Aquatic product culture economic prediction method based on multi-source data and sentiment analysis

Abstract

The invention discloses an aquatic product culture economic prediction method based on multi-source data and sentiment analysis. The method comprises the following steps: acquiring structured data and unstructured data of a set aquatic product fishery in a required time period; an emotion analysis model generated based on a dynamic semantic vector is used for extracting emotion features in the unstructured data, an emotion time sequence is generated, and the emotion features comprise emotion tendency and emotion intensity; the method comprises the following steps: constructing an aquatic product culture economic prediction model, and inputting structured prediction variables in a required time period during preliminary training, including one or more of meteorological parameters, market price parameters and culture scale parameters; and inputting the emotional time sequence as an additional predictive variable, and further training the aquatic product culture economic prediction model in combination with structured data to optimize the model performance.

Description

Technical Field

[0001] The present invention relates to the technical field of economic forecasting, and in particular to an economic forecasting method for aquaculture based on multi-source data fusion and sentiment analysis, which is suitable for making accurate economic forecasts when policy regulation or market sentiment fluctuations are significant. Background Art

[0002] With the development of deep learning technology, economic forecasting has gradually shifted from relying on structured data such as historical market data and meteorological conditions to utilizing unstructured data such as social media sentiment fluctuations and policy comments. However, current aquaculture economic forecasts are still mainly based on traditional structured data analysis, ignoring the impact of non-quantitative factors such as market sentiment and policy dynamics, which leads to poor performance of forecasting models in the face of sudden market fluctuations.

[0003] Existing aquaculture economic forecasting models, such as time series forecasting methods based on ARIMA and RNN, are usually difficult to effectively capture the complex emotional fluctuations in the market, especially unstructured data cannot be used for effective forecasting. In addition, existing sentiment analysis methods mainly rely on static word vectors, which are difficult to capture the dynamic changes of sentiment characteristics, thus affecting the accurate understanding of complex emotions, which leads to poor prediction results of the model in complex market environments.

[0004] Based on the above challenges, there is an urgent need for an economic forecasting method that can integrate structured and unstructured data and dynamically reflect fluctuations in market sentiment, so as to improve the accuracy of economic forecasts for aquaculture and its adaptability to market changes. Summary of the invention

[0005] The purpose of the present invention is to provide a method for economic forecasting of aquatic product breeding based on multi-source data and sentiment analysis, so as to solve the problem that it is difficult for relevant aquatic product breeding price forecasting methods to reflect the impact of unstructured data on aquatic product breeding price forecasting.

[0006] In order to solve the above technical problems, the present invention provides a method for predicting the economics of aquaculture based on multi-source data and sentiment analysis, comprising the following steps:

[0007] Obtain structured and unstructured data of the aquatic product market for a set aquatic product fishery within the required time period;

[0008] Extracting sentiment features from unstructured data using a sentiment analysis model based on dynamic semantic vector generation to generate sentiment time series, wherein the sentiment features include sentiment tendency and sentiment intensity;

[0009] Constructing an aquatic product breeding economic forecasting model, inputting structured forecasting variables within the required time period, wherein the variables include one or more of meteorological parameters, market price parameters and breeding scale parameters, and performing preliminary training of the model;

[0010] The emotional time series is then input as a prediction variable, combined with structured data, to further train the aquatic product breeding economic prediction model to optimize model performance.

[0011] Preferably, the structured data for setting a fishery for aquatic products within the required time period includes meteorological data, market price data and aquaculture scale historical records.

[0012] Preferably, a web crawler is used to obtain relevant information containing preset keywords, and unstructured data related to aquaculture of aquatic products is collected from public data sources.

[0013] Preferably, the crawled unstructured data is preprocessed, including text standardization, sentiment word tagging, word segmentation and stop word filtering, and data screening and cleaning, and then a sentiment analysis model based on dynamic semantic vector generation is used to extract sentiment features in the unstructured data to generate a sentiment time series.

[0014] Preferably, the aquatic product breeding economic forecasting model includes a convolutional neural network, a bidirectional long short-term memory network and a genetic algorithm combined model, which is used to predict aquatic product breeding economic data.

[0015] Preferably, the genetic algorithm is used to optimize the hyperparameters of the aquatic product farming economic forecasting model to improve the accuracy and generalization ability of the forecast.

[0016] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0017] Compared with the prior art, the present invention significantly improves the prediction accuracy of the aquaculture economic forecasting model by integrating structured data and unstructured data, especially in a complex and changing market environment. By introducing advanced sentiment analysis technology, the model can capture subtle changes in market sentiment in real time, thereby achieving a more accurate prediction of the economic performance of aquaculture. This dynamic sentiment feedback mechanism enables the forecasting model to not only cope with fluctuations in historical data, but also adapt to sudden market changes.

[0018] In addition, the present invention uses a genetic algorithm to optimize the hyperparameters of the model, which greatly improves the generalization ability of the model and makes its performance on unseen data more stable. This optimization strategy ensures the adaptability and robustness of the model under different market conditions, thereby improving the overall prediction accuracy and reliability. Therefore, the aquaculture economic forecasting method of the present invention is not only efficient and practical, but also provides strong technical support for practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is an implementation flow chart of an embodiment of the present invention.

[0020] Figure 2 It is a flow chart of sentiment analysis generated based on dynamic semantic vectors in the present invention.

[0021] Figure 3 It is a price prediction model diagram of the present invention. DETAILED DESCRIPTION

[0022] The aquaculture economic forecasting method based on multi-source data and sentiment analysis proposed by the present invention will be further described in detail below in conjunction with the accompanying drawings and specific examples. The advantages and features of the present invention will become clearer based on the following description.

[0023] The research of the present invention shows that the economic forecast of aquatic product farming relies on a large amount of data and information. The current economic forecasting models of aquatic product farming mainly focus on the use of structured data, such as meteorological data, market prices and farming scale for forecasting. However, these models often ignore the impact of unstructured data such as market sentiment and policy dynamics. In response to market fluctuations, especially policy changes or drastic changes in market sentiment, traditional models are difficult to quickly and accurately reflect these changes. In addition, existing sentiment analysis methods mostly rely on static word vectors, which make it difficult to capture the dynamic changes of sentiment characteristics, resulting in poor prediction results of the model in a complex market environment. Therefore, the existing economic forecasting methods of aquatic product farming have great limitations in dealing with sudden market fluctuations and complex sentiment information, and cannot fully predict the economic performance of the aquatic product market.

[0024] Based on this, the core idea of ​​the present invention is to construct an economic forecasting model for aquatic product farming by combining multi-source data with sentiment analysis. The method first obtains structured and unstructured data, extracts sentiment features and generates sentiment time series, and then integrates them into model training as additional predictive variables to improve the accuracy of prediction and model performance. This method not only effectively captures the fluctuations in market sentiment, but also better responds to sudden changes in the market, thereby improving the accuracy of aquatic product farming economic forecasts and the generalization ability of the model, significantly improving the timeliness and accuracy of forecasts, and providing a scientific basis for economic decision-making in the aquatic product farming market.

[0025] For specific implementation, please refer to Figure 1 , which is a schematic diagram of an embodiment of the present invention. Figure 1 As shown in the figure, the economic forecasting method of aquaculture based on multi-source data and sentiment analysis includes the following steps:

[0026] Step S1: Obtain structured data and unstructured data of the aquatic product market within a required time period;

[0027] Specifically, structured data includes historical records such as meteorological data, market price data, and breeding scale. These data can be obtained from public statistical data, weather stations, market analysis reports, and other data sources.

[0028] Specifically, unstructured data was obtained from public data sources such as news, market comments and policy documents through web crawlers. Using preset keywords such as "aquatic product prices" and "market fluctuations", the initial links from September 1, 2021 to September 1, 2024 were crawled to obtain 29,987 relevant texts.

[0029] Step S2: Use the sentiment analysis model based on dynamic semantic vector generation to extract sentiment features from unstructured data and generate sentiment time series.

[0030] For details, please refer to Figure 2 , Figure 2 This is a flow chart of sentiment analysis based on dynamic semantic vector generation in the present invention. In a possible implementation, S2 specifically includes sub-steps S2.1 to S2.3:

[0031] Step S21: Specifically, the crawled unstructured data is preprocessed, including the following steps: first, through text standardization, the original text is converted into a unified format to ensure data consistency; second, the sentiment words in the text are annotated using a sentiment dictionary to lay the foundation for subsequent sentiment analysis; then, the text is segmented and stop words are filtered, and the text is divided into independent words through the word segmentation operation, and meaningless stop words are removed to reduce noise in the data; finally, data screening and cleaning are performed to eliminate redundant and invalid information, and screen out data that is valuable for sentiment analysis and prediction, thereby obtaining a high-quality and effective data set.

[0032] Step S22: Specifically, the data is processed using a dynamic semantic vector generation method to obtain a context-related dynamic semantic vector. Compared with traditional static word vector generation methods such as Word2Vec or GloVe, the dynamic semantic vector generation method can dynamically adjust the semantic representation of each token according to the context through the self-attention mechanism of GPT, especially when dealing with polysemous words and complex contexts, which can significantly improve the accuracy of sentiment analysis. The traditional method assigns a fixed vector to each word and cannot distinguish its meaning in different contexts, resulting in poor analysis performance in complex emotional scenarios. Through dynamic semantic vector generation, this method can not only capture long-distance dependencies in the text, but also handle subtle emotional fluctuations, thereby providing more accurate sentiment analysis results. This advantage is particularly suitable for dynamic and changeable market sentiment analysis, and can provide more reliable input data for economic forecasting.

[0033] In a possible implementation, step S22 specifically includes sub-steps S221 to S223:

[0034] Step S221: First, a text set D consisting of unstructured data (such as news reports, market reviews, policy documents, etc.) is constructed. N} Input to the dynamic semantic vector generation module. Since the input text data is unstructured, it must be tokenized, that is, the text is broken down into small units that the model can recognize.

[0035] During the implementation, the tokenization operation is performed using the GPT model’s own tokenizer (such as GPT2Tokenizer). i ∈D, the tokenization process can be expressed as:

[0036]

[0037] Where T(·) represents the tokenization function, L i It is a document i The number of tokens.

[0038] This process breaks down the input text into words, subwords, or characters to suit the model's processing needs. For example, for the input "aquatic product prices have risen", the model will break it down into a token sequence:

[0039] S 示例 =["water","product","price","increase"]

[0040] Step S222: After the text tokenization is completed, the GPT model will use its self-attention mechanism to process the input sequence. This mechanism allows each token to interact with the context of the entire sentence to generate context-dependent dynamic semantic representations. Through the multi-layer Transformer structure of GPT, the input token sequence is processed, and each layer will output a more accurate semantic vector. The specific implementation steps are as follows:

[0041] 1. Embedding vector mapping: The tokenized token sequence S i Input to the embedding layer of the GPT model to generate the corresponding embedding vector sequence X i By calling the embedding function model.get_input_embeddings(), each token is mapped to a vector representation of fixed dimension: X i =Embedding(S i )

[0042] 2. Multi-layer Transformer encoding: Using the multi-layer Transformer structure of the GPT model, embedding the vector sequence X i The model internally calls the self-attention function and the feedforward network function to process the output of each layer. The self-attention function calls model.transformer.h[l].attn; the feedforward network function calls model.transformer.h[l].mlp for nonlinear transformation. After multiple layers of encoding, the context vector sequence is obtained: H i = Transformer(X i )

[0043] 3. Context vector generation: Finally, the model generates a context-dependent dynamic semantic vector H for each token i , the output vector sequence is: Z i =H i

[0044] Step S223: After the context encoding is completed, the model generates a context-related dynamic semantic vector Z for each token i . These vectors will be used for subsequent sentiment feature extraction.

[0045] Step S23: The extracted dynamic semantic vector is input into the convolutional neural network for further emotional feature extraction, which specifically includes: convolution operation, converting the dynamic semantic vector sequence Z iInput into the one-dimensional convolution layer to extract local sentiment features; pooling operation, pooling the convolved features to obtain a feature vector of fixed length; sentiment feature mapping, mapping the feature vector to the sentiment space to obtain the sentiment score; sentiment time series generation, arranging the sentiment scores according to the timestamps of the text to form a sentiment time series S.

[0046] Step S3: construct an economic forecasting model for aquatic product breeding, input structured forecasting variables within the required time period, including one or more of meteorological parameters, market price parameters and breeding scale parameters, and perform preliminary training of the model; then input the emotional time series as a forecasting variable, combine with structured data, further train the economic forecasting model for aquatic product breeding, and optimize model performance.

[0047] For details, please refer to Figure 3 , which is a structural diagram of the aquatic product breeding economic forecasting model of the present invention. The constructed aquatic product breeding economic forecasting model adopts a convolutional neural network, a bidirectional long short-term memory network and a genetic algorithm combination model to construct an aquatic product breeding economic forecasting model.

[0048] In a possible implementation, step S3 specifically includes sub-steps S31 to S33:

[0049] The specific steps are as follows:

[0050] Step S31: Structured data input: First, the acquired structured data, including meteorological data, market prices, breeding scale, etc., are input into the model for preliminary training.

[0051] Step S32: Sentiment time series input: In the process of model optimization, the sentiment time series is used as an additional input feature and combined with structured data for further training to optimize the prediction performance of the model. The sentiment time series can help the model understand how market sentiment fluctuations affect the price of aquatic products.

[0052] Step S33: Model optimization: Use genetic algorithm to optimize the hyperparameters of the aquatic product breeding economic forecasting model to improve the model's prediction accuracy and generalization ability. The main optimization steps include:

[0053] 1. Population initialization: First, multiple hyperparameter combinations are randomly generated to form the initial population. These hyperparameters may include the convolution kernel size, pooling layer configuration, learning rate of the convolutional neural network, and the number of hidden units and time step of the bidirectional long short-term memory network (BiLSTM). Considering the complexity of multi-source data such as meteorology, market prices, and breeding scale, the diversity of the initial population is crucial to ensure that the model has enough flexibility to explore different hyperparameter combinations.

[0054] 2. Fitness evaluation: Use the training set and validation set to evaluate each hyperparameter combination in the population, and calculate its fitness based on the performance of the model in predicting the price of aquatic products. Common fitness evaluation indicators include mean square error (MSE), mean absolute error (MAE) or root mean square error (RMSE) to measure the prediction error of the model under structured data and sentiment time series input. In this way, the performance of each hyperparameter combination can be accurately evaluated to screen out individuals with better performance.

[0055] 3. Selection, crossover and mutation: After fitness evaluation, individuals with higher fitness are selected for genetic operations. The selection process ensures the retention of excellent genes, and the crossover operation combines the advantageous hyperparameter combinations of different parents to generate new hyperparameter individuals. The mutation operation introduces a small amount of random perturbations and changes the values ​​of certain hyperparameters to avoid the population from falling into a local optimal solution. Through multiple generations of genetic operations, the model hyperparameters will continue to evolve and optimize model performance.

[0056] In this optimization process, the sentiment time series, as one of the important input features, can provide the genetic algorithm with dynamic market sentiment fluctuation information. For example, when the sentiment time series shows drastic market sentiment fluctuations, the genetic algorithm can adjust the hyperparameters of BiLSTM (such as learning rate, number of hidden units, etc.) accordingly to ensure that the model can flexibly adapt to the impact of sentiment changes on aquatic product price forecasts. This combination not only optimizes the model's hyperparameters, but also improves the model's ability to cope with market complexity and uncertainty.

[0057] Step S4: Comparison of economic forecast accuracy for aquatic product farming. After the model is optimized, the forecast result is finally output as the changing trend of aquatic product market prices. During the implementation of the present invention, we systematically evaluated the accuracy of the economic forecast model for aquatic product farming. By comparing the model with the traditional ARIMA and RNN models, the results show that the forecast accuracy of this method is significantly improved when facing market sentiment fluctuations. Specifically, in the test of the same data set, the prediction error of the CNN-GA-BiLSTM model based on multi-source data and sentiment analysis was reduced by 0.15 compared with the traditional model, showing higher stability and reliability.

[0058] In the experiment, the model was used to evaluate the prediction accuracy of aquatic product prices. The results showed that the average absolute error of the CNN-GA-BiLSTM model after the introduction of sentiment features was 0.35, which was significantly better than the traditional model that only relied on structured data, with an average absolute error of 0.50. This result proves that incorporating market sentiment data into model training can effectively improve the accuracy of aquatic product economic forecasts, especially during policy changes and market fluctuations.

[0059] In summary, the economic forecasting method for aquaculture of the present invention combines dynamic sentiment analysis and multi-source data, successfully solves the problem that traditional models do not respond to market sentiment in a timely manner, significantly improves the forecasting accuracy, and provides a scientific basis for economic decision-making in the aquaculture market. By optimizing the model structure and introducing genetic algorithms for hyperparameter adjustment, this model shows better adaptability and forecasting performance in a complex market environment, providing an important reference for industry development.

Claims

1. A method for predicting the economics of aquaculture based on multi-source data and sentiment analysis, characterized by: The following steps are involved: Obtain structured and unstructured data of a set aquatic product fishery within the required time period; Extracting sentiment features from unstructured data using a sentiment analysis model based on dynamic semantic vector generation to generate sentiment time series, wherein the sentiment features include sentiment tendency and sentiment intensity; Construct an economic forecasting model for aquaculture, and input structured forecasting variables within the required time period during the initial training, including one or more of meteorological parameters, market price parameters, and aquaculture scale parameters; The sentiment time series is input as an additional predictor variable and combined with structured data to further train the aquaculture economic forecasting model to optimize model performance.

2. The aquaculture economic forecasting method based on multi-source data and sentiment analysis according to claim 1, characterized in that: The structured data for setting aquatic product fisheries within the required time period include meteorological data, market price data and breeding scale historical records.

3. The aquaculture economic forecasting method based on multi-source data and sentiment analysis according to claim 1 is characterized by: Use a web crawler to obtain relevant information containing preset keywords, and collect unstructured data related to the aquaculture of the aquatic product from public data sources.

4. The aquaculture economic forecasting method based on multi-source data and sentiment analysis according to claim 2 is characterized by: The crawled unstructured data is preprocessed, including text standardization, sentiment word tagging, word segmentation and stop word filtering, and data screening and cleaning. Then, a sentiment analysis model based on dynamic semantic vector generation is used to extract sentiment features from unstructured data and generate sentiment time series.

5. The aquaculture economic forecasting method based on multi-source data and sentiment analysis according to claim 1, characterized in that: The aquatic product breeding economic forecasting model includes a convolutional neural network, a bidirectional long short-term memory network and a genetic algorithm combined model, which is used to forecast aquatic product breeding economic data.

6. The aquaculture economic forecasting method based on multi-source data and sentiment analysis according to claim 1, characterized in that: The genetic algorithm is used to optimize the hyperparameters of the aquatic product breeding economic forecasting model to improve the accuracy and generalization ability of the forecast.

Images