Research On Predicting Oil Price Trends Based On Sentiment Analysis

As the blood of industry,oil plays an important role in the global economic market,and its price fluctuations have a significant impact on the political and economic activities of countries around the world.Being able to accurately predict fluctuations in crude oil prices can enable policy makers to make forward-looking economic and political decisions,gaining an advantage in a complex international environment.However,the uncertainty and complexity of the oil market(such as complex supply and demand relationships,geopolitical factors,natural disasters,and substitution effects of other energy products)make it difficult to obtain accurate predictions.With the rapid development of the Internet and big data technology,rich online data(including press releases)can help predict the trend of oil prices.Therefore,this study introduces sentiment analysis as a big data analysis tool,mining relevant information,and taking social events such as geopolitical and unexpected events as one of the factors that affect the real-time trend of crude oil prices.The specific tasks for predicting oil prices are as follows:Firstly,in constructing an emotional analysis task based on oil market public opinion,an improved FinBERT based emotional analysis model is proposed.In response to the problem of all hidden state vectors being compressed into one vector in the pooling layer of the FinBERT model,which may result in information loss,this article adds a new pooling layer architecture based on FinBERT.Different pooling schemes express different attributes of the input sequence,the classification token pooling layer provides the overall prediction of the FinBERT model,and the average pooling layer adds the prediction part of each token,The maximum pooling layer tends to input the most expressive token,and finally uses a weighted sum with trainable weights to summarize the results.Secondly,in the oil price prediction task based on the "decomposition integration" framework,a TSD-BiTGRU oil price trend prediction model is proposed.In the decomposition task,a two-stage decomposition feature extraction model(TSD)is proposed to address the problem of incomplete decomposition and insufficient extraction of oil price sequences by a single decomposition model,in order to complete the extraction of internal features of oil prices.The model first uses two stages to decompose the oil price sequence.The first stage uses ICEEMDAN to decompose the oil price sequence,and the second stage uses VMD to decompose the high-frequency sequences in the decomposition results of the first stage.In the integration task,a bi-directional anti output gated loop unit(BiTGRU)model is proposed to improve the prediction accuracy and robustness of the GRU model due to the problem of oversaturation during the training process.Then,this article combines the two improved models and proposes the TSD-BiTGRU oil price prediction model,which is validated on the oil price dataset.Finally,this article constructs a complete Fin PBERT-TSD-ATT-BiTGRU model for predicting oil prices based on public opinion in the oil market.Combining actual data,this article selects 12 influencing indicators from the supply demand relationship and financial market influencing factors,and integrates them into market public opinion indicators.Predict short-term,medium-term,and long-term oil price trends by comparing the predictions of different models at the levels of MAPE,RMSE,and R2.The experimental results show that the combined prediction model performs well in oil price prediction tasks.