Commodity evaluation monitoring method and system based on multi-dimensional data

Abstract

The invention discloses a commodity evaluation monitoring method and system based on multi-dimensional data, and relates to the technical field of evaluation monitoring, and the method comprises the steps: collecting the multi-dimensional data, and carrying out the preprocessing of the multi-dimensional data; performing feature extraction on the preprocessed multi-dimensional data, and fusing the extracted features into a comprehensive feature vector; based on the comprehensive feature vector, through a space-time convolutional neural network, identifying and evaluating the space-time distribution of the hot spots; based on the spatial and temporal distribution of the evaluation hotspots, identifying areas and time periods of the evaluation hotspots, and evaluating the trend of the hotspots according to an identification result; and performing dynamic early warning according to the trend of predicting and evaluating the hot spots, and generating an evaluation report. According to the method, spatial-temporal feature modeling is carried out on data through STCNN, and the distribution conditions of evaluation hotspots in different time and spaces are identified. Based on the spatio-temporal characteristics, the area and time period of the hot spot can be accurately positioned and evaluated, and the diffusion trend of the hot spot can be predicted and evaluated.

Description

Technical Field

[0001] The present invention relates to the technical field of evaluation and monitoring, and in particular to a method and system for monitoring product evaluations based on multi-dimensional data. Background Art

[0002] Although existing evaluation and monitoring methods can analyze consumer feedback to a certain extent, there are still obvious deficiencies. First, existing methods often rely too much on text data and cannot effectively process multi-modal information such as images, videos, and audio. Consumers' ways of expressing opinions through social platforms and O2O platforms are becoming increasingly diverse, and text data is often only a part of it. Therefore, evaluation and monitoring methods that rely solely on text analysis are difficult to comprehensively capture the full picture of user feedback. Second, existing evaluation and monitoring methods usually cannot dynamically perceive the spatio-temporal changes of evaluations. Especially on O2O platforms, user feedback has obvious regional and temporal characteristics, and consumer feedback in different regions and different time periods may vary significantly. Existing technologies are difficult to accurately identify the spatio-temporal distribution of evaluation hotspots and their changing trends. These deficiencies seriously affect the brand owners' comprehensive control of market dynamics and also limit their ability to respond promptly to negative evaluations. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides a method for monitoring product evaluations based on multi-dimensional data to solve the problems that existing technologies cannot effectively process multi-modal information and dynamically perceive the spatio-temporal changes of evaluations.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a method for monitoring product evaluations based on multi-dimensional data, which includes collecting multi-dimensional data and preprocessing the multi-dimensional data; extracting features from the preprocessed multi-dimensional data and fusing the extracted features into a comprehensive feature vector; based on the comprehensive feature vector, identifying the spatio-temporal distribution of evaluation hotspots through a spatio-temporal convolutional neural network; based on the spatio-temporal distribution of evaluation hotspots, identifying the regions and time periods of evaluation hotspots and predicting the trends of evaluation hotspots according to the identification results; dynamically warning according to the predicted trends of evaluation hotspots and generating an evaluation report.

[0006] As a preferred solution of the method for monitoring product evaluations based on multi-dimensional data according to the present invention, wherein: the multi-dimensional data includes text data, image data, video data, audio data, and time and geographical location information.

[0007] As a preferred solution of the method for monitoring product evaluations based on multi-dimensional data according to the present invention, wherein: the preprocessing of the multi-dimensional data is specifically carried out as follows, Denoise the text data by removing special symbols and stop words; Adjust the scale of the image data by interpolation method to unify the size; Reduce the redundancy of video data by frame sampling method; Remove the background noise of the audio data by filter; Format the time and geographical location information by standardization rules.

[0008] As a preferred solution of the commodity evaluation monitoring method based on multi-dimensional data according to the present invention, wherein: extract features from the preprocessed multi-dimensional data, and fuse the extracted features into a comprehensive feature vector. The specific steps are as follows. Use BERT to extract the deep semantic feature vector of the text data; Use ResNet-50 to extract the spatial feature vector of the image data; Use C3D to extract the temporal and spatial feature vectors of the video data; Extract the frequency feature vector of the audio data by Mel spectrogram conversion and combining with CNN; Extract the periodic feature vector of the time and geographical location information by sine time encoding combined with regional embedding encoding; Introduce the attention mechanism to dynamically allocate the attention weights of different feature vectors; Based on the attention weights of different feature vectors, use the multi-modal fusion algorithm to perform feature interaction calculation, capture the non-linear relationship between different features, and generate a comprehensive feature vector.

[0009] As a preferred solution of the commodity evaluation monitoring method based on multi-dimensional data according to the present invention, wherein: based on the comprehensive feature vector, identify the spatio-temporal distribution of evaluation hotspots through a spatio-temporal convolutional neural network. The specific steps are as follows. Divide the comprehensive feature vector according to the time window into different time dimension data; Divide the comprehensive feature vector according to the geographical location and longitude and latitude into different geographical space dimension data; Use the spatial convolutional layer of ST-CNN to slide the geographical space dimension data on the spatial dimension, learn the correlation between geographical grids, and extract the spatial distribution characteristics of evaluation hotspots in geographical regions; Use the temporal convolutional layer of ST-CNN to slide the time dimension data along the time dimension, capture the short-term temporal dependence relationship of evaluation hotspots, and extract the temporal distribution characteristics of evaluation hotspots in the time series; By alternately stacking spatial convolutional layers and temporal convolutional layers, the spatial distribution characteristics and temporal distribution characteristics of hotspots are jointly evaluated. Through the fully connected layer of ST-CNN, the spatio-temporal distribution of evaluation hotspots is identified.

[0010] As a preferred solution of the commodity evaluation monitoring method based on multi-dimensional data according to the present invention, wherein: based on the spatio-temporal distribution of evaluation hotspots, the evaluation hotspot intensity in different regions and times is identified, and the trend of evaluation hotspots is analyzed according to the identification result. The specific steps are as follows. Based on the spatio-temporal distribution of evaluation hotspots, aggregation in the time dimension and space dimension is performed, the hotspot intensity in the time dimension and the hotspot intensity in the space dimension are respectively calculated, and the hotspot intensities in the time dimension and the space dimension are combined to predict the evaluation hotspot intensity in different regions and times. ; At the current time point for the current region of the evaluation hotspot intensity time series analysis is performed to monitor the fluctuation of the hotspot intensity and identify the change rate of the hotspot intensity at consecutive time points. ; When the value is positive, it is considered that the evaluation hotspot is in a diffusion trend; When the value is negative, it is considered that the evaluation hotspot is in a decay trend.

[0011] As a preferred solution of the commodity evaluation monitoring method based on multi-dimensional data according to the present invention, wherein: the dynamic early warning is carried out according to the predicted trend of evaluation hotspots, and an evaluation report is generated. The specific steps are as follows. Based on historical evaluation hotspot data, an evaluation hotspot intensity threshold H1 and a diffusion rate threshold H2 are defined; According to the statistical analysis of the influence degree on historical public opinion hotspots, the early warning intensity is divided into level 1 early warning, level 2 early warning and level 3 early warning in ascending order; When > H1, a level 1 early warning is triggered; When > H2, a level 2 early warning is triggered; When > H1 and > H2, a level 3 early warning is triggered; The evaluation report includes the analysis of the current situation of evaluation hotspots, trend prediction, dissemination scope and countermeasure suggestions.

[0012] In a second aspect, the present invention provides a product evaluation monitoring system based on multi-dimensional data, including a data collection module, a feature fusion module, a spatio-temporal distribution recognition module, a trend prediction module, and a report generation module; the data collection module is used to collect multi-dimensional data and preprocess the multi-dimensional data; the feature fusion module is used to extract features from the preprocessed multi-dimensional data and fuse the extracted features into a comprehensive feature vector; the spatio-temporal distribution recognition module is used to identify the spatio-temporal distribution of evaluation hotspots based on the comprehensive feature vector through a spatio-temporal convolutional neural network; the trend prediction module is used to identify the regions and time periods of evaluation hotspots based on the spatio-temporal distribution of evaluation hotspots, and evaluate the trend of evaluation hotspots according to the recognition results; the report generation module is used to perform dynamic early warning based on the predicted trend of evaluation hotspots and generate an evaluation report.

[0013] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the product evaluation monitoring method based on multi-dimensional data as described in the first aspect of the present invention is implemented.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the product evaluation monitoring method based on multi-dimensional data as described in the first aspect of the present invention is implemented.

[0015] The beneficial effects of the present invention are as follows: By collecting multi-dimensional data, preprocessing these data, extracting their key features, and then fusing them into a comprehensive feature vector, and then using STCNN to perform spatio-temporal feature modeling on the data to identify the distribution of evaluation hotspots in different times and spaces. Based on these spatio-temporal features, the evaluation hotspot regions and time periods can be accurately located, and the diffusion trend of evaluation hotspots can be predicted. Finally, dynamic early warning is generated according to the prediction results, and a detailed evaluation analysis report is output to help the brand party timely respond to and resolve potential evaluation crises. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0017] Figure 1 It is a flowchart of the product evaluation monitoring method based on multi-dimensional data in Embodiment 1.

[0018] Figure 2Schematic diagram of the product evaluation monitoring system based on multi-dimensional data in Embodiment 1. Detailed implementation manners

[0019] To make the above objects, features and advantages of the present invention more obvious and understandable, the following detailed description of the specific implementation manners of the present invention will be given with reference to the accompanying drawings of the specification.

[0020] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0021] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" appearing in different places in this specification does not all refer to the same embodiment, nor is it a separate or alternative embodiment that excludes other embodiments.

[0022] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a product evaluation monitoring method based on multi-dimensional data, including the following steps: S1: Collect multi-dimensional data and preprocess the multi-dimensional data.

[0023] The multi-dimensional data includes text data, image data, video data, audio data, and time and geographical location information.

[0024] Furthermore, the multi-dimensional data is collected by integrating online resources such as social media platforms, e-commerce platforms, and content sharing websites, and using API interfaces, web crawler technologies, and mobile application data collection functions.

[0025] Denoise the text data by removing special symbols and stop words; Specifically, for the denoising process of text data, the process of removing special symbols and stop words means identifying and deleting all non-alphanumeric characters and those common words that contribute little or no contribution to semantic understanding, such as "of", "is", "in", etc. in the original text, so as to streamline the text content.

[0026] Adjust the scale of the image data by interpolation method to unify the size; Specifically, for the preprocessing of image data, the process of adjusting the scale through interpolation to unify the size means that when the image sizes are different, a mathematical interpolation algorithm is used to calculate the values of new pixel points, so that all images can be scaled to a predetermined standard size, thus ensuring the consistency of image input.

[0027] Reduce video data redundancy through frame sampling; Specifically, for the preprocessing of video data, the process of reducing redundancy through frame sampling means that representative frames are selected from a continuous sequence of video frames according to certain rules, and visually repetitive or minimally changing frames are removed to reduce the data volume while retaining the core information of the video content and improving the processing efficiency.

[0028] Remove background noise from audio data through a filter; Specifically, for the preprocessing of audio data, the process of removing background noise through a filter means that filtering algorithms in digital signal processing technology are applied to selectively weaken or eliminate unwanted sound components, such as environmental noise, wind noise, etc., and retain clear human voices or other target sounds to improve the audio quality.

[0029] Format time and geographical location information through standardization rules.

[0030] Specifically, for the preprocessing of time and geographical location information, the process of formatting through standardization rules means that according to established data standards, timestamps and geographical coordinates from different sources are converted into a unified expression form, such as using the international standard time format (ISO 8601) and the latitude and longitude representation method, to ensure that this information can be accurately compared and analyzed within the same framework.

[0031] S2: Extract features from the preprocessed multi-dimensional data and fuse the extracted features into a comprehensive feature vector.

[0032] Use BERT to extract deep semantic feature vectors of text data; Specifically, the input text data is tokenized and then fed into a pre-trained BERT model. The multi-layer bidirectional Transformer encoder captures context information, thereby generating vector representations containing rich semantic information for each vocabulary. These vectors can reflect the subtle differences and complex relationships of vocabulary in different contexts.

[0033] Use ResNet-50 to extract spatial feature vectors of image data; Specifically: The original image is first adjusted to a fixed size, and then low-level to high-level spatial features in the image are gradually extracted through a series of convolutional layers, residual blocks, and pooling layers. ResNet-50 effectively solves the problem of gradient disappearance in deep networks using the skip connection mechanism, enabling ResNet-50 to learn more abstract and complex visual patterns.

[0034] C3D is used to extract the temporal and spatial feature vectors of video data; Specifically: The video sequence is input into C3D as a three-dimensional tensor, where each frame is regarded as a two-dimensional plane, and the time dimension constitutes the third axis. By applying 3D convolutional kernels, C3D can capture both the temporal dynamics between frames and the spatial structure within frames in one operation, thus comprehensively representing the spatio-temporal characteristics of video content.

[0035] The frequency feature vectors of audio data are extracted through Mel spectrogram conversion and combined with CNN; Specifically: First, the audio signal is converted into a Mel spectrogram, which is a spectral representation method that maps the audio signal to a spectrum closer to human auditory perception. Then, this spectrogram is input into the CNN, and the convolutional layer automatically learns the local patterns in the spectrogram, such as the features of phonemes or speech segments, to obtain the frequency features of the audio.

[0036] The periodic feature vectors of time and geographical location information are extracted through sine time encoding combined with regional embedding encoding; Specifically: The timestamp is converted into sine function values to reflect the periodic changes within a day or a year; at the same time, the geographical location coordinates are transformed into high-dimensional vectors through embedding encoding. These vectors not only retain the relative distance information between positions but may also contain region-specific attributes. The combination of the two encoding methods can effectively capture the periodic and regional characteristics in time and geographical location information.

[0037] The attention mechanism is introduced to dynamically allocate attention weights to different feature vectors, and the expression is: ; Among them, is the attention weight of the th feature vector, is the transpose of the key vector of the th feature vector, is the query vector of the th feature vector, is the dimension of the th feature vector, is the index variable of the feature vector; It should be noted that the feature vectors include the deep semantic feature vectors of text data, the spatial feature vectors of image data, the temporal and spatial feature vectors of video data, the time-frequency feature vectors of audio data, and the periodic feature vectors of time and geographical location information.

[0038] Furthermore, by calculating the similarity between the query vector and the key vectors of each feature vector, the importance of each feature vector is determined. Specifically, the dot product operation is performed between the query vector and the key vector of each feature vector, and the result of the dot product is scaled by dividing it by the square root of the feature vector dimension. Subsequently, the softmax function is applied to convert these scaled dot product values into a probability distribution form, namely the attention weights. In this way, each feature vector obtains a weight value reflecting its relative importance, enabling more attention to be paid to the information considered more important in subsequent processing.

[0039] Based on the attention weights of different feature vectors, a multimodal fusion algorithm is adopted to perform feature interaction calculation, capture the non-linear relationships between different features, and generate a comprehensive feature vector. The expression is: ; where, is the comprehensive feature vector, is the total amount of all features, is the th feature vector, is the th feature vector, is the attention weight of the th feature vector, is different from is the index variable of another feature vector.

[0040] Furthermore, first, each feature vector is weighted and summed according to its attention weight. At the same time, considering the differences between all other feature vectors and the current feature vector, a normalization process is carried out through a denominator term to ensure reasonable combination even if there are large differences between features. Second, further considering the pairwise interaction between different feature vectors, they are multiplied after multiplying their respective attention weights, and then the sum of all possible interaction combinations is calculated and also normalized. Finally, the results obtained by these two methods are added together to generate a comprehensive feature vector, which not only retains the key information of the original multi-modal features but also incorporates the interactions between cross-modalities, providing a richer and more comprehensive representation.

[0041] For example, in the context of product review monitoring, suppose there is a discussion thread about a certain mobile phone product, which contains text comments (text data) posted by users, uploaded product usage photos (image data), an unboxing video (video data), and voice evaluation recordings of users (audio data). The attention mechanism will automatically evaluate the importance of these different sources of information according to the context. For example, text comments may emphasize more on product performance issues, while pictures or videos can more intuitively show the appearance design or usage experience. The multi-modal fusion algorithm will then combine this information, together with background information such as time and location, to generate a comprehensive feature vector that can comprehensively reflect the overall attitude of users towards the product.

[0042] S3: Based on the comprehensive feature vector, through a spatio-temporal convolutional neural network, identify the spatio-temporal distribution of evaluation hotspots.

[0043] Divide the comprehensive feature vector into different time dimension data according to time windows; Specifically: Divide the comprehensive feature vector into a series of time segments according to the time order. Each time segment represents a specific time period, such as the evaluation information within one day, one week, or one month. This process ensures that the evaluation changes in different time periods can be analyzed separately, so as to more accurately capture the trend of evaluation evolution over time.

[0044] Divide the comprehensive feature vector into different geographical space dimension data according to geographical location and longitude and latitude; Specifically: Allocate the comprehensive feature vector to different geographical grids according to geographical coordinates. Each geographical grid corresponds to a specific geographical location, and all evaluation hotspots related to this location will be classified into the corresponding grid. In this way, the evaluation status of each region can be accurately analyzed, and the evaluation differences between different regions can be identified.

[0045] Use the spatial convolutional layer of the ST-CNN to slide the geographical space dimension data on the spatial dimension, learn the correlation between geographical grids, and extract the spatial distribution characteristics of evaluation hotspots in geographical regions; Specifically: Input the data in the geographical grid into the spatial convolutional layer. The convolutional operation will slide on the geographical grid to learn the correlation patterns between adjacent grids. Through this sliding operation, it can be identified how evaluation hotspots spread between different geographical locations, and then extract the spatial distribution characteristics of evaluation hotspots in geographical regions.

[0046] Use the temporal convolutional layer of the ST-CNN to slide the time dimension data along the time dimension, capture the short-term time dependence relationship of evaluation hotspots, and extract the time distribution characteristics of evaluation hotspots in the time series; Specifically: The evaluation hot spot data divided by time window is input into the temporal convolutional layer along the time axis. The temporal convolutional layer moves along the time series in the way of sliding window, and extracts local features for each time point and its adjacent time points. In this process, the convolutional kernel captures the dependencies between adjacent time points, and identifies the short-term patterns and trends of the evolution of evaluation hot spots over time. In this way, the temporal convolutional layer can effectively extract the variation rules of evaluation hot spots from the time series data, including characteristics such as the occurrence frequency, duration, and intensity of events, so as to reveal the distribution characteristics of evaluation hot spots in time.

[0047] By alternately stacking the spatial convolutional layer and the temporal convolutional layer, jointly evaluating the spatial distribution characteristics and temporal distribution characteristics of evaluation hot spots, and through the fully connected layer of ST-CNN, the spatio-temporal distribution of evaluation hot spots is identified.

[0048] Specifically: In the fully connected layer of ST-CNN, all information from the spatial and temporal dimensions is integrated, and each neuron comprehensively considers all input information of the previous layer, so as to identify the distribution of evaluation hot spots in the entire spatio-temporal range. In this way, the fully connected layer can generate an output result that comprehensively reflects the spatio-temporal characteristics of evaluation hot spots.

[0049] It should be noted that evaluation hot spots refer to public topics that arouse extensive attention or discussion in time and space, and these evaluation hot spots are usually collected through user interactions on social media, news media, forums, and O2O platforms.

[0050] S4: Based on the spatio-temporal distribution of evaluation hot spots, identify the regions and time periods of evaluation hot spots, and evaluate the trends of evaluation hot spots according to the identification results.

[0051] Based on the spatio-temporal distribution of evaluation hot spots, perform aggregation in the time dimension and the space dimension, calculate the hot spot intensity in the time dimension and the hot spot intensity in the space dimension respectively, and combine the hot spot intensities in the time dimension and the space dimension to predict the evaluation hot spot intensities in different regions and times , the expression is: ; Where is the weight coefficient in the time dimension, is the weight coefficient in the space dimension, is the set of historical time points within the time window, represents the current time point, represents the historical time point, represents the historical time point when in the current region the discussion volume of the evaluation hot spot, is the time decay coefficient, represents the current region The set of domain regions, Express the current region And the domain region The distance between Denote the spatial attenuation coefficient, Denote the domain region At the current time point The discussion volume of the tone hot spot Denote at the time point The current region at The comprehensive evaluation hot spot intensity; Based on the spatio-temporal distribution of the evaluation hot spots, perform aggregation in the time dimension and the space dimension. The specific process is as follows: Based on the spatio-temporal distribution of the evaluation hot spots, first, in the time dimension, segment and aggregate the evaluation hot spots according to time windows (such as hours, days, weeks), calculate the hot spot intensity within each time period, and capture the changing trend of the evaluation over time; second, in the space dimension, take geographical regions as units, aggregate the evaluation hot spots in different regions, and combine the proximity relationship and the attenuation coefficient between regions , calculate the hot spot intensity and its diffusion in each region; finally, combine the aggregation results in the time dimension and the space dimension to generate a spatio-temporal distribution map of the evaluation hot spots, comprehensively showing the evolution of the hot spots over time and the propagation law in space.

[0052] Furthermore, first calculate the hot spot intensity for the time dimension. This process involves collecting the set of historical time points Within the time window before the current time point All the data. For each historical time point , according to the gap between this time point and the current time point t, apply a decay function to measure the influence of the historical discussion volume , and thus calculate a weighted average as the hot spot intensity in the time dimension.

[0053] Next, when calculating the hot spot intensity in the space dimension, consider the current region And its set of neighborhood regions, All the relevant data in. For each domain region , according to the distance between it and the current region , also use a decay function to evaluate the influence of the discussion volume of adjacent regions at the current time point t On the current region, and then obtain the hot spot intensity in the space dimension.

[0054] Finally, in order to predict the comprehensive evaluation hot spot intensity in different regions and times , combine the hotspot intensities in the time dimension and the space dimension calculated above. The weight coefficient in the time dimension is used to adjust the importance of the time factor, while the weight coefficient in the space dimension is used to balance the role of the space factor. In this way, the influences of both the time and space aspects are comprehensively considered, and a comprehensive evaluation hotspot intensity that can reflect the current time point t and the current region is generated of the current region . It not only reflects the discussion heat of the current region changing over time but also considers the influence of the surrounding regions on the evaluation of the current region, providing a more comprehensive understanding of the evaluation dynamics.

[0055] According to the current time point for the current region of the evaluation hotspot intensity perform time series analysis to monitor the fluctuations of the hotspot intensity and identify the change rate of the hotspot intensity at consecutive time points , and the expression is: ; where represents the comprehensive evaluation hotspot intensity of the current region at the time point ; When the value of is positive, it is considered that the evaluation hotspot is in a spreading trend; When the value of is negative, it is considered that the evaluation hotspot is in a decaying trend.

[0056] S5: Conduct dynamic early warning based on the predicted trend of the evaluation hotspot and generate an evaluation report.

[0057] Based on historical evaluation hotspot data, define the evaluation hotspot intensity threshold H1 and the diffusion rate threshold H2; According to the statistical analysis of the influence degree on historical public opinion hotspots, divide the early warning intensity into level 1 early warning, level 2 early warning, and level 3 early warning in ascending order; When > H1, trigger a level 1 early warning; It should be noted that when triggering a level 1 early warning, it is recommended to initiate preliminary data verification and regional investigation to promptly understand the specific reasons for abnormal evaluations and prevent the further expansion of abnormal evaluations.

[0058] When > H2, trigger a level 2 early warning; It should be noted that when triggering a level 2 early warning, it is recommended to immediately strengthen the monitoring intensity and evaluate potential impacts.

[0059] When > H1 and ​When it is greater than H2, a third-level warning is triggered; It should be noted that when a third-level warning is triggered, emergency actions need to be taken to control the spread of the evaluation.

[0060] The evaluation report includes the analysis of the current situation of evaluation hotspots, trend prediction, dissemination scope, and countermeasure suggestions.

[0061] This embodiment also provides a commodity evaluation monitoring system based on multi-dimensional data, including: a data acquisition module, a feature fusion module, a spatio-temporal distribution recognition module, a trend prediction module, and a report generation module; The data acquisition module is used to collect multi-dimensional data and preprocess the multi-dimensional data; The feature fusion module is used to extract features from the preprocessed multi-dimensional data and fuse the extracted features into a comprehensive feature vector; The spatio-temporal distribution recognition module is used to identify the spatio-temporal distribution of evaluation hotspots based on the comprehensive feature vector through a spatio-temporal convolutional neural network; The trend prediction module is used to identify the regions and time periods of evaluation hotspots based on the spatio-temporal distribution of evaluation hotspots and evaluate the trends of evaluation hotspots according to the recognition results; The report generation module is used to perform dynamic warning according to the predicted trends of evaluation hotspots and generate an evaluation report.

[0062] This embodiment also provides a computer device applicable to the situation of the commodity evaluation monitoring method based on multi-dimensional data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the commodity evaluation monitoring method based on multi-dimensional data as proposed in the above embodiment.

[0063] This computer device can be a terminal. This computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of this computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of this computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0064] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for monitoring product evaluations based on multi-dimensional data proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, a magnetic disk or an optical disc.

[0065] In summary, the present invention: collects multi-dimensional data, preprocesses these data, extracts their key features, and then fuses them into a comprehensive feature vector. Next, STCNN is used to model the spatio-temporal features of the data to identify the distribution of evaluation hotspots in different times and spaces. Based on these spatio-temporal features, the evaluation hotspot areas and time periods can be accurately located, and the diffusion trend of evaluation hotspots can be predicted. Finally, a dynamic warning is generated according to the prediction results, and a detailed evaluation analysis report is output to help the brand party timely respond to and resolve potential evaluation crises.

[0066] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for monitoring product evaluations based on multi-dimensional data, characterized in that: Including, Collecting multi-dimensional data and preprocessing the multi-dimensional data; Extracting features from the preprocessed multi-dimensional data and fusing the extracted features into a comprehensive feature vector; Based on the comprehensive feature vector, through a spatio-temporal convolutional neural network, identifying the spatio-temporal distribution of evaluation hotspots; Based on the spatio-temporal distribution of evaluation hotspots, identifying the regions and time periods of evaluation hotspots, and predicting the trend of evaluation hotspots according to the identification results; Conducting dynamic early warning according to the predicted trend of evaluation hotspots and generating an evaluation report.

2. The method for monitoring product evaluations based on multi-dimensional data according to claim 1, wherein: The multi-dimensional data includes text data, image data, video data, audio data, and time and geographical location information.

3. The method for monitoring product evaluations based on multi-dimensional data according to claim 2, wherein: The preprocessing of the multi-dimensional data is as follows specifically, Denosing the text data by removing special symbols and stop words; Adjusting the scale of the image data through an interpolation method to unify the size; Reducing video data redundancy through frame sampling; Removing background noise from the audio data through a filter; Formatting the time and geographical location information through a standardization rule.

4. The method for monitoring product evaluations based on multi-dimensional data according to claim 3, wherein: The feature extraction of the preprocessed multi-dimensional data and the fusion of the extracted features into a comprehensive feature vector are as follows specifically, Using BERT to extract the deep semantic feature vector of the text data; Using ResNet-50 to extract the spatial feature vector of the image data; Using C3D to extract the temporal and spatial feature vectors of the video data; Extracting the frequency feature vector of the audio data through Mel spectrogram conversion and combining with CNN; Extracting the periodic feature vector of the time and geographical location information through sinusoidal time encoding combined with regional embedding encoding; Introducing an attention mechanism to dynamically allocate attention weights for different feature vectors; Based on the attention weights of different feature vectors, using a multi-modal fusion algorithm for feature interaction calculation, capturing the non-linear relationship between different features, and generating a comprehensive feature vector.

5. The method for monitoring product evaluations based on multi-dimensional data according to claim 4, wherein: The identification of the spatio-temporal distribution of evaluation hotspots based on the comprehensive feature vector through a spatio-temporal convolutional neural network is as follows specifically, Divide the comprehensive feature vector according to the time window into different time dimension data; Divide the comprehensive feature vector according to geographical location and longitude and latitude into different geographical space dimension data; Using the spatial convolutional layer of ST-CNN, sliding the geographical space dimension data on the spatial dimension to extract the spatial distribution features of evaluation hotspots in geographical regions; Using the temporal convolutional layer of ST-CNN, sliding the time dimension data along the time dimension to extract the temporal distribution features of evaluation hotspots in the time series; By alternately stacking the spatial convolutional layer and the temporal convolutional layer, combining the spatial distribution features and temporal distribution features of evaluation hotspots, and through the fully connected layer of ST-CNN, identifying the spatio-temporal distribution of evaluation hotspots.

6. The method for monitoring product evaluations based on multi-dimensional data according to claim 5, wherein: Based on the spatio-temporal distribution of evaluation hotspots, identifying the intensity of evaluation hotspots in different regions and times, and analyzing the trend of evaluation hotspots according to the identification results, as follows specifically, Based on the spatio-temporal distribution of evaluation hotspots, perform aggregation in the time dimension and the space dimension, calculate the hotspot intensity in the time dimension and the hotspot intensity in the space dimension respectively, and combine the hotspot intensities in the time dimension and the space dimension to predict the evaluation hotspot intensity in different regions and at different times ; According to the current time point For the current area The evaluation hot spot intensity Perform time series analysis to monitor the fluctuations of the hot spot intensity and identify the change rate of the hot spot intensity at consecutive time points ; When is positive, it is considered that the evaluation hotspot is in a spreading trend; When has a negative value, it is considered that the evaluation hot spot is in a decaying trend.

7. The method for monitoring product evaluations based on multi-dimensional data according to claim 6, wherein: The dynamic early warning according to the predicted trend of evaluation hotspots and the generation of an evaluation report are as follows specifically, Based on historical evaluation hotspot data, defining an evaluation hotspot intensity threshold H1 and a diffusion rate threshold H2; According to the statistical analysis of the influence degree on historical public opinion hotspots, dividing the early warning intensity into level 1 early warning, level 2 early warning, and level 3 early warning in ascending order; When > H1, a first-level warning is triggered; When > H2, a secondary warning is triggered; When > H1 and > H2, a third-level warning is triggered; The evaluation report includes the analysis of the current situation of evaluation hotspots, trend prediction, dissemination scope, and coping suggestions.

8. A commodity evaluation monitoring system based on multi-dimensional data, based on the commodity evaluation monitoring method based on multi-dimensional data according to any one of claims 1 to 7, characterized in that: It includes a data collection module, a feature fusion module, a spatio-temporal distribution recognition module, a trend prediction module, and a report generation module; The data collection module is used to collect multi-dimensional data and preprocess the multi-dimensional data; The feature fusion module is used to extract features from the preprocessed multi-dimensional data and fuse the extracted features into a comprehensive feature vector; The spatio-temporal distribution recognition module is used to identify the spatio-temporal distribution of evaluation hotspots based on the comprehensive feature vector through a spatio-temporal convolutional neural network; The trend prediction module is used to identify the regions and time periods of evaluation hotspots based on the spatio-temporal distribution of evaluation hotspots and evaluate the trends of evaluation hotspots according to the recognition results; The report generation module is used to perform dynamic early warning based on the predicted trends of evaluation hotspots and generate an evaluation report.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for monitoring commodity evaluations based on multi-dimensional data according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for monitoring commodity evaluations based on multi-dimensional data according to any one of claims 1 to 7.

Images