable profit attribution mechanism through the fusion of semantic and structural bimodal features, individualized causal effect estimation and multi-objective revenue optimization. The lightweight Transformer model is used to jointly represent review text and structured attributes, achieving unified encoding of semantics and business attributes; and a strategic effect estimation module based on dual-robust deep causal learning is designed to identify individualized income differences under different delivery strategies. Under budget constraints, revenue maximization and risk balance are achieved through multi-objective optimization. Empirical results on two public datasets show that the model proposed in this paper significantly outperforms existing methods in both profit prediction accuracy and causal effect identification. RMSE is reduced by at least 2.6%, while Qini and the overall optimization score are improved by 5.8% and 4.9%. It achieves a better balance between income stability and portfolio diversity, verifying its application potential in intelligent cultural tourism decision-making scenarios.