A promotion management method and system based on brand evaluation data analysis

Abstract

The present invention relates to the technical field of brand evaluation data processing, and specifically discloses a promotion management method and system based on brand evaluation data analysis, wherein the method collects brand evaluation data from multiple channels, uses a learning model to perform sentiment analysis on the brand evaluation data to obtain a sentiment tendency matrix, and extracts features from the brand evaluation data to obtain a key feature vector; performs feature fusion based on the sentiment tendency matrix and the key feature vector, performs data screening in the frequency domain, and determines a promotion strategy based on the screened data. Through the present invention, brand evaluation data can be used more efficiently, brand promotion strategies can be optimized, and market competitiveness can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of brand evaluation data processing, and in particular to a promotion management method and system based on brand evaluation data analysis. Background Art

[0002] In existing technologies, brand promotion often relies on traditional market research and advertising strategies. However, with the development of the Internet and social media, consumers' evaluation and feedback on brands are becoming more diverse and real-time. These evaluation data contain a lot of valuable information, but how to efficiently collect, analyze and apply these data to optimize brand promotion strategies remains an urgent problem to be solved.

[0003] Therefore, the present invention provides a promotion management method and system based on brand evaluation data analysis. Summary of the invention

[0004] The purpose of the present invention is to provide a promotion management method and system based on brand evaluation data analysis to solve the following technical problems: to improve brand promotion strategies and enhance brand awareness and market competitiveness by collecting, analyzing and utilizing brand evaluation data. The system uses a machine learning model to perform sentiment analysis on brand evaluation data, extract key feature vectors from brand evaluation data, analyze brand evaluation data from different angles, and filter data in the frequency domain, which can reduce interference during data filtering, improve the accuracy of data filtering, and thus improve the effectiveness of brand promotion strategy formulation.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A promotion management method based on brand evaluation data analysis, the method comprising:

[0007] Obtain brand evaluation data from multiple channels;

[0008] Using a natural language processing algorithm to classify the brand evaluation data into categories to obtain multiple categories of evaluation data;

[0009] Using a pre-trained machine learning model to perform sentiment analysis on the multiple categories of evaluation data to obtain a sentiment tendency matrix; wherein each row element of the sentiment tendency matrix represents a sentiment analysis result of a category of evaluation data;

[0010] Extracting features based on each of the multiple categories of evaluation data to obtain a key feature vector for each category of evaluation data;

[0011] The sentiment tendency matrix and the key feature vector are fused based on the following formula to obtain a fused feature matrix M:

[0012] ;

[0013] Wherein, M is the fusion feature matrix, σ is the activation function, f and g are nonlinear transformation functions, m is the sentiment tendency matrix, Vi is the key feature vector, α and βi are weight coefficients;

[0014] Perform frequency domain conversion based on the fusion feature matrix to obtain frequency domain feature data;

[0015] Perform frequency analysis based on the frequency domain feature data, discard frequencies that do not meet preset conditions, and obtain target frequency domain data;

[0016] Performing frequency domain inverse conversion based on the target frequency domain data to obtain effective brand evaluation data;

[0017] A promotion strategy is determined based on the effective brand evaluation data.

[0018] As a further solution of the present invention: the feature extraction is performed on each category of the multiple categories of evaluation data to obtain a key feature vector of each category of evaluation data, including:

[0019] Extracting keywords from each category of the multiple categories of evaluation data to obtain keyword features;

[0020] The keyword features corresponding to each type of evaluation data are mapped into a vector space to obtain a key feature vector of each type of evaluation data.

[0021] As a further solution of the present invention: performing frequency analysis based on the frequency domain feature data, discarding frequencies that do not meet preset conditions, and obtaining target frequency domain data, includes:

[0022] Determining the correspondence between the frequency domain components in the frequency domain feature data and the user emotion trends in the brand evaluation data;

[0023] Based on the corresponding relationship, determining a preset condition for frequency screening;

[0024] The frequency domain feature data is screened based on the preset condition of the frequency screening, and the screened data is used as the target frequency domain data.

[0025] As a further solution of the present invention: determining the correspondence between the frequency domain components in the frequency domain feature data and the user emotion trends in the brand evaluation data includes:

[0026] Obtain statistical analysis results of historical brand evaluation data corresponding to historical users;

[0027] Based on the statistical analysis results, determining the user's long-term emotional trend and short-term emotional trend;

[0028] Based on the long-term emotional trend and the short-term emotional trend, a corresponding relationship between the frequency domain component and the user emotional trend is determined.

[0029] As a further solution of the present invention: determining the correspondence between the frequency domain component and the user's emotional trend based on the long-term emotional trend and the short-term emotional trend includes:

[0030] Extracting features based on historical evaluation content corresponding to the long-term sentiment trend to obtain a first sentiment feature;

[0031] Extracting features based on historical evaluation content corresponding to the short-term sentiment trend to obtain a second sentiment feature;

[0032] Performing frequency domain conversion on the first emotion feature to obtain first frequency domain data;

[0033] Performing frequency domain conversion on the second emotion feature to obtain second frequency domain data;

[0034] A frequency domain component analysis is performed based on the first frequency domain data and the second frequency domain data, and based on a result of the frequency domain component analysis, a corresponding relationship between the frequency domain component and the user emotion trend is determined.

[0035] As a further solution of the present invention: performing frequency domain component analysis based on the first frequency domain data and the second frequency domain data includes:

[0036] Determining a first ratio of low-frequency components to high-frequency components in the first frequency domain data;

[0037] determining a second ratio of low-frequency components to high-frequency components in the second frequency domain data;

[0038] Based on the first ratio and the second ratio, a result of the frequency domain component analysis is determined.

[0039] As a further solution of the present invention: the determining the correspondence between the frequency domain component and the user emotion trend includes:

[0040] When the first ratio is greater than 0.5, determining that the low-frequency domain corresponds to the long-term emotional trend;

[0041] When the second ratio is less than 0.5, it is determined that the high-frequency frequency domain corresponds to the short-term emotional trend.

[0042] A promotion management system based on brand evaluation data analysis, the system comprising:

[0043] Data acquisition module, used to obtain brand evaluation data from multiple channels;

[0044] A classification module, used to classify the brand evaluation data by type using a natural language processing algorithm to obtain multiple categories of evaluation data;

[0045] A sentiment analysis module, used to perform sentiment analysis on the multiple categories of evaluation data using a pre-trained machine learning model to obtain a sentiment tendency matrix; wherein each row element of the sentiment tendency matrix represents a sentiment analysis result of a category of evaluation data;

[0046] A feature extraction module, used to extract features based on each of the multiple categories of evaluation data to obtain a key feature vector for each category of evaluation data;

[0047] The feature fusion module is used to perform feature fusion on the sentiment tendency matrix and the key feature vector based on the following formula to obtain a fused feature matrix M:

[0048] ;

[0049] Wherein, M is the fusion feature matrix, σ is the activation function, f and g are nonlinear transformation functions, m is the sentiment tendency matrix, Vi is the key feature vector, α and βi are weight coefficients;

[0050] A frequency domain conversion module, used for performing frequency domain conversion based on the fusion feature matrix to obtain frequency domain feature data;

[0051] A data screening module, used to perform frequency analysis based on the frequency domain feature data, discard frequencies that do not meet preset conditions, and obtain target frequency domain data;

[0052] A frequency domain inverse conversion module, used to perform frequency domain inverse conversion based on the target frequency domain data to obtain effective brand evaluation data;

[0053] A strategy determination module is used to determine the promotion strategy based on the effective brand evaluation data.

[0054] Beneficial effects of the invention: The promotion management method based on brand evaluation data analysis of the invention can realize accurate sentiment analysis and feature extraction of brand evaluation data, filter noise data, identify periodic changes in sentiment, and formulate accurate promotion strategies based on high-quality sentiment data by comprehensively utilizing natural language processing, machine learning, frequency domain analysis and other technical means. It effectively ensures the accuracy and reliability of data analysis and provides strong technical support for brand management and market promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The present invention will be further described below in conjunction with the accompanying drawings.

[0056] Figure 1 It is a flow chart of a promotion management method based on brand evaluation data analysis of the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] See also Figure 1 As shown, the present invention is a promotion management method based on brand evaluation data analysis. Figure 1 is a flow chart of a promotion management method based on brand evaluation data analysis of the present invention. In some embodiments, Figure 1 The method shown may be executed by a processing device (eg, a processor) or a promotion management system based on brand evaluation data analysis. Exemplarily, process 100 may include the following operations.

[0059] Step 101, obtaining brand evaluation data from multiple channels.

[0060] In some embodiments, the processor may collect brand evaluation data from multiple data sources (channels). For example, the multiple data sources may include social media, e-commerce platforms, product review websites, questionnaires, forum discussions, etc.

[0061] Since data from different channels may have different formats, after obtaining brand evaluation data from multiple channels, the brand evaluation data may be preprocessed, for example, data cleaning and integration.

[0062] Step 102: Classify the brand evaluation data by type using a natural language processing algorithm to obtain multiple categories of evaluation data.

[0063] In step 102 , the criteria for classifying brand evaluation data using natural language processing (NLP) technology may be the subject of the evaluation content, the emotional tendency of the evaluation (eg, positive, negative, neutral), and the like.

[0064] In some embodiments, the natural language processing technology used may include word segmentation, part-of-speech tagging, and sentiment analysis.

[0065] For example, the LDA (Latent Dirichlet Allocation) topic model can be used to classify brand evaluation data into categories such as product quality, after-sales service, and logistics.

[0066] For example, the pre-trained BERT model can be used to perform sentiment classification on brand evaluation data, dividing it into positive, negative, and neutral evaluations.

[0067] Step 103: Use the pre-trained machine learning model to perform sentiment analysis on the multiple category evaluation data to obtain a sentiment tendency matrix.

[0068] In step 103, a pre-trained machine learning model is used to perform sentiment analysis on the classified evaluation data to generate a sentiment tendency matrix. Each row element of the sentiment tendency matrix represents a sentiment analysis result for a class of evaluation data. The sentiment analysis result for a class of evaluation data represented by each row element can be a positive sentiment score or a negative sentiment score, etc.

[0069] For example, a sentiment analysis model (such as an LSTM-based neural network, etc.) can be pre-trained to output a sentiment score for each review.

[0070] Each row in the sentiment matrix represents the sentiment score (such as positive, negative, neutral) of each brand evaluation data in a certain category.

[0071] The pre-trained machine learning model may also be other models, for example, a bidirectional LSTM network, a deep neural network, etc.

[0072] Step 104 , performing feature extraction based on each of the multiple categories of evaluation data to obtain a key feature vector for each category of evaluation data.

[0073] In step 104, feature extraction is performed on the evaluation data of each category to extract representative key feature vectors. Feature extraction methods may include bag-of-words (BoW), TF-IDF vectors, word embedding (such as Word2Vec, GloVe), etc.

[0074] For example, TF-IDF can be used to extract keyword features in each type of evaluation data, and then the Word2Vec model can be used to map each evaluation into a vector space to obtain the key feature vector of each type of evaluation data.

[0075] In some embodiments, feature extraction is performed on each of the multiple categories of evaluation data to obtain a key feature vector for each category of evaluation data, including: keyword extraction is performed on each category of the multiple categories of evaluation data to obtain keyword features; vector space mapping is performed on the keyword features corresponding to each category of evaluation data to obtain the key feature vector for each category of evaluation data.

[0076] Step 105, performing feature fusion on the sentiment tendency matrix and the key feature vector based on the following formula to obtain a fused feature matrix M:

[0077] ;

[0078] Among them, M is the fusion feature matrix, σ is the activation function, f and g are nonlinear transformation functions, m is the sentiment tendency matrix, Vi is the key feature vector, α and βi are weight coefficients.

[0079] In some embodiments, an MLP may be used to perform nonlinear transformation on the sentiment tendency matrix and key feature vectors.

[0080] In some embodiments, ReLU can be used as an activation function and combined with weight coefficients to fuse features.

[0081] In this embodiment, the sentiment tendency matrix m and the vector Vi are subjected to nonlinear transformations f and g, and weighted averaged by weights α and βi, and nonlinearity is introduced by the activation function σ, thereby obtaining a new fusion feature matrix M. The fusion process can capture the complex relationship between the sentiment tendency matrix and the key feature vectors, and improve the expressiveness of the sentiment analysis features.

[0082] By extracting features from each type of evaluation data, we obtain key feature vectors, and based on a specific formula, we fuse the sentiment matrix and key feature vectors to obtain the fused feature matrix M. This method can comprehensively consider sentiment features and evaluation content features, making the analysis results more comprehensive and accurate.

[0083] Step 106: Perform frequency domain conversion based on the fusion feature matrix to obtain frequency domain feature data.

[0084] Frequency domain conversion refers to converting the fused feature matrix from time / space data to frequency domain data. For example, Fourier transform, wavelet transform, etc. can be used to convert the time / space domain data to the frequency domain in order to analyze the frequency components of the data.

[0085] Step 107: Perform frequency analysis based on the frequency domain feature data, discard frequencies that do not meet preset conditions, and obtain target frequency domain data.

[0086] By analyzing the frequency domain characteristic data, frequency components that do not meet the preset conditions (such as low-frequency noise, high-frequency interference, etc.) can be filtered out and the effective frequency components can be retained.

[0087] For example, you can set a frequency threshold to filter out frequency components below the threshold, or you can retain frequency components within a specific frequency band and remove other frequency components.

[0088] The target frequency domain data is the frequency domain data after discarding the frequencies that do not meet the preset conditions.

[0089] Through frequency domain conversion, the fusion feature matrix is ​​converted into frequency domain feature data, and frequency analysis is performed to discard frequencies that do not meet the preset conditions. This can effectively filter out noise data, retain valuable emotional information, enhance data quality, ensure data purity, and improve the reliability of analysis results.

[0090] In some embodiments, the performing frequency analysis based on the frequency domain feature data, discarding frequencies that do not meet preset conditions, and obtaining target frequency domain data includes:

[0091] Determining the correspondence between the frequency domain components in the frequency domain feature data and the user emotion trends in the brand evaluation data;

[0092] Based on the corresponding relationship, determining a preset condition for frequency screening;

[0093] The frequency domain feature data is screened based on the preset condition of the frequency screening, and the screened data is used as the target frequency domain data.

[0094] User sentiment trends refer to the trend of user sentiment changes over time, which can be divided into long-term sentiment trends and short-term sentiment trends. Long-term sentiment trends reflect the overall direction of user sentiment, which may require months or even years of data to observe. Short-term sentiment trends, on the other hand, reflect the fluctuations of user sentiment in a shorter period of time, which can usually be observed within a few days or weeks.

[0095] The correspondence relationship refers to the relationship between different frequency components in the frequency domain feature data and the user's emotional trend (long-term or short-term). Through statistical analysis of historical brand evaluation data, it is possible to determine which frequency components correspond to long-term emotional trends and which correspond to short-term emotional trends.

[0096] The preset condition is a standard used to filter frequency domain feature data. For example, if low-frequency components correspond to long-term emotional trends and high-frequency components correspond to short-term emotional trends, the preset condition may set a threshold, and frequency components below this threshold are considered useful, otherwise they will be discarded.

[0097] According to the correspondence between the frequency domain feature data and the user's emotional trend, a specific frequency screening criterion can be determined. For example, if the analysis finds that the low-frequency components of the frequency domain feature data mainly correspond to the long-term emotional trend, the preset condition can be to retain all frequency components below the preset frequency domain.

[0098] In this embodiment, according to the corresponding relationship of the user's emotional trend, the preset conditions of frequency screening are set to screen the frequency domain feature data, so that the frequency domain screening conditions can be flexibly adjusted according to specific analysis requirements to ensure that the target frequency domain data finally obtained has high relevance and representativeness.

[0099] In some embodiments, determining the correspondence between the frequency domain components in the frequency domain feature data and the user emotion trends in the brand evaluation data includes:

[0100] Obtain statistical analysis results of historical brand evaluation data corresponding to historical users;

[0101] Based on the statistical analysis results, determining the user's long-term emotional trend and short-term emotional trend;

[0102] Based on the long-term emotional trend and the short-term emotional trend, a corresponding relationship between the frequency domain component and the user emotional trend is determined.

[0103] Historical brand evaluation data refers to user evaluation information collected by a brand in a certain period of time in the past. This data usually includes user comments, feedback, ratings, etc., and can come from social media, evaluation platforms, questionnaires and other channels.

[0104] The purpose of statistical analysis of historical evaluation data is to identify the overall trends and fluctuations in user sentiment. Statistical analysis usually includes sentiment analysis (sentiment classification as positive, neutral or negative) and time series analysis to observe how user sentiment towards the brand changes over time.

[0105] After the brand evaluation data is classified by sentiment through natural language processing (NLP) technology and machine learning models, the results are sorted by time dimension to form a sentiment time series. After that, the basic characteristics of user sentiment trends can be obtained by statistical analysis of these sentiment data.

[0106] In this embodiment, by determining the correspondence between the frequency domain components in the frequency domain feature data and the user emotion trends in the brand evaluation data, the long-term and short-term trends in the user emotions can be identified. By using historical brand evaluation data for statistical analysis, extracting long-term and short-term emotion features, and performing frequency domain conversion and component analysis, the periodic changes in brand emotions can be effectively identified, thereby providing strong data support for brand management and strategy formulation.

[0107] In some embodiments, determining the correspondence between the frequency domain component and the user emotion trend based on the long-term emotion trend and the short-term emotion trend includes:

[0108] Extracting features based on historical evaluation content corresponding to the long-term sentiment trend to obtain a first sentiment feature;

[0109] Extracting features based on historical evaluation content corresponding to the short-term sentiment trend to obtain a second sentiment feature;

[0110] Performing frequency domain conversion on the first emotion feature to obtain first frequency domain data;

[0111] Performing frequency domain conversion on the second emotion feature to obtain second frequency domain data;

[0112] A frequency domain component analysis is performed based on the first frequency domain data and the second frequency domain data, and based on a result of the frequency domain component analysis, a corresponding relationship between the frequency domain component and the user emotion trend is determined.

[0113] The first emotional feature is a feature extracted from the historical evaluation content corresponding to the long-term emotional trend, which can reflect the user's emotional state in the long term.

[0114] The second emotional feature is a feature extracted from the historical evaluation content corresponding to the short-term emotional trend, which can reflect the user's emotional fluctuations in the short term.

[0115] In some embodiments, performing frequency domain component analysis based on the first frequency domain data and the second frequency domain data includes:

[0116] Determining a first ratio of low-frequency components to high-frequency components in the first frequency domain data;

[0117] determining a second ratio of low-frequency components to high-frequency components in the second frequency domain data;

[0118] Based on the first ratio and the second ratio, a result of the frequency domain component analysis is determined.

[0119] In some embodiments, determining the correspondence between the frequency domain component and the user emotion trend includes:

[0120] When the first ratio is greater than 0.5, determining that the low-frequency domain corresponds to the long-term emotional trend;

[0121] When the second ratio is less than 0.5, it is determined that the high-frequency frequency domain corresponds to the short-term emotional trend.

[0122] In this embodiment, by analyzing the ratio of low-frequency and high-frequency components in the first frequency domain data and the second frequency domain data, long-term emotional trends and short-term emotional trends can be distinguished, so that brands can more clearly understand the dynamic changes in user emotions and adopt appropriate promotion strategies for different emotional cycles.

[0123] Step 108: Perform frequency domain inverse conversion based on the target frequency domain data to obtain effective brand evaluation data.

[0124] Inverse frequency domain conversion is to convert frequency domain data from the frequency domain to the time / space domain.

[0125] In some embodiments, effective brand evaluation data may be obtained by applying inverse Fourier transform to target frequency domain data.

[0126] Step 109: Determine a promotion strategy based on the effective brand evaluation data.

[0127] By using the processed effective brand evaluation data, we can analyze the brand’s performance and user feedback in different aspects, and then formulate corresponding promotion strategies, including content marketing, advertising, product improvement, etc.

[0128] For example, you can analyze the positive and negative sentiment scores in valid brand evaluation data to identify hot issues that users are concerned about. Based on the analysis results, you can optimize advertising strategies and focus on promoting product features that users are satisfied with. In response to negative feedback, you can develop product improvement plans to improve user satisfaction.

[0129] The present invention uses natural language processing (NLP) algorithms to classify brand evaluation data from multiple channels, and uses a pre-trained machine learning model to perform sentiment analysis to obtain a sentiment tendency matrix. This ensures accurate sentiment analysis of different categories of evaluation data, and helps to fully understand users' emotional attitudes towards different aspects of the brand.

[0130] Working principle of the invention: The invention uses natural language processing and machine learning models to realize automatic sentiment analysis and feature extraction of brand evaluation data, and realizes intelligent processing of sentiment data through frequency domain analysis and conversion. The whole process is highly automated, reducing human intervention and subjective bias, and improving analysis efficiency and accuracy. And through these steps, brand evaluation data can be systematically analyzed, effective information can be extracted, and data-driven promotion strategies can be formulated to improve brand influence and market competitiveness.

[0131] Based on the same inventive concept, this specification also provides a promotion management system based on brand evaluation data analysis, the system comprising:

[0132] Data acquisition module, used to obtain brand evaluation data from multiple channels;

[0133] A classification module, used to classify the brand evaluation data by type using a natural language processing algorithm to obtain multiple categories of evaluation data;

[0134] A sentiment analysis module, used to perform sentiment analysis on the multiple categories of evaluation data using a pre-trained machine learning model to obtain a sentiment tendency matrix; wherein each row element of the sentiment tendency matrix represents a sentiment analysis result of a category of evaluation data;

[0135] A feature extraction module, used to extract features based on each of the multiple categories of evaluation data to obtain a key feature vector for each category of evaluation data;

[0136] The feature fusion module is used to perform feature fusion on the sentiment tendency matrix and the key feature vector based on the following formula to obtain a fused feature matrix M:

[0137] ;

[0138] Wherein, M is the fusion feature matrix, σ is the activation function, f and g are nonlinear transformation functions, m is the sentiment tendency matrix, Vi is the key feature vector, α and βi are weight coefficients;

[0139] A frequency domain conversion module, used for performing frequency domain conversion based on the fusion feature matrix to obtain frequency domain feature data;

[0140] A data screening module, used to perform frequency analysis based on the frequency domain feature data, discard frequencies that do not meet preset conditions, and obtain target frequency domain data;

[0141] A frequency domain inverse conversion module, used to perform frequency domain inverse conversion based on the target frequency domain data to obtain effective brand evaluation data;

[0142] A strategy determination module is used to determine the promotion strategy based on the effective brand evaluation data.

[0143] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A promotion management method based on brand evaluation data analysis, characterized in that: The method comprises: Obtain brand evaluation data from multiple channels; Using a natural language processing algorithm to classify the brand evaluation data into categories to obtain multiple categories of evaluation data; Using a pre-trained machine learning model to perform sentiment analysis on the multiple categories of evaluation data to obtain a sentiment tendency matrix; wherein each row element of the sentiment tendency matrix represents a sentiment analysis result of a category of evaluation data; Extracting features based on each of the multiple categories of evaluation data to obtain a key feature vector for each category of evaluation data; The sentiment tendency matrix and the key feature vector are fused based on the following formula to obtain a fused feature matrix M: ； Wherein, M is the fusion feature matrix, σ is the activation function, f and g are nonlinear transformation functions, m is the sentiment tendency matrix, Vi is the key feature vector of the i-th category evaluation data, α and βi are weight coefficients, k is the total number of categories of evaluation data, i∈[1, k] and is a positive integer; Perform frequency domain conversion based on the fusion feature matrix to obtain frequency domain feature data; Perform frequency analysis based on the frequency domain feature data, discard frequencies that do not meet preset conditions, and obtain target frequency domain data; Perform frequency domain inverse conversion based on the target frequency domain data to obtain effective brand evaluation data; A promotion strategy is determined based on the effective brand evaluation data.

2. A promotion management method based on brand evaluation data analysis according to claim 1, characterized in that: The feature extraction is performed on each of the multiple categories of evaluation data to obtain a key feature vector of each category of evaluation data, including: Extracting keywords from each category of the multiple categories of evaluation data to obtain keyword features; The keyword features corresponding to each type of evaluation data are mapped into a vector space to obtain a key feature vector of each type of evaluation data.

3. A promotion management method based on brand evaluation data analysis according to claim 1, characterized in that: The frequency analysis is performed based on the frequency domain feature data, and frequencies that do not meet the preset conditions are discarded to obtain target frequency domain data, including: Determining the correspondence between the frequency domain components in the frequency domain feature data and the user emotion trends in the brand evaluation data; Based on the corresponding relationship, determining a preset condition for frequency screening; The frequency domain feature data is screened based on the preset condition of the frequency screening, and the screened data is used as the target frequency domain data.

4. A promotion management method based on brand evaluation data analysis according to claim 3, characterized in that: The determining of the correspondence between the frequency domain components in the frequency domain feature data and the user emotion trends in the brand evaluation data includes: Obtain statistical analysis results of historical brand evaluation data corresponding to historical users; Based on the statistical analysis results, determining the user's long-term emotional trend and short-term emotional trend; Based on the long-term emotional trend and the short-term emotional trend, a corresponding relationship between the frequency domain component and the user emotional trend is determined.

5. A promotion management method based on brand evaluation data analysis according to claim 4, characterized in that: The determining, based on the long-term emotional trend and the short-term emotional trend, the corresponding relationship between the frequency domain component and the user emotional trend comprises: Extracting features based on historical evaluation content corresponding to the long-term sentiment trend to obtain a first sentiment feature; Extracting features based on historical evaluation content corresponding to the short-term sentiment trend to obtain a second sentiment feature; Performing frequency domain conversion on the first emotion feature to obtain first frequency domain data; Performing frequency domain conversion on the second emotion feature to obtain second frequency domain data; A frequency domain component analysis is performed based on the first frequency domain data and the second frequency domain data, and based on a result of the frequency domain component analysis, a corresponding relationship between the frequency domain component and the user emotion trend is determined.

6. A promotion management method based on brand evaluation data analysis according to claim 5, characterized in that: Performing frequency domain component analysis based on the first frequency domain data and the second frequency domain data includes: Determining a first ratio of low-frequency components to high-frequency components in the first frequency domain data; determining a second ratio of low-frequency components to high-frequency components in the second frequency domain data; Based on the first ratio and the second ratio, a result of the frequency domain component analysis is determined.

7. A promotion management method based on brand evaluation data analysis according to claim 6, characterized in that: The determining the correspondence between the frequency domain component and the user emotion trend includes: When the first ratio is greater than 0.5, determining that the low-frequency domain corresponds to the long-term emotional trend; When the second ratio is less than 0.5, it is determined that the high-frequency frequency domain corresponds to the short-term emotional trend.

8. A promotion management system based on brand evaluation data analysis, characterized in that: The system comprises: Data acquisition module, used to obtain brand evaluation data from multiple channels; A classification module, used to classify the brand evaluation data by type using a natural language processing algorithm to obtain multiple categories of evaluation data; A sentiment analysis module, used to perform sentiment analysis on the multiple categories of evaluation data using a pre-trained machine learning model to obtain a sentiment tendency matrix; wherein each row element of the sentiment tendency matrix represents a sentiment analysis result of a category of evaluation data; A feature extraction module, used to extract features based on each of the multiple categories of evaluation data to obtain a key feature vector for each category of evaluation data; The feature fusion module is used to perform feature fusion on the sentiment tendency matrix and the key feature vector based on the following formula to obtain a fused feature matrix M: ； Wherein, M is the fusion feature matrix, σ is the activation function, f and g are nonlinear transformation functions, m is the sentiment tendency matrix, Vi is the key feature vector of the i-th category evaluation data, α and βi are weight coefficients, k is the total number of categories of evaluation data, i∈[1, k] and is a positive integer; A frequency domain conversion module, used for performing frequency domain conversion based on the fusion feature matrix to obtain frequency domain feature data; A data screening module, used to perform frequency analysis based on the frequency domain feature data, discard frequencies that do not meet preset conditions, and obtain target frequency domain data; A frequency domain inverse conversion module, used to perform frequency domain inverse conversion based on the target frequency domain data to obtain effective brand evaluation data; A strategy determination module is used to determine a promotion strategy based on the effective brand evaluation data.

Images